\
"
The
information
in
this
document
has
been
funded
wholly
or
in
part
by
the
United
States
Environmental
Protection
Agency
under
Contract
No.
68­
01­
5838.
It
has
been
subject
to
the
Agency's
peer
and
administrative
review,
and
it
has
been
approved
for
publication
as
an
EPA
document.
Mention
of
trade
names
or
commercial
products
does
not
constitute
endorsement
or
recommendation
for
use."

i
March
1983
A
Comparison
of
Alternative
Approaches
for
Estimating
Recreation
and
Related
Benefits
of
Water
Quality
Improvements
Prepared
for
U.
S.
Environmental
Protection
Agency
Economic
Analysis
Division
Washington,
DC
20460
Dr.
Ann
Fishe~
Project
Officer
Prepared
by
Dr.
William
H.
Desvousges
Research
Triangle
Institute
Research
Triangle
Park,
NC
27709
Dr.
V.
Kerry
Smith
University
of
North
Carolina
Chapel
Hill,
NC
27514
and
Matthew
P.
McGivney
Research
Triangle
Institute
Research
Triangle
Park,
NC
27709
EPA
Contract
No.
68­
01­
5838
This
research
project
was
initiated
and
supported
under
work
agkeement
68­
01­
5838
by
the
Benefits
Staff
in
the
Office
of
Policy
Analysis
at
the
U.
S.
Environmental
Protection
Agency
(
EPA).

Throughout
this
research
effort,
the
authors
of
this
report
were
fortunate
enough
to
take
advantage
of
research
activities
already
in
progress.
One
author
had
partially
completed
an
analysis
of
the
problems
of
defining
and
measuring
option
value,
for
example,
and
another
had
partially
completed
research
to
design
a
generalized
travel
cost
site
demand
model.
In
addition
the
authors
also
benefited
from
free
access
to
any
array
of
related
working
papers
 
many
of
which
have
subsequently
been
published
 
that
improved
the
final
research
design
beyond
that
possible
otherwise.
Finally,
access
to
an
independently
developed
estimator
for
ranked
data
improved
the
authors'
ability
to
make
certain
types
of
comparisons
for
the
contingent
ranking
component
of
the
survey.
Although
none
of
these
complementary
activities
was
contemplated
when
the
project
was
initially
proposed,
each
has
played
a
substantial
role
in
the
final
results,
We
would
not
expect
these
same
circumstances
to
be
easily
replicated
in
future
projects
of
comparable
scale
and
duration.

This
final
report
has
been
substantially
improved
through
the
constructive
comments
of
many
reviewers.
In
particular
we
would
like
to
thank
Ann
Fisher,
the
EPA
project
officer,
for
her
careful
commentary
and
continuous
support.
In
addition,
as
part
of
the
EPA's
review,
six
other
individuals
furnished
detailed
comments:

Richard
Bishop,
University
of
Wisconsin
Rick
Freeman,
Bowdoin
College
Bill
Lott,
University
of
Connecticut
Robert
Mitchell
and
Richard
Carson,
Resources
for
the
Future
(
RFF)
Bill
Schuize,
University
of
Wyoming.

In
addition,
useful
comments
were
also
received
from
the
following
individuals:

Tayler
Bingham,
Research
Triangle
Institute
(
RTI)
Peter
Caulkins,
U.
S.
EPA
Warren
Fisher,
U.
S.
Fish
and
Wildlife
Service
David
Gallagher,
University
of
New
South
Wales
Debbie
Gibbs,
Bureau
of
Reclamation
Jerry
Hausman,
Massachusetts
Institute
of
Technology
Reed
Johnson,
U.
S.
Naval
Academy
and
U.
S.
EPA
John
Loomis,
U.
S.
Forest
Service
Glenn
Morris,
RTI
Doug
Rae,
Charles
River
Associates
Liz
Wilman,
RFF.

The
authors
also
have
benefited
from
the
comments
of
participants
at
presentations
given
at
RFF;
Vanderbilt
University;
the
University
of
Missouri­
Rolla;
Dillon,
Colorado
(
Visual
Values
Workshop);
Research
Triangle
Park
(
Triangle
Econometrics
Seminar);
and
Washington,
DC.
(
EPA).
We
pre
most
grateful
for
all
these
efforts.
Finally,
we
are
most
appreciative
of
the
efforts
of
our
editor,
Hall
Ashmore,
and
of
Jan
Shirley,
Supervisor
of
RTI's
Word
Processing
Center.

111
.
,
_
.
.
..­.
.­­_
 
 
.
...­
C
O
N
T
E
N
T
S
Page
Figures
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Tables.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

1
Introduction,
Objectives,
and
Summary
.
.
.
.
.
.
.
.
.
.
.
.
1.1
Introduction
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
1.2
Objectives
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
1.3
SummaryofResults.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
1.3.1
Overview.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
1
.3.2
Contingent
Valuation
Approach
.
.
.
.
.
.
.
.
.
.
1
.3.3
Travel
Cost
Approach.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
1
.3.4
Approach
Comparison.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
1
.3.5
Considerations
for
Future
Research
.
.
.
.
.
.
.
.
l.
4Guide
to
the
Report...
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

2
A
Brief
Review
of
the
Conceptual
Basis
for
the
Benefit
Estimation
Approaches
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
2.11ntroduction
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
2.2
A
Brief
Review
of
the
Conventional
Theory
of
Benefits
Measurement
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
2.3
A
Framework
for
Comparing
Alternative
Benefit
Measurement
Approaches.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
2.4
The
Nature
of
the
Benefits
Measured
in
the
Alternative
Approaches
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
2.4.1
Travel
Cost
Approach.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
2.4.2
Contingent
Valuation
Approach
.
.
.
.
.
.
.
.
.
.
2.4.3
Contingent
Ranking
Approach.
.
.
.
.
.
.
.
.
.
.
2.5
Summary
.
.
.
.
.

3
Survey
Design
.
.
.
.
.
3.1
Introduction
.
.
.
3.2
General
Description
3.2.1
Geography
.
3.2.2
Uses
.
.
.
.
3.2.3
Recreation
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
of
the
Monongahela
River
Basin
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
3.2.4
Socioeconomic
Profile
3.3
Sampling
Plan
.
.
.
.
.
.
3.3.1
Target
Population
.
3.3.2
Sample
Selection
and
3.3.3
Sampling
Weights
.
.
3.4
Survey
Plan
.
.
.
.
.
.
.
3.4.1
Questionnaire
Desicjn
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Survey
Design.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
and
Limited
Local
Pretest
.
.
3.4.2
Retaining
Field
Su~
ervisors
and
Hiring
Interviewers
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

v
ix
xi
1­
1
1­
1
1
­
4
1­
5
1­
5
1­
5
1­
7
1­
8
1­
11
1­
15
2­
1
2­
1
2
­
2
2
­
9
2­
12
2­
12
2­
13
2­
14
2­
14
3­
1
3­
1
3­
1
3­
1
3­
3
3
­
3
3
­
4
3
­
5
3
­
5
3
­
6
3
­
6
3­
6
3­
7
3
­
9
F.

CONTENTS
(
continued)

3.4.3
Counting
and
Listing
of
Sample
Segments
.
3.4.4
Developing
Field
Manuals
and
Conducting
Interviewer
Training
.
.
.
.
.
.
.
.
.
.
.
3.4.5
Training
Session
.
.
.
.
.
.
.
.
.
.
.
.
.
3.4.6
Conducting
Household
Interviews
.
.
.
.
.
3.4.7
Initial
Contacts
and
Obtaining
Cooperation
3.4.8
Household
Enumeration
.
.
.
.
.
.
.
.
.
.
3.4.9
Interviewing
Procedures
.
.
.
.
.
.
.
.
.
3.4.10
Interviewer
Debriefing
.
.
.
.
.
.
.
.
.
.
3.4.11
Data
Receipt,
Editing,
and
Keypunching
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

4
Contingent
Valuation
Design
and
Results:
Option
Price
and
User
Values
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
4.11ntroduction
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
4.2
A
Review
of
Design
Issues
in
Contingent
Valuation
Survey
s
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
4.2.1
Hypothetical
Bias
.
.
.
.
.
.
.
.
.
.
.
.
.
4.2.2
Strategic
Bias....
.
.
.
.
.
.
.
.
.
.
.
4.2.3
Payment
Vehicle
Bias.
.
.
.
.
.
.
.
.
.
.
.
4.2.4
Starting
Point
Bias
.
.
.
.
.
.
.
.
.
.
.
.
4.2.5
Information
Bias
.
.
.
.
.
.
.
.
.
.
.
.
.
.
4.2.6
Interviewer
Bias
.
.
.
.
.
.
.
.
.
.
.
.
.
.
4.2.7
Summary
and
Implications
for
Contingent
Valuation
Research
Design
.
.
.
.
.
.
.
.
.
4.3
Questionnaire
Design
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
4.3.1
Questionnaire
Design:
Part
A
.
.
.
.
.
.
.
4.3.2
Benefits
Measures:
Part
B.
.
.
.
.
.
.
.
.
4.4
Profiles
of
Survey
Respondents
.
.
.
.
.
.
.
.
.
.
4.50ption
Price
Results
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
4.6
User
Value
Results
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
4.7
Summary
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
,..
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

5
Contingent
Valuation
Design
and
Results:
Option
and
Existence
Values
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
5.1
Introduction
5
.
2
C
o
n
t
i
n
g
e
n
t
Cla;
m&
"
Marke~
s
and.
the.
M.
odel.
ing
of.
.
"
.
.
.

Uncertainty
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
5.3
Option
Value:
The
"
Timeless"
Analyses
.
.
.
.
.
.
.
.
.
5.4
The
Time­
Sequenced
Analyses
.
.
.
.
.
.
.
.
.
.
.
.
.
.
5.5
Recent
Estimates
of
Nonuser
Values
.
.
.
.
.
.
.
.
.
.
.
5.6
Measuring
Option
Value:
Survey
Design
.
.
.
.
.
.
.
.
5.7
Survey
Results­­
Option
Value
.
.
.
.
.
.
.
.
.
.
.
.
.
.
5.7.1
Option
Value­­
Demand
Uncertainty
.
.
.
.
.
.
.
.
5.7.2
Option
Value­­
Supply
Uncertainty
.
.
.
.
.
.
.
.
5.8
Existence
Value
Estimates
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
5.9
Summary
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

vi
Page
3­
9
3­
1o
3­
11
3­
11
3­
13
3­
14
3­
14
3­
16
3­
18
4­
1
4­
1
4­
2
4
­
2
4
­
4
4
­
6
4­
6
4­
7
4­
7
4
­
9
4
­
9
4
­
9
4­
11
4­
20
4­
27
4­
36
4­
38
5­
1
5­
1
5
­
3
5­
7
5­
14
5­
16
5­
21
5­
25
5­
25
5­
29
5­
31
5­
33
.
.
.
..­.
.
.
.
.
.
CONTENTS
(
continued)

Page
6
Contingent
Ranking
Design
and
Results:
Option
Prices.
.
.
.
6.11ntroduction
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
6.2
Consumer
Behavior
and
the
Contingent
Ranking
Framework
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
6.3
Estimation
of
Random
Utility
Models
with
Ordered
Alternatives
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
6.4
Past
Applications
of
Contingent
Ranking
.
.
.
.
.
.
.
.
.
6.5
Monongahela
Contingent
Ranking
Experiment:
Design
and
Estimates
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
6.6
Benefit
Estimates
with
Contingent
Ranking
Models
.
.
.
.
6.7
Implications
and
Further
Research
.
.
.
.
.
.
.
.
.
.
.
.

7
A
Generalized
Travel
Cost
Model
for
Measuring
the
Recreation
Benefits
of
Water
Quality
Improvements
.
.
.
.
.
.
7.1
Introduction
7.2
Travel
Cost
Model
JI
III
j::::::::::;:::
7.3
The
Travel
Cost
Model
for
Heterogeneous
Recreation
Sites
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
7.4
Sources
of
Data.....
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
7.5
Empirical
Results
for
Site­
Specific
Travel
Cost
Models
.
.
7.5.1
The
Treatment
of
Onsite
Time
.
.
.
.
.
.
.
.
.
.
7.5.2
The
Opportunity
Cost
of
Travel
Time
.
.
.
.
.
.
.
7.5.3
Results
for
the
Basic
Model
.
.
.
.
.
.
.
.
.
7.5.4
Results
for
the
Tailored
Modeis"
.
.
.
.
.
.
.
.
.
7.5.5
Evaluation
of
Measures
of
the
Opportunity
Cost
of
Travel
Time.....
.
.
.
.
.
.
.
.
.
.
.
.
.
7.6
Further
Evaluation
of
the
Travel
Cost
Models
.
.
.
.
.
.
7.7
Analyzing
the
Role
of
Water
Quality
for
Recreation
Demand
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
7.8
A
Measure
of
the
Benefits
of
a
Water
Quality
Change
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
7.9
Summary
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

8
A
Comparison
of
the
Alternative
Approaches
for
Estimating
Recreation
and
Related
Benefits
.
.
.
.
.
.
.
.
,
.
.
.
.
.
.
8.1
Introduction
8.2
The
Conceptual
;
ram.
ewo;
k.
for.
a.
Cornpar;
s&
.
o;
"
.
"
.
.

Recreation
Benefit
Estimation
Approaches
.
.
.
.
.
.
.
.
8.2.1
Background
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
8.2.2
Research
Design
and
Comparative
Analysis
.
.
.
.
8.2.3
Past
Comparisons
of
Benefit
Estimation
Methods
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
8.3
A
Comparative
Evaluation
of
the
Contingent
Valuation,
Travel
Cost,
and
Contingent
Ranking
Benefit
Estimation
Methods
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
8.41mplications
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
6­
1
6­
1
6­
2
6­
7
6­
9
6­
16
6­
25
6­
28
7­
1
7­
1
7
­
2
7­
1o
7­
22
7­
30
7­
31
7­
32
7­
32
7­
36
7­
38
7­
43
7­
51
7­
57
7­
64
8­
1
8­
1
8­
2
8­
2
8­
3
8­
9
8­
12
8­
20
vii
,
CONTENTS
(
continued)

9
References
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

Appendixes
A
Sample
Design
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
B
Survey
Forms
and
Procedures
.
.
.
.
.
.
.
.
.
.
.
.
Part
l­­
Household
Control
Form.
.
.
.
.
.
.
.
.
.
.
Part
2­­
Counting
and
Listing
Examples
.
.
.
.
.
.
.
Part
3­­
Debriefing
Agenda
.
.
.
.
.
.
.
.
.
.
.
.
.
Part
4­­
Quality
Control
Procedures
.
.
.
.
.
.
.
.
.
c
Survey
Analysis:
Supporting
Tables
.
.
.
.
.
.
.
.
.
D
Survey
Questionnaires
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Part
l­­
Survey
Questionnaire
.
.
.
.
.
.
.
.
.
.
.
Part
2­­
Suggestions
for
Improving
the
Questionnaire
Future
Use
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
E
Technical
Water
Quality
Measures:
An
Economist's
Perspective
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
F
Travel
Cost:
Supporting
Tables
.
.
.
.
.
.
.
.
.
.
.
G
Alternative
Regression
Models
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
for
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
Paqe
9­
1
A­
1
B­
1
B­
1
B
­
4
B­
7
B­
9
c
­
1
D­
1
D­
2
D­
28
E­
1
F­
1
G­
1
.
.
.
Vlll
FIGURES
Number
1­
1
1­
2
2­
1
2­
2
2­
3
2­
4
2­
5
3­
1
3­
2
3­
3
3
­
4
4­
1
4­
2
4­
3
4­
4
4­
5
4
­
6
4­
7
4
­
8
5­
1
5
­
2
5­
3
5
­
4
5­
5
6­
1
7­
1
Effects
and
responses
to
water
quality
regulatory
actions
.
.
A
spectrum
of
water
quality
benefits
.
.
.
.
.
.
.
.
.
.
.
.
.

The
demand
function
and
the
consumer
surplus
welfare
measure
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
A
comparison
of
alternative
welfare
measures
.
.
.
.
.
.
.
.
.
Surplus
measures
for
a
change
in
quantity
.
.
.
.
.
.
.
.
.
.
Smith
­
Krutilla
framework
for
classifying
the
measurement
bases
and
approaches
of
economic
benefits
resulting
from
improved
water
quality.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Travel
cost
demand
function
with
water
quality
improvement
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

Map
of
Monongahela
River
and
other
area
recreation
sites
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Geographic
location
of
survey
area
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Field
interviewer
training
session
agenda
.
.
.
.
.
.
.
.
.
.
Summary
of
completed
interviews
.
.
.
,
.
.
.
.
.
.
.
.
.
.
.

Activity
card
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Site
activity
matrix
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Map
of
Monongahela
River
and
other
area
recreation
sites.
.
.
Recreation
sites
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Water
quality
ladder
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Value
card
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Payment
card
Rankordercardjjj;
l::
j:::::::::::::::

Optimal
allocation
of
choice
with
contingent
claims
.
.
.
.
.
Optimal
allocation
of
choices
of
contingent
claims
without
uniqueness
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Option
value
in
Cicchetti­
Freeman
analysis
.
.
.
.
.
.
.
.
.
.
Option
value
in
Cicchetti­
Freeman
with
"
no
demand"
.
.
.
.
.
Option
value
with
contingent
claims
in
Graham's
analysis
.
.
.

Rank
order
card
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

Measurement
of
consumer
surplus
increment
due
to
water
quality
improvement
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.

ix
Page
1­
2
1­
3
2
­
3
2
­
5
2
­
6
2­
1o
2­
12
3
­
2
3
­
5
3­
12
3­
13
4­
1o
4­
1o
4­
12
4­
12
4­
13
4­
14
4­
17
4­
20
5
­
6
5
­
7
5­
12
5­
12
5­
13
6­
18
7­
59
­.­,­­
I
ABLtb
Number
1­
1
1­
2
2­
1
3­
1
3­
2
4­
1
4
­
2
4­
3
4­
4
4­
5
4­
6
4­
7
4
­
8
4
­
9
4­
1o
4­
11
4­
12
4­
13
4­
14
5­
1
5
­
2
5­
3
5
­
4
5­
5
A
Comparison
of
Mean
Benefit
Estimates
.
.
.
.
.
.
.
.
.
.
.
Regression
Comparisons
of
Contingent
Valuation
and
Travel
Cost
Benefit
Estimates
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

Alternative
Welfare
Measures
and
Types
of
Consumer
Surplus
Measures
for
Contingent
Valuation
Studies
.
.
.
.
.
.

Questionnaire
Development
Activity
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Final
Distribution
of
Sample
Housing
Units
.
.
.
.
.
.
.
.
.
.

Summary
of
Biases
in
Contingent
Valuation
Experiments
.
.
.
.
Summary
of
Option
Price
Question
Formats
by
Interview
Type
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Characteristics
of
Key
Respondent
Groups
.
.
.
.
.
.
.
.
.
.
Reasons
for
Zero
Bids
by
Elicitation
Method
.
.
.
.
.
.
.
.
.
Degree
of
Importance
of
Water
Quality
by
Key
Respondent
Groups
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Respondent
Attitudes
About
Self
by
Key
Respondent
Groups
.
Logit
Estimation
of
Zero
Bids
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Profile
of
Outliers
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Estimated
Option
Price
for
Changes
in
Water
Quality:
Effects
of
Instrument
and
Type
of
Respondent­­
Protest
Bids
and
Outliers
Excluded.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Student
t­
Test
Results
for
Option
Price­­
Protest
Bids
and
Outliers
Excluded..
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Regression
Results
for
Option
Price
Estimates­­
Protest
Bids
and
Outliers
Excluded
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Student
t­
Test
Results
for
Option
Price­­
Protest
Bids
and
Outliers
Excluded..
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Estimated
User
Values­­
Protest
Bids
and
Outliers
Excluded
.
.
Regression
Results
for
User
Value
Estimates
of
Water
Quality
Changes
­­
Protest
Bids
and
Outliers
Excluded
.
.
.
.
.

Summary
of
Mitchell­
Carson
Estimated
Mean
Annual
Willingness
to
Pay
by
Version
and
Water
Quality
.
.
.
.
.
.
.
Summary
of
Willingness­
to­
Pay
Questions
by
Type
oflnterview
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Summary
of
User,
Supply
Uncertainty,
and
Existence
Value
Questions
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Estimated
Option
Values
for
Water
Quality
Change:
Effects
of
Instrument
and
Type
of
Respondent­­
Protest
Bids
and
Outliers
Excluded..
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Student
t­
Test
Results
for
Question
Format.
.
.
.
.
­
.
.
.
.
Paqe
1­
9
1­
1o
2­
7
3
­
8
3­
15
4­
8
4­
18
4­
21
4­
23
4­
24
4­
25
4­
26
4­
30
4­
32
4­
33
4­
34
4­
36
4­
37
4­
38
5­
19
5­
24
5­
24
S­
26
5­
27
xi
TABLES
(
continued)

Number
Page
5­
6
5­
7
5­
8
5
­
9
6­
1
6
­
2
6
­
3
6
­
4
6­
5
7­
1
7
­
2
7
­
3
7
­
4
7
­
5
7
­
6
7­
7
7­
8
7
­
9
7­
1o
7­
11
7­
12
7­
13
7­
14
7­
15
7­
16
7­
17
7­
18
7­
19
Regression
Results
for
Option
Value
Estimates­­
Protest
Bids
and
Outliers
Excluded..
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Effects
of
Supply
Uncertainty
on
Option
Price
.
.
.
.
.
.
.
.
Student
t­
Tests
for
the
Effects
of
Supply
Uncertainty
for
Users...
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Estimated
Existence
Values
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

Summary
of
Rae/
CRA
Contingent
Ranking
Studies
.
.
.
.
.
,
Combinations
of
Water
Quality
and
Payments
for
Monongahela
Contingent
Ranking
Survey
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Selected
Results
for
the
Random
Utility
Model
With
Ranked
Logit
Estimator
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Comparison
of
Ordered
Logit
and
Keener­
Waldman
Ordered
Normal
ML
Estimator
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Benefit
Estimates
from
Contingent
Ranking
Models
.
.
.
.
.
.

Hedonic
Wage
Models
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Summary
of
Predicted
Hourly
Wage
Rates
.
.
.
.
.
.
.
.
.
.
.
The
Characteristics
of
the
Sites
and
the
Survey
Respondents
Selected
from
the
Federal
Estate
Survey
.
.
.
.
.
Regression
Results
of
General
Model,
by
Site
.
.
.
.
.
.
.
.
.
Summary
of
Cicchetti,
Seneca,
and
Davidson
[
1969]
Participation
Models
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Comparison
of
Basic
Model
with
Tailored
Model
:
Coefficient
for(
TC+
MC)
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
F­
Test
for
Restriction
of
General
Model
.
.
.
.
.
.
.
.
.
.
.
F­
Test
for
Restriction
of
Tailored
Models
.
.
.
.
.
.
.
.
.
.
Effects
of
Truncation
on
the
Travel
Cost
Models'
Estimates
.
.
Two­
Stage
Least­
Squares
Estimates
for
Selected
Travel
Cost
Site
Demand
Models...
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Comparison
of
Ordinary
Least­
Squares
and
Two­
Stage
Least­
Squares
Estimates
of
Travel
Cost
(
TC.
+
MC.
)
Parameters
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Hausman
~
est
f~
r
Differences
Between
Two­
Stage
Least­
Squares
and
Ordinary
Least­
Squares
Estimates
.
.
.
.
.
Description
of
U.
S.
Army
Corps
of
Engineers
Data
on
Site
Characteristics
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Generalized
Least­
Squares
Estimates
of
Determinants
of
Site
Demand
Parameters
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Recreation
Sites
on
the
Monongahela
River
.
.
.
.
.
.
.
.
.
.
Dissolved
Oxygen
Levels
for
Recreation
Activities
.
.
.
.
.
.
Mean
and
Range
of
Benefit
Estimates
for
Water
Quality
Scenarios
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Consumer
Surplus
Loss
Due
to
the
Loss
of
Use
of
the
Monongahela
River
by
Survey
Users'
Income
.
.
.
.
.
.
.
.
.
Consumer
Surplus
Loss
Due
to
Loss
of
Use
of
the
Monongahela
River
by
Survey
Users'
Travel
Cost
.
.
.
.
.
.
.
5­
28
5­
30
5­
31
5­
32
6­
11
6­
18
6­
21
6­
24
6­
27
7­
26
7­
27
7­
28
7­
33
7­
36
7­
37
7­
39
7­
42
7­
45
7­
48
7­
49
7­
50
7­
53
7­
56
7­
57
7­
60
7­
61
7­
62
7­
62
xii
TABLES
(
continued)

Number
7­
20
7­
21
8­
1
8­
2
8­
3
8
­
4
8
­
5
Consumer
Surplus
Increments
Due
to
Water
Quality
lmprovement­­
Boatable
to
Fishable
by
Survey
Users'
Income
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

Consumer
Surplus
Increment
Due
to
Water
Quality
lmprovement­­
Boatable
to
Swimmable
by
Survey
Users'
Income
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

Predicted
Demand
Parameters
for
Monongahela
Sites.
.
.
.
.
.
Bishop­
Heberlein
Comparative
Results
for
Benefit
Approaches
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
A
Comparison
of
Benefit
Estimates
for
Water
Quality
Improvements
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
A
Comparison
of
Contingent
Valuation
and
Generalized
Travel
Cost
Benefit
Estimates
.
.
.
.
,
.
.
.
.
.
.
.
.
.
.
.
A
Comparison
of
Contingent
Valuation
and
Contingent
Ranking
Benefit
Estimates
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Page
7­
63
7­
63
8
­
8
8­
10
8­
13
8­
16
8­
19
.
.
.
X111
."
 
 
7
CHAPTER
1
INTRODUCTION,
OBJECTIVES,
AND
SUMMARY
1.1
INTRODUCTION
This
Research
Triangle
Institute
(
RTI
)
report
to
the
U.
S.
Environmental
protection
Agency
(
EpA)
compares
alternative
approaches
for
estimating
the
recreation
and
related
benefits
of
water
quality
improvements.
The
results
provide
information
on
the
performance
of
various
ways
to
estimate
the
benefits
of
environmental
quality
improvements,
so
EPA
can
use
such
methods
in
preparing
the
regulatory
impact
analyses
required
by
Executive
Order
12291
and
in
evaluating
other
regulatory
proposals.
This
report
is
also
relevant
to
the
proposed
revision
of
the
Federal
water
quality
standards
regulations,
which
recommends
that
States
consider
incremental
benefits
and
costs
in
setting
their
water
quality
standards.
Site­
specific
water
quality
standards
are
likely
to
play
an
important
role
in
future
water
policy
issues
because
they
bring
together
the
crucial
elements
of
appropriate
stream
uses
and
advanced
treatment
requirements
for
municipalities
and
industries.
Benefit­
cost
assessments
can
yield
valuable
information
for
these
decisions.

Evaluations
of
benefits
and
costs
depend
on
a
determination
of
the
links
between
regulatory
policy,
technical
effects,
and
behavioral
responses.
Figure
1­
1
illustrates
one
set
of
linkages
­­
in
this
case
for
the
proposed
water
quality
standards
regulations.
This
report
addresses
the
last
component
of
Figure
1­
1,
which
involves
estimating
monetized
benefits
for
regulatory
policy.
One
of
the
difficulties
in
such
a
task
arises
from
the
absence
of
organized
markets
for
many
of
the
services
derived
from
water
resources.

The
benefits
of
water
resource
regulations
are
usually
measured
with
one
of
three
types
of
approaches:
(
1)
market­
based
approaches,
which
use
indirect
I
in
kages
between
the
environmental
goods
and
some
commodities
exchanged
in
markets;
(
2)
contingent
valuation
approaches,
which
establish
an
institutional
framework
for
a
hypothetical
market;
and
(
3)
public
referenda.
This
report
considers
the
first
two
approaches;
the
last
is
omitted
since
it
is
beyond
EPA's
mandate.

Some
opponents
argue
that
benefit­
cost
analysis
is
invalid
because
it
cannot
measure
all
of
the
benefits
of
environmental
regulations.
Nevertheless,
this
report
describes
the
measurement
of
several
benefits
from
water
quality
improvements,
including
some
regarded
as
unmeasurable
in
earlier
environmental
benefits
research
efforts.
Specifically,
as
highlighted
in
Figure
1­
2,
this
study
considers
both
the
recreation
benefits
that
accrue
to
users
of
a
recreation
site
and
the
intrinsic
benefits*
that
accrue
to
both
users
and
nonusers.

*
This
classification
modifies
the
one
in
Mitchell
and
Carson
[
1981]
.

1­
1
rl
Changs
Oasignatad
Lb(
s)

II
Watar
@
diW
Regulatory
Action(
s)

v
i
Modify
Critaria
to
Provida
for
Osaignatad
Usa(
s)

f
t
I
rl
Chan~
in
Effluants
I
I
I
Cfsangas
in
Vktar
Ouality
I
­'
O"
A"
iOn(
s~
*
Taohnioal
Effects
of
Watar
Ouality
L
+
Effaots
on
Economic
Agarsts
­
w=(')<
nEiiicl
Bahaviod
Effacts
of
Watar
Ouallw
Figure
1­
1.
Effects
and
responses
to
water
quality
regulatory
actions.

User
benefits
arise
from
recreation
uses
of
the
river
and
are
measured
by
users'
willingness
to
pay
for
the
water
quality
levels
necessary
to
permit
these
recreation
uses.
That
is,
the
valuation
depends
on
the
use
of
the
river.
In
this
case,
clean
water
in
a
river
is
worth
something
because
recreationists
are
going
to
fish,
boat,
swim
in,
or
picnic
along
the
river.

Intrinsic
benefits
consist
of
two
value
types:
option
value
and
existence
value.
Relevant
to
both
current
users
and
potential
future
users,
option
value
is
the
amount
an
individual
would
be
willing
to
pay
for
improved
water
quality
(
over
his
expected
user
values)
to
have
the
right
to
use
the
river
in
the
future
when
there
is
uncertainty
either
in
the
river's
availability
at
a
particular
quality
level
or
in
his
use
of
it
(
with
the
river
meeting
specified
water
quality
conditions).
For
example,
if
an
individual
might
use
the
river,
but
is
not
sure
he
will,
he
may
pay
some
amount
each
year
for
the
right
(
or
option)
to
use
it
(
with
the
river
meeting
specified
water
quality
conditions).
Under
some
conditions,
this
payment,
the
option
price,
will
exceed
his
expected
consumer
surplus
­­
the
value
he
would
derive
from
anticipated
use.
This
excess­­
the
amount
that
the
option
price
exceeds
the
expected
consumer
surplus­­
is
defined
as
the
option
value.

1­
2
potential
Water
Quality
Benefits
Current
User
Benefits
Intrinsic
Benefits
Direct
use
Indirect
use
Potential
use
No
Use
+
Recreational
l
 
fishing,
swimming,
boating,
rafting,
etc.
In
Stream
Commercial
 
fishing,
navigation
{
Municipal
 
drinking
water,
waste
disposal
Withdrawal
Agricultural
 
irrigation
Industrial/
Commercial
 
cooling,
process
treatment,
waste
disposal,
steam
generation
{
Recreational*
 
hiking,
picnicking,
birdwatching,
photography,
etc.
Near
Stream
Relaxation*
 
viewing
Aesthetic"
 
enhancement
of
adjoining
site
amenities
+
Near­
term
potential
use
Option*

Long­
term
potential
use
­
i
Stewardship
 
maintaining
a
good
environment
for
everyone
to
enjoy
(
including
future
Existence*
family
use
 
bequest)

Vicarious
consumption
 
enjoyment
from
the
knowledge
that
others
are
using
the
resource.

l
 
Considerad
in
this
project.

Figure
1­
2.
A
spectrum
of
water
quality
benefits.

Existence
value,
on
the
other
hand,
is
an
individual's
willingness
to
pay
for
the
knowledge
that
a
resource
exists.
That
is,
an
individual­­
either
a
user
or
a
nonuser­­
might
be
willing
to
pay
something
to
maintain
a
high
level
of
water
quality
at
a
recreation
site
in
a
particular
area,
even
though
he
will
not
use
it,
so
that
his
children
may
have
future
use
of
the
site
or
simply
to
know
that
the
ecosystem
at
the
site
will
be
maintained.

This
study's
comparison
of
alternative
benefits
measurement
approaches
estimates
user
values
by
the
travel
cost
approach
(
indirect
method),
by
four
different
ways
of
eliciting
option
price
in
a
contingent
valuation
experimental
design
(
direct
method),
and
by
a
contingent
ranking
of
water
quality
outcomes
and
option
price
amounts.
The
central
comparison
evaluates
whether
there
are
differences
between
approaches
because
"
true"
values
for
each
of
these
types
of
benefits
are
unknowns.
In
addition,
since
the
other
methods
are
not
suitable
for
measuring
them,
option
and
existence
values
are
compared
only
in
terms
of
alternative
ways
for
posing
the
hypothetical
questions.

A
distinguishing
feature
of
this
project
is
its
use
of
a
case
study
of
the
Pennsylvania
portion
of
the
Monongahela
River
as
the
point
of
reference
for
1­
3
both
the
comparison
of
approaches
and
the
estimation
of
option
and
existence
values.
The
Monongahela
is
representative
of
a
number
of
rivers
in
the
country,
has
multiple
uses,
and
has
recently
been
the
focus
of
effluent
guidelines
for
the
iron
and
steel
industry.
The
survey
design
for
the
Monongahela,
calling
for
a
household
survey,
is
a
middle
ground
between
the
macro
approach
for
estimating
benefits
of
water
po{
lution
controls
(
see
Mitchell
and
Carson
[
1981
]
)
and
the
user
orientation
of
many
micro
contingent
valuation
efforts
(
see
Schulze,
d'Arge,
and
Brookshire
[
1981]).
The
design
uses
a
representative
sample
of
households
for
the
region
and,
similar
to
Mitchell­
Carson,
includes
both
user
and
intrinsic
benefits.
It
also
is
a
specific
application,
considering
individuals'
willingness
to
pay
for
a
specific
river
basin's
water
quality.

1.2
OBJECTIVES
The
potential
implications
of
this
study
for
water
policy
dictated
clearly
defined
objectives
and
a
project
design
to
achieve
them.
The
overall
objective
of
this
project
was
to
conduct
a
study
comparing
alternative
approaches
for
estimating
the
recreation
and
related
benefits
of
different
water
quality
levels.
In
particular,
the
study
sought
to
measure
user,
option,
and
existence
values
for
the
Pennsylvania
segment
of
the
Monongahela
River
and
to
estimate
the
recreation
and
related
benefits
that
would
be
derived
from
providing
different
use
classifications
(
fishable,
swimmable,
beatable)
for
this
river
segment.

In
addition
to
meeting
its
own
specific
objectives,
an
environmental
benefits
research
project
ideally
would
fit
the
needs
of
those
involved
in
the
evaluation
of
public
policy
questions
and
the
needs
of
the
research
community
in
general.
Since
the
most
important
direct
use
of
natural
environments
is
for
water­
based
recreation
(
see
Freeman
[
1979a]
),
this
project's
general
research
area
considers
one
of
the
primary
components
of
environmental
benefits
research.
In
addition
to
its
water
quality
orientation,
the
project
is
also
relevant
to
two
areas
Freeman
identified
for
future
research:

I
think
that
a
major
research
effort
should
be
made
to
select
an
appropriate
area
and
water
bodies
for
a
study,
to
develop
a
properly
specified
model,
and
to
gather
the
necessary
data.
Until
such
an
effort
is
made,
t
h
e
p
r
a
c
t
i
c
a
l
i
t
y
o
f
t
h
e
Clawson­
Knetsch
[
1
9
6
6
]
[
travel
cost]
technique
for
estimating
recreation
benefits
will
remain
an
open
question.
[
p.
256]

There
should
be
carefully
conducted
experiments
with
the
survey
techniques
for
estimating
willingness
to
pay
for
reduction
in
pol
Iu
­
tion.
These
experiments
should
be
coordinated
with
studies
based
on
other
analytical
techniques
in
an
effort
to
provide
a
cross­
check
or
validation
of
benefit
estimates
obtained
by
different
approaches.
[
p.
265]

.

1
­
4
1
.
3
S
U
M
M
A
R
Y
O
F
R
E
S
U
L
T
S
This
section
summarizes
at­
e
presented
for
individual
preaches.

1
.3.1
Overview
the
major
findings
of
the
research.
The
findings
approaches
and
for
the
comparison
between
ap
­

The
results
of
this
project
strongly
support
the
feasibility
of
measuring
the
recreation
and
related
benefits
of
water
quality
improvements.
Moreover,
the
benefits
measurement
approaches
­­
several
contingent
valuation
formats
and
the
travel
cost
method
­­
show
consistent
results
for
comparable
changes
in
water
quality.
Indeed,
the
range
of
variation
is
generally
less
than
that
expected
in
models
used
to
transiate
the
effects
of
effluents
in
a
water
body
into
the
corresponding
water
quality
parameters.
In
addition,
the
results
also
clearly
show
that
the
intrinsic
benefits
of
water
quality
improvements­­
especially
option
values
­­
can
be
measured
and
that
they
are
a
sizable
portion
­­
greater
than
half
­­
of
the
total
recreation
and
related
benefits
total
.

1
.3.2
Contingent
Valuation
Ap
preach
Based
on
the
results
of
the
Monongahela
River
case
study,
the
general
prognosis
is
good
for
the
continued
use
of
the
contingent
valuation
approach
to
estimate
the
benefits
of
water
quality
improvements.
Statistical
analysis
using
regression
methods
to
evaluate
the
determinants
of
the
variation
in
the
option
price
bids
gave
little
indication
that
individual
interviewers
influenced
the
results.
The
consistently
plausible
signs
and
magnitudes
of
key
economic
variables
suggest
that
the
respondents
perceived
the
survey
structure
as
realistic
and
did
not
experience
problems
with
the
hypothetical
nature
of
some
of
the
questions.
These
findings
were
realized
despite
the
fact
that
the
sample
included
households
whose
socioeconomic
profile
was
comparable
to
demographic
groups
that
were
found
to
be
more
difficult
respondents
in
past
contingent
valuation
surveys.
On
average,
the
respondents
were
older,
less
educated,
and
poorer
than
those
in
the
most
successful
contingent
valuation
studies.

The
contingent
valuation
estimates
of
the
option
price
for
water
quality
improvements,
which
include
user
and
option
values
,
are
consistently
plausible
across
the
various
analytical
approaches,
with
estimates
for
the
combined
water
quality
levels
ranging
from
roughly
$
50
to
$
120
per
year
per
household
sampled
in
the
Monongahela
River
basin.
Nonetheless,
the
empirical
results
do
indicate
that
the
methods
used
to
elicit
the
willingness­
to­
pay
amount
have
a
statistically
significant
effect
on
the
estimates
of
willingness
to
pay.
For
example,
both
the
direct
question
with
a
payment
card
and
the
bidding
game
with
a
$
125
starting
point
produced
higher
willingness­
to­
pay
estimates
than
either
the
direct
question
without
an
aid
or
the
bidding
game
with
a
$
25
starting
point.
Thus,
there
is
some
evidence
of
starting
point
bias
in
the
bidding
game,
but
the
statistical
analyses
are
not
conclusive.
The
results
comparing
the
two
bidding
game
methods
as
a
set
(
i.
e.
,
those
with
$
25
and
$
125
starting
points)
with
the
non
bidding
games
(
direct
question
and
payment
card
combined)
indicated
no
differences
between
these
two
sets
of
approaches.

1­
5
The
findings
provide
clear
support
for
a
positive,
statistically
significant,
and
sizable
option
value
for
water
quality
improvements
along
the
Monongahela
River.
The
estimated
option
values
for
loss
of
the
use
of
the
area
in
its
current
condition
(
i
.
e.
,
boatable)
range
from
approximately
$
15
to
$
60
per
year
per
household,
and
the
option
values
for
improving
water
quality
to
a
swimmable
level
range
from
approximately
$
20
to
$
45
per
year
per
household.
Thus,
option
value
is
a
substantial
fraction
of
the
user's
option
price,
and
the
value
of
this
change
in
water
quality
generally
exceeds
user
values.

The
survey
also
provided
estimates
of
existence
values.
Unfortunately,
respondents
did
not
necessarily
understand
the
distinction
sought.
Many
bid
the
same
amounts
as
they
had
earlier
on
the
option
price
for
a
comparable
change
in
water
quality.
It
is
not
clear
whether
these
responses
were
deliberate
or
a
reflection
of­
misunderstanding
of
the
questions.
Thus,
while
the
findings
suggest
that
these
values
are
positive
and
statistically
significant,
prudence
requires
they
be
interpreted
cautiously.

Of
course,
it
should
also
be
acknowledged
that
the
available
estimates
of
intrinsic
values
are
quite
limited.
Most
can
be
criticized
for
problems
in
the
research
design,
including
possible
flaws
in
the
survey.
The
design
of
the
Monongahela
River
study
relies
on
the
use
of
a
schematic
classification
of
the
sources
of
an
individual's
valuation
of
the
river
(
i.
e.
,
a
card
showing
different
types
of
values)
in
eliciting
a
division
of
user
and
other
benefits.
Because
this
is
the
first
application
of
this
device,
it
was
not
possible
to
evaluate
its
effectiveness.

In
addition
to
the
more
widely
used
bidding
game
and
direct
question
formats
for
contingent
valuation
experiments,
t
h
e
Monongahela
R
i
v
e
r
b
a
s
i
n
survey
also
applied
the
contingent
ranking
format.
This
format
requires
only
that
individuals
rank
combinations
of
water
quality
levels
and
option
prices
and
uses
a
statistical
procedure
(
ranked
order
logit)*
to
estimate
willingness
to
pay.
While
other
contingent
valuation
formats
require
that
individuals
directly
provide
willingness
to
pay,
contingent
ranking
asks
them
to
rank
hypothetical
outcomes.
In
effect,
it
asks
a
simpler
task
of
the
respondent­­
only
to
rank
outcomes
­­
but
requires
more
sophisticated
and
less
direct
techniques
to
estimate
the
value
of
the
outcomes.

Since
use
of
the
contingent
ranking
format
to
estimate
the
benefits
of
environmental
quality
improvements
is
quite
new,
the
behavioral
model
underlying
its
estimation
procedures
is
also
early
in
its
development.
Although
this
project
provides
a
description
of
these
underpinnings,
its
evaluation
of
the
theoretical
properties
and
practical
issues
is
incomplete.
Overall,
the
findings
of
this
study
suggest
that,
even
though
the
behavioral
models
used
to
derive
benefits
estimates
with
the
contingent
ranking
format
were
somewhat
arbitrary,
the
results
from
the
ranking
format
closely
parallel
other
contingent
valuation
estimates.

*
ln
more
technical
terms,
the
procedure
uses
a
specification
for
the
indirect
utility
function
together
with
a
maximum
Ii
kelihood
estimator.

1­
6
The
mean
estimates
derived
from
the
contingent
ranking
format­­
roughly
$
6
0
annuallY
Per
household
for
imProvin9
water
quality
in
the
Monongahela
to
fishable
and
approxlamteiy
$
50
more
annUallY
for
improving
it
to
swimmable­­
appear
l
a
r
g
e
r
t
h
a
n
those
derived
with
other
contingent
valuation
formats.
However,
these
differences
are
not
statistically
significant.
In
addition,
the
benefit
estimates
from
all
continent
vaiuation
formats
are
comparable
across
individuals,
with
the
primary
differences
between
contingent
ranking
and
other
methods
stemmin9
from
the
questioning
format
used
in
the
other
methods.

1
.3.3
Travel
Cost
Ap
preach
This
study
also
developed
and
used
a
generalized
travel
cost
model
to
predict
the
recreation
benefits
of
water
quality
improvements
at
recreation
sites.

The
travel
cost
model
assumes
that
site
features
or
attributes
affect
both
an
individual's
ability
to
participate
in
recreation
activities
at
any
particular
site
and
the
quality
of
his
recreation
experiences
at
the
site.
In
considering
the
demand
for
a
recreation
site
as
a
derived
demand,
the
common
sense
rationale
of
the
model
suggests
that
a
recreation
site's
features
or
attributes
will
influence
the
demand
for
its
services.
Since
the
level
of
water
quality
i
s
a
site
attribute,
a
basis
is
established
for
relating
water
changes
to
shifts
in
demand
for
a
recreation
site's
services.

The
generalized
model
was
estimated
from
data
on
43
water­
based
recreation
sites
in
the
Federal
Estate
Survey
component
of
the
1977
National
Outdoor
Recreation
Survey.
This
survey
provided
information
on
recreation
use
patterns
at
each
site
during
a
single
season.
Based
on
sample
sizes
for
each
site
that
ranged
from
approximately
30
to
several
hundred
respondents,
t
h
e
survey
described
individuals'
recreation
behavior,
socioeconomic
characteristics
travel
time
necessary
to
reach
the
site
of
other
factors.
,
residential
location,
and
a
variety
Several
advantages
of
this
travel
cost
model
include:

.
Deriving
individual
estimates
for
the
time
associated
with
traveling
to
the
site
as
well
as
the
roundtrip
distance
for
each
t
r
i
p
.

.
Using
the
opportunity
cost
of
time
to
evaluate
travel
time
and
estimating
opportunity
cost
for
each
individual
based
on
his
characteristics,
inclu­
ding
age,
education,
race,
occupation.

.
Considering
for
each
site
the
potential
effects
of
differences
in
onsite
time
per
visit.

The
generalized
model
was
used
to
estimate
the
benefits
Monongahela
R
i
v
e
r
,
as
identified
in
the
survey
of
the
basin.
model
predicted
a
value
of
$
83
per
year
per
user
household
sex,
and
individuals'

for
users
of
the
The
travel
cost
if
a
decrease
in
1­
7
­.
 
 
.
..
water
quality
is
avoided
and
a
value
of
$
15
per
year
for
each
user
household
[
if
water
quality
is
improved
to
a
swimmable
level.

i$

Several
features
of
the
generalized
travel
cost
model
are
of
particular
importance:
it
provides
a
framework
for
estimating
the
value
of
water
quality
improvements
for
a
substantial
range
of
sites,
and
its
site­
specific
orientation
is
especially
relevant
for
water
quality
standards
applications.
Finally,
it
includes
the
effect
of
key
site
features­­
like
access
and
facilities­­
and
can
use
data
frequently
available
in
the
public
domain.

1
.3.4
Approach
Comparison
One
of
the
primary
objectives
of
this
research
has
been
to
compare
available
approaches
for
measuring
the
benefits
of
water
quality
improvement.
Such
a
comparison
­­
reflecting
the
assumptions
inherent
in
each
approach­­
will
show
the
plausibility
of
the
required
assumptions
as
descriptions
of
real­
world
behavior
and
constraints.
However,
since
the
"
true"
value
of
water
quality
improvement
benefits
is
unknown,
a
comparison
cannot
be
interpreted
as
a
validation
of
any
one
approach.
On
the
other
hand,
an
evaluation
of
the
c
o
m
­
parability
of
estimates
across
approaches
that
considers
the
reasons
for
their
consistencies
and
differences
provides
a
basis
for
an
improved
use
of
benefit
methodologies.
Consistency
also
would
give
increased
flexibility
in
matching
a
method
to
available
data
for
each
particular
application.

Based
on
the
research
for
the
Monongahela
River
basin
case
study,
the
comparison
between
the
travel
cost
and
contingent
valuation
approaches
is
the
most
interesting.
Estimates
of
benefits
from
water
quality
improvement
are
compared
for
the
69
users
identified
in
the
survey
of
households
in
the
basin
/

area.
Previous
comparisons
of
approaches
relied
on
the
use
of
mean
estimates
c
from
each
method.
When
these
means
are
compared,
it
is
assumed
that
all
individuals
can
be
treated
as
drawing
from
populations
with
the
same
mean
benefits.
Differences
in
individuals
or
error
in
the
pairing
of
means
can
lead
to
a
confounding
of
the
benefit
comparisons.
In
contrast,
this
study
compared
each
household's
user
value,
derived
from
the
contingent
valuation
survey,
with
the
corresponding
estimate
for
that
household
from
the
travel
cost
model.
Thus,
this
study
gives
a
"
more
controlled
comparison
than
was
possible
in
earlier
studies.
~
Table
1­
1
shows
the
mean
benefit
estimates
of
user
values
for
the
travel
cost
and
contingent
valuation
approaches.
On
theoretical
grounds,
the
contin­
[
gent
valuation
estimates
of
compensating
surplus
should
be
less
than
the
travel
1
cost
estimates
based
on
ordinary
consumer
surplus,
but
the
differences
should
be
slight
due
to
the
small
income
effects
found
in
the
research.
However,
\

this
is
not
the
case
for
three
out
of
four
contingent
valuation
estimates
for
improvements
in
water
quality.
Only
the
estimates
derived
with
the
$
25
bidding
game
format
are
less
than
the
travel
cost
estimates,
although
the
travel
cost
estimates
are
within
the
range
of
contingent
valuation
estimates.
For
the
loss
of
the
area,
the
means
comparison
conforms
to
theoretical
expectations,
with
the
travel
cost
estimates
larger
than
the
contingent
valuation
estimates.
Most
of
the
differences
between
approaches
exceed
the
size
expected
from
theory.
At
best,
simple
comparisons
of
mean
estimates­­
augmented
by
a
priori
information
­­
are
rough
judgments
of
plausibility.
On
the
basis
of
this
comPar­

11
1­
8
Table
1­
1.
A
Comparison
of
Mean
Benefit
Estimates
(
1981
Dollars)
a
Water
quality
changeb
Boatable
Boatable
Approach
Loss
of
area
to
fishable
to
swimmable
Contingent
valuation
Direct
question
19.71
(
17)
21.18
(
17)
31.18
(
17)

payment
card
19.71
(
17)
30.88
(
17)
51.18
(
17)

iterative
bidding
($
25)
6.59
(
19)
4.21
(
19)
10.53
(
19)

iterative
bidding
($
125)
36.25
(
16)
20.31
(
16)
48.75
(
16)

Generalized
travel
cost
82.65
(
94)
7.01
(
94)
14.71
(
94)

a
The
travel
cost
estimates
were
converted
from
1977
to
1981
dollars
using
the
consumer
price
index
for
December
1981,
the
last
month
of
the
survey.

bThe
numbers
in
parentheses
after
the
means
are
the
number
of
observations
on
which
each
of
these
estimates
was
based.
The
number
for
the
travel
cost
estimates
exceeds
the
sum
of
the
sample
size
for
the
contingent
valuation
results
because
some
users
visited
more
than
one
Monongahela
River
site.

ison,
however,
the
Monongahela
River
basin
precise.

A
more
discriminating
comparison
of
the
estimates
are
plausible,
but
not
travel
cost
and
contingent
valuation
approaches,
one
that
judges
how
the
two
approaches
compare
across
individuals
is
also
possible
with
the
Monongahela
River
basin
benefit
estimates.
In
this
comparison,
presented
in
Table
1­
2,
the
contingent
valuation
measure
of
user
value
was
regressed
on
the
travel
cost
estimate
(
see
Chapter
8
for
details).
The
~
priori
expectations
of
comparability
in
methods
can
be
structured
as
two
statistical
tests.
These
models
also
take
account
of
the
effect
of
question
formats
used
in
the
contingent
valuation
survey.

The
results
from
the
regression
tests
generally
reinforce
the
earlier
conclusions
based
on
comparing
the
means
estimated
from
each
method.
Several
additional
conclusions
are
possible
from
these
comparisons:

.
The
contingent
valuation
estimates
of
water
quality
improvements
overstate
willingness
to
pay­­
in
contrast
to
the
theoretical
expectation
s­­
but
the
results
do
not
permit
a
judgment
of
statistically
significant
differences
between
the
two
sets
of
estimates
Some
caution
is
required,
however,
because
the
properties
of
the
statistical
tests
are
approximate.

.
The
travel
cost
model
overstates­
­
by
an
amount
greater
than
theory
would
predict­­
willingness
to
pay
for
the
loss
of
the
area,
and
the
estimates
are
not
comparable
to
the~
ntingent
valuation
estimates.
,

1­
9
r
Table
1­
2.
Regression
Comparisons
of
Contin~
ent
Valuation
and
Travel
Cost
Benefit
Estimates
Water
quality
change
Independent
Boatable
to
Boatable
variables
Loss
of
area
game
fishing
to
swimming
Intercept
Travel
cost­
benefit
estimate
Qualitative
variables
Payment
card
Direct
question
Iterative
bid
($
25)

R
2
F
2
1
.
8
6
(
1.37)

0.33
(
1.17)
b
(
­
4
.
3
6
)

­
32.64
(
­
2
,
5
5
)

­
14.60
(­
1
.27)

­
31.82
(
­
2
.
5
5
)

0.10
2.42
(
0.05)
C
33.99
(
1.90)

5
1
.
7
6
(
2.64)

12.96
(
0.75)

­
11.24
(
­
0
.
6
0
)

0.12
3.00
(
0.02)
C
59.57
(
2.02)

­
2.71
(­
1.14)
b
(­
1
.79)

77.01
(
2.36)

21.00
(
0.73)

­
21.82
(
­
0
.
6
9
)

0.11
2.62
(
0.04)
C
a
The
numbers
below
the
estimated
coefficients
are
t­
ratios
for
the
null
hypothesis
of
no
association.
b
These
statistics
are
the
t­
ratios
for
the
hypothesis
equivalent
to
unity
for
the
slope
coefficientl%
r
Ordinary
Consumer
Surplus
(
OCS)
after
adjustment
is
made
for
the
fact
that
Compensating
Surplus
(
CS)
is
measured
in
1981
dollars
and
OCS
in
1977
dollars.

c
The
number
in
parentheses
below
the
reported
F­
statistic
is
the
level
of
significance
for
rejection
of
the
null
hypothesis
of
no
association
between
the
dependent
and
independent
variables.

.
The
comparative
performance
of
the
contingent
valuation
approach
in
relationship
to
the
travel
cost
method
is
sensitive
to
differences
in
question
format­­
with
the
clearest
distinctions
found
between
the
payment
card
and
the
bidding
game
with
the
$
125
starting
point.

.
The
explanatory
power
of
the
models
used
in
the
comparison
are
not
high,
but
the
null
hypothesis
of
no
association
between
methods
is
clearly
rejected
at
high
levels
of
significance
(
based
on
the
F­
tests
reported
at
the
bottom
of
the
table).

1­
10
1
.3.5
Considerations
for
Future
Research
T
h
e
findings
of
this
project
also
suggest
that
there
are
a
number
of
areas
for
future
benefits
research,
including
both
general
and
specific
issues­­
especially
those
concerned
with
particular
benefits
measurement
approaches.

Genera]
!
SSUeS
Option
and
existence
values
remain
the
most
difficult
general
issues
to
address
adequately.
The
research
design
for
this
project
relied
on
the
individual
to
divide
the
hypothetical
option
price
payment
into
its
user
and
option
value
components
and
then
to
add
existence
values
to
these
option
price
bids
as
an
incremental
premium.
Other
studies
(
Brookshire,
Cummings,
et
al.
[
1982]
and
Randall,
Hoehn,
and
Tolley
[
1981
]
)
have
elicited
preservation
values
­­
including
both
option
and
existence
values­­
as
additions
to
user
values.
Mitchell
and
Carson
[
1981
]
found
user
values
by
subtracting
nonuser's
option
price
payments
from
user
option
price
payments.
Regardless
of
the
procedures,
however,
all
these
studies
have
found
option
and
existence
values
to
be
substantial­­
greater
than
half
of
the
total
benefits
of
environmental
improvements.
The
choice
among
elicitation
procedures,
remains
an
open
question.

One
question
that
arises
from
the
results
of
this
and
other
recent
studies
of
intrinsic
benefits
is,
"
Why
worry
about
measuring
option
value
when
it
is
possible
to
elicit
option
price
bids
that
include
it?"
Empirical
estimates
are
of
interest
because
of
the
controversy
over
the
sign
and
magnitude
of
option
value
that
has
arisen
in
the
theoretical
literature.
I
n
addition,
many
practical
applications
of
benefit
methods
do
not
measure
intrinsic
benefits,
suggesting
a
need
for
empirical
estimates
to
gauge
the
extent
of
the
omitted
portion
of
benefits
from
particular
environmental
policies.
The
early
theoretical
work
seemed
to
imply
(
without
explicitly
stating
this
conclusion)
that
option
values
would
be
small
in
comparison
to
user
values.
Recent
theoretical
work
by
Freeman
[
1982]
makes
a
case
for
positive
option
values
and
confirms
this
presumption
by
suggesting
that
option
values
should
be
small
under
almost
all
conditions.
Only
by
attempting
to
distinguish
between
option
and
other
intrinsic
values
will
it
be
possible
to
bring
some
empirical
evidence
to
bear
on
this
question.

Proportional
relationships
between
user
and
intrinsic
values
from
earlier
studies
have
often
been
used
in
attempts
to
infer
the
size
of
the
omitted
benefits
when
the
intrinsic
values
are
not
directly
estimated.
The
limited
resources
available
for
many
public
policy
evaluations
is
the
primary
reason
for
the
widespread
use
of
the
proportional
approach.
Since
it
is
unlikely
that
these
constraints
on
evaluations
will
ease
in
the
future,
more
empirical
research
on
the
use
and
size
of
these
proportions
might
be
productive.
For
instance,
determining
how
(
and
if)
the
proportions
differ
for
certain
classes
of
assets
­­
ranging
from
unique
natural
environments
to
waterbodies
with
numerous
substitutes­­
would
provide
useful
guidance
for
applying
these
proportions.
Moreover,
attempting
to
distinguish
between
option
and
existence
values
for
different
classes
of
environmental
assets
may
indicate
the
feasibility­­
and
need­­
for
such
distinctions
(
see
Fisher
and
Raucher
[
1982]
for
a
review).

1­
11
.
The
research
in
this
project
has
skirted
another
important
issue­­
benefits
aggregation.
The
travel
cost
model
used
in
this
project
predicts
recreation
site
benefits
for
"
the
representative"
user.
By
assuming
that
all
sites
are
possible
substitutes
(
because
one
site's
attributes
can
be
"
repackaged"
to
be
equivalent
to
any
other
site),
it
implicitly
maintains
a
simplistic
view
of
the
relationship
between
recreation
sites
within
a
region.
Individuals
always
select
the
site
providing
the
desired
mix
of
attributes
at
the
lowest
implicit
price.
Clearly,
not
all
sites
adhere
to
these
relationships.
For
example,
a
historical
monument
at
the
site
may
make
it
unique.
What
is
needed
is
a
more
general
characterization
that
would
accommodate
sites
not
conforming
to
the
aggregation
rule
used
to
relate
effective
site
services
to
site
attributes.
Such
a
framework
would
explain
the
relationship
between
an
individual's
patterns
of
site
usage
for
facilities
permitting
very
different
types
of
recreation
activities
(
e.
g.,
water­
based
recreation
versus
skiing).
Nevertheless,
consistent
regional
and
national
benefit
estimates
will
require
a
careful
description
of
the
interrelationships
between
the
individual's
demands
for
different
types
of
recreation
sites.

Another
unresolved
issue
involves
regional
aggregation
of
local
benefits
estimated
with
the
contingent
valuation
approach.
Conventional
practice
in
statistical
surveys
is
to
use
statistical
weights,
which
reflect
the
probability
of
selecting
a
particular
sampling
unit,
to
estimate
aggregate
benefits
for
t
h
e
representative
population
(
see
Mitchell
and
Carson
[
1981
]).
H
o
w
e
v
e
r
,
t
h
i
s
approach
raises
fundamental
problems
with
the
conventional
practice
in
economic
modeling
that
assumes
common
(
and
constant)
parameters
across
individuals
for
correctly
specified
behavioral
models.
The
definition
of
a
representative
sample
is
often
based
on
a
description
of
statistical
models,
leading
to
observed
data
that
are
at
variance
with
conventional
economic
modeling.
More
research
following
the
work
of
Porter
[
1973]
is
needed
to
consider
t
h
e
relevance
of
this
issue
for
the
extrapolation
of
contingent
valuation
estimates.

Another
general
research
issue
involves
comparing
alternative
benefit
estimation
approaches.
This
project's
comparison,
which
examines
benefits
predicted
with
the
generalized
travel
cost
model
and
contingent
valuation
will
­
ingness­
to­
pay
estimates
for
the
same
individuals,
permitted
a
fairly
direct
comparison
of
estimates
with
theoretical
bounds.
However,
this
study
used
estimates
from
only
69
users
of
the
Monongahela
River.
A
comparison
having
a
larger
number
of
users
and
based
on
a
water­
based
recreation
site
with
a
greater
diversity
of
users
would
provide
a
more
revealing
comparison.
Indeed,
following
Bishop
and
Heberlein,
attempts
to
compare
simulated
market
results
with
the
results
of
this
project
also
may
shed
light
on
the
relationships
among
the
estimation
approaches.
Before
these
comparisons
are
made,
however,
more
systematic
attention
should
be
given
to
the
theoretical
underpinnings
of
t
h
e
approaches,
following
the
work
of
Schulze
et
al.
[
1981],
Smith
and
Krutilla
[
1982],
and
Bockstael
and
McConnell
[
1982].

Future
research
should
also
reconsider
the
economic
principles
underlying
comparisons
of
economic
welfare­­
particularly
the
measurement
basis
(
ordinary
consumer
surplus
and
the
more
precise
Hicksian­
based
measures).
The
comparisons
made
in
this
project
have
involved
expenditures
of
such
a
small
percentage
of
individuals'
budgets
that
the
differences
between
the
measures
is
1­
12
insignificant.
Since
some,
and
perhaps
many,
environmental
issues
may
involve
large
price
and
quantity
changes
with
more
significant
income
effects,
the
emPirical
application
of
various
measures
becomes
significant.
Bockstael
a
n
d
McConnell
[
19801
have
raised
some
empirically
b
a
s
e
d
i
s
s
u
e
s
,
b
u
t
a
m
o
r
e
extensive
effort
such
as
Willig's
[
19761,
comparing
recent
approaches
proposed
by
Hausman
[
1981],
McKenzie
and
Pearce
[
1982]
,
and
Takayama
[
1982],
may
yield
guidance
for
applications
with
these
large
changes.

A
final
general
issue
on
the
research
agenda
that,
unfortunately,
was
beyond
the
scope
of
this
project­­
and
too
many
other
benefits
analyses­­
is
the
distribution
aspect
of
benefit
policies.
By
neglecting
distribution
concerns,
economists­
are
unable
to
appreciate
many
policy
objections
expressed
in
the
political
arena.
For
example,
attention
to
the
distributional
effects
of
alternative
water
pollution
policies
would
be
a
valuable
complement
to
the
efficiencyoriented
questions
that
constitute
the
primary
focus
of
benefits
analysis.
Further
rationale
for
such
efforts
stems
from
Executive
Order
12291,
which
recognizes
the
importance
of
distribution
effects
by
requiring
them
in
regulatory
impacts
analyses.

The
future
research
agenda
for
the
individual
benefits
estimation
approaches
contains
items
ranging
in
subject
from
experimental
design
and
sampling
to
the
behavioral
models
that
underlie
several
approaches.
Some
of
the
agenda
items
are
already
being
studied
in
various
quarters,
while
others
will
involve
substantial
funding
­­
e.
g.
,
basic
data
collection­­
for
any
progress
to
be
made.

Specific
Research
Issues
The
travel
cost
model
developed
in
the
project
raises
as
many
research
questions
as
it
answers.
The
main
answer
is
that
the
model
can
be
used
to
estimate
the
benefits
of
water
quality
improvements
in
a
way
~
sistent
with
economic
theory.
*
However,
many
problems
were
encountered
on
the
way
to
answering
this
fundamental
question.
For
example,
in
the
survey
data
used
to
estimate
the
travel
cost
model,
as
in
many
surveys
involving
noneconomic
data,
the
data
were
heaped
at
specific
points,
possibly
presenting
problems
for
ordinary
least­
squares
regression
analysis.
Specifically,
all
visitors
who
made
only
one
visit
to
a
site
were
heaped
at
the
zero
point
for
the
logarithmic
transformation
of
the
dependent
variable
,
while
the
visitors
who
made
the
maximum
were
heaped
at
the
other
end
point.
The
maximum
is
the
value
(
8)
assigned
to
the
open
interval
for
five
or
more
visits.
The
remaining
visitors
were
arrayed
at
specific
intervals
in
between.
The
need,
obviously,
is
for
a
statistical
estimator
that
can
handle
this
problem.
I
n
terms
of
the
absolute
magnitude
of
the
estimated
values,
which
is
important
for
estimating
benefits,
the
differences
may
be
small,
but
this
is
a
fundamental
question
requiring
statistical
analysis
rather
than
judgment.
Equally
important,
the
fact
that
all
respondents
have
used
the
site
at
least
once
implies
that
this
study
fails
to
consider
the
demands
of
individuals
whose
maximum
willingness
to
pay
falls
below
their
travel
cost.
This
truncation
can,
as
suggested
in
the
report,
lead
to
biased
estimates
of
*
This
is
one
of
the
items
on
Freeman's
[
1979a]
research
agenda
cited
earlier.

1­
13
!

site
demands.
It
is
important
to
evaluate
the
implications
of
amending
the
statistical
models
to
directly
account
for
these
effects
for
the
benefit
estimates
derived
for
water
quality
improvements.

Many
of
the
items
on
the
travel
cost
research
agenda
stem
from
limited
data.
This
project
used
the
1977
Outdoor
Recreation
Survey's
Federal
Estate
component,
which
surveyed
visitors
at
various
recreation
sites
on
Federal
lands.
Although
in
many
ways
these
data
are
far
better
than
those
in
earlier
survey
efforts,
they
omit
many
items
important
for
the
travel
cost
model.
For
example,
there
were
no
questions
on
substitute
sites
that
respondents
had
considered­­
or
even
visited
at
other
times­­
before
visiting
a
particular
site.
While
the
generalized
model
assumes
that
site
attributes
are
capable
of
reflecting
substitution
potential,
the
model
would
be
considerably
improved
if
it
had
a
better
measure
of
substitutes.

The
travel
cost
model
also
assumed
that
the
sole
purpose
of
an
individual's
trip
was
to
visit
a
particular
site.
However,
Haspel
and
Johnson
[
1982]
point
out
the
potential
for
overstating
benefits
when
there
are
multiple
purposes
for
a
trip,
suggesting
the
need
for
more
research
using
itinerary
information
to
assess
the
importance
of
multipurpose
trips.
Also
needed
for
the
travel
cost
model
are
more
data
on
the
types
of
time
allocations
the
individual
considered
in
making
the
trip.
For
example,
was
work
time
forgone
or
compulsory
vacation
time?
Each
may
have
a
different
opportunity
cost.
With
answers
to
these
questions,
it
will
be
possible
to
improve
the
calculation
of
an
individual's
time
costs
for
recreation.

Including
site
attributes
in
the
travel
cost
model
created
several
datarelated
questions.
Specifically,
because
water
quality
data
from
the
standard
storage
system
(
STORET)
were
inadequate
for
many
recreation
sites,
observations
were
missing
on
key
parameters
,
and
the
monitoring
station
information
was
frequently
unreliable.
Clearly,
more
comprehensive
data
are
needed,
especially
for
water
quality
parameters
relevant
to
recreation
activities.
Data
on
other
site
attributes
such
as
access
or
size
were
available
for
the
U
.
S.
Army
Corps
of
Engineers'
sites
through
the
Corps'
Resource
Management
System.
However,
to
apply
the
model
to
other
recreation
sites­­
e.
g.
,
sites
managed
by
the
U.
S.
Forest
Service­­
would
require
similar
information
on
important
site
attributes.
Present!
y,
such
data
are
not
readily
available.

The
future
research
agenda
for
the
contingent
valuation
approach
is
aimed
at
a
more
systematic
treatment
of
issues
involving
the
design
of
the
hypothetical
market.
The
research
questions
are
in
the
general
area
that
economists
have
termed
"
framing
the
question"
(
see
Brookshire,
Cummings,
et
al.
[
1982]
)­­
an
area
generally
called
"
context"
in
the
psychological
literature
The
definition
of
the
commodity
to
be
valued,
the
question
format
used
to
elicit
the
value,
the
ordering
of
various
valuation
and
nonvaluation
questions
the
means
of
payment
in
the
market,
and
the
information
provided
in
the
survey
questionnaire
are
all
important
elements
in
this
framing
process.
More
attention
to
these
issues
is
likely
to
substantially
improve
the
understanding
of
the
approach
and
provide
results
that
are
easier
to
interpret.

1­
14
This
project
addressed
several
general
contingent
valuation
issues
by
comparing
several
question
formats
­­
bidding
games
with
two
starting
points,

direct
question,
and
the
unachored
payment
card
­­
both
to
each
other
and
to
results
from
the
contingent
ranking
format.
Different
payment
cards,
such
as
the
anchored
card
USed
bY
Mitchell
and
Carson
[
1981
1,
were
not
compared.
In
addition,
the
contingent
ranking
format
was
always
used
in
conjunction
with
another
question
format,
which
limits
the
independence
of
the
conclusions.
Both
of
these
are
good
candidates
for
future
research.

This
survey
was
conducted
in
a
specific
river
basin,
making
the
orientation
more
micro
in
scope
than
Mitchell
and
Carson's
[
1981].
A
more
systematic
comparison
of
their
results
for
overall
national
water
quality
and
the
results
of
this
study
for
the
Monongahela
River
basin
may
be
useful.
Moreover,
the
general
framin9
questions
are
especially
relevant
to
the
macro
approach,
where
it
is
more
difficult
to
define
the
hypothetical
commodity.
if
policy
decisions
require
basin­
specific
results
,
either
specific
surveys
(
or
the
ability
to
transfer
results
between
basins)
or
the
ability
to
infer
estimates
for
specific
river
basins
from
the
macro
approach
will
be
required.

Recently,
Brookshire,
Cummings,
et
al.
[
1982]
introduced
the
ideas
of
environmental
accounts
and
budget
constraints
as
part
of
the
framing
issue.
The
accounts
question
aims
at
determining
whether
people
give
an
overall
environmental
quality
bid
in
a
survey
or
a
bid
for
the
specific
hypothetical
commodity.
The
budget
constraint
requires
that
individuals
provide
rough
budget
shares
for
their
monthly
incomes
and
then
reallocate
these
categories
to
provide
the
budget
amount
for
the
hypothetical
commodity.
The
preliminary
results
in
Brookshire,
Cummings,
et
al.
[
1982]
suggest
this
is
a
useful
avenue
for
learning
more
about
framing
influences.

Finally,
improving
efficiency
in
defining
hypothetical
markets
is
a
neglected
area
in
the
contingent
valuation
literature.
One
promising
approach
is
the
use
of
focus
groups
(
from
market
research
literature)
to
obtain
impressions
about
terminology,
visual
aids,
and
other
framing
issues.
Applying
these
marketing
research
ideas
to
contingent
valuation
may
indicate
their
overall
merits.

Research
agendas
must
continually
evolve,
producing
new
avenues
from
deadends
that
once
offered
promise.
The
present
agenda
tries
to
map
some
of
these
new
avenues.
The
passage
of
time
and
the
fruits
of
future
research
will
mark
its
ultimate
usefulness.

1.4
GUIDE
TO
THE
REPORT
This
chapter
has
introduced
the
report
by
highlighting
the
project
object
i
v
e
s
a
n
d
s
u
m
m
a
r
i
z
i
n
g
t
h
e
f
i
n
d
i
n
g
s
o
f
t
h
e
r
e
s
e
a
r
c
h
.
Chapter
2
provides
a
brief
review
of
some
of
the
theoretical
issues
of
comparing
alternative
benefit
estimation
approaches.
After
describing
the
Monongahela
River
basin,
Chapter
3
summarizes
the
sampling
and
survey
plans
for
the
contingent
valuation
and
contingent
ranking
approaches
used
in
the
case
study.
Chapter
4
builds
on
the
contingent
valuation
foundation
laid
in
Chapter
3
by
presenting
the
research
design
for
the
contingent
valuation
survey,
by
profiling
key
groups
of
respondents
,
and
by
summarizing
the
empirical
option
price
results,
includ
­

1­
15
ing
the
effects
of
question
format,
starting
point,
and
interviewer
bias.
Chapter
5
synthesizes
the
theoretical
underpinnings
of
option
value,
giving
particular
attention
to
the
role
of
supply
uncertainty,
and
presents
empirical
results
for
both
option
and
existence
values.
Chapter
6
reviews
the
theory
underlying
the
contingent
ranking
approach,
provides
a
critical
summary
of
its
previous
applications,
considers
appropriate
measures
of
benefits,
and
summarizes
the
empirical
findings
from
its
application
to
the
Monongahela
River
basin.
Chapter
7
presents
the
development
of
a
generalized
travel
cost
model
and
describes
its
application
to
predict
the
recreation
benefits
of
water
quality
improvements
in
the
Monongahela
River.
The
development
of
the
model
treats
the
empirical
significance
of
model
specification,
site
time
costs,
simultaneity
in
visit/
site
time
decisions,
and
statistical
biases
in
its
predicted
values.
Chapter
8
compares
the
alternative
approaches
for
estimating
recreation
and
related
benefits,
in
light
not
only
of
previous
comparison
attempts
but
also
of
g
priori
expectations.
In
addition,
Chapter
8
also
describes
paired
comparisons
of
the
contingent
valuation
and
travel
cost
approaches
and
of
the
contin
­
gent
valuation
and
contingent
ranking
approaches
using
multivariate
regression
techniques.

1­
16
CHAPTER
2
A
BRIEF
REVIEW
OF
THE
CONCEPTUAL
BASIS
FOR
THE
BENEFIT
ESTIMATION
APPROACHES
2.1
INTRODUCTION
An
ideal
comparison
of
benefit
estimation
approaches
would
begin
with
a
detailed
theoretical
appraisal
of
each
approach,
showing
how
each
is
derived
from
a
common
conceptual
framework.
However,
this
kind
of
appraisal
is
beyond
the
scope
of
this
project.
Instead,
this
chapter
highlights
the
assumptions
information,
and
types
of
benefits
measured
by
each
of
three
approaches
and
compares
these
features
on
general,
rather
than
on
common,
theoretical
grounds.

The
definition
of
economic
benefits
based
on
theoretical
welfare
economics
has
closely
followed
the
model
of
consumer
behavior,
which
suggests
that
individuals
can
acquire
utility
only
through
consuming
goods
or
services.
This
framework
leads
to
definitions
of
economic
benefits
best
suited
for
describing
user
benefits
associated
with
improvements
in
environmental
quality.
However,
since
the
work
of
Krutilla
[
1967],
nonuser,
or
intrinsic,
benefits
have
been
increasingly
recognized
as
playing
an
important
role
in
the
aggregate
values
for
certain
environmental
resources.

Intrinsic
benefits
are
generally
viewed
as
arising
from
two
sources.
The
first
source
suggests
that
an
individual
can
realize
utility
without
direct
consumption
of
a
good
or
service.
Rather,
other
motives
can
be
satisfied
with
allocation
patterns
for
certain
resources,
and
these
motives
­­"
stewardship"
and
"
vicarious
consumption"
in
Freeman's
[
1981
]
terms­­
can
lead
to
utility,
therefore
providing
nonuser
benefits.
An
alternate
view
can
be
derived
by
redefining
the
act
of
consumption
to
admit
what
might
be
considered
indirect
use
of
the
services
of
an
environmental
amenity.

The
second
source
of
intrinsic
benefits
is
derived
by
relaxing
one­
of
the
assumptions
underlying
conventional
consumer
behavior
models.
The
simplest
treatment
of
the
conditions
for
efficient
resource
allocation
assumes
that
all
goods
and
services­­
whether
they
provide
positive
increments
to
utility
or
decrease
it­­
are
available
with
certainty.
Of
course,
this
is
not
the
case
in
the
real
world.
Indeed,
in
some
circumstances­­
e.
g.
,
the
degree
of
reversibility
in
water
quality
conditions
­­
uncertainty
may
well
be
the
most
important
element
of
the
public
policy
problem.
I
n
these
cases,
therefore,
consumer
behavior
models
must
be
amended
to
reflect
how
households
react
to
uncertainty
and
whether
they
would
be
willing
to
pay
for
action
that
would
reduce
i
t
.
e
2­
1
A
second
relevant
feature
of
the
definitions
of
economic
benefits
presumably
arises
from
the
early
focus
on
goods
or
services
exchanged
in
private
markets.
These
definitions
developed
measures
of
benefits
for
price
changes.
Since
environmental
policy
has
dealt
mostly
with
quantity
(
or
quality)
changes
in
services
provided
outside
of
private
markets,
these
measures
must
be
adapted
to
meet
policy
needs.

The
purpose
of
this
chapter
is
to
briefly
review
the
theoretical
concepts
generally
used
in
measuring
economic
benefits,
Specifically,
Section
2.2
deals
with
the
theoretical
basis
of
benefit
measurement
based
on
the
concept
of
an
individual's
willingness
to
pay
and
describes
alternative
ways
to
measure
changes
in
consumer
welfare.
Section
2.3
outlines
the
framework
for
comparing
different
benefit
estimation
approaches
­­
an
adaptation
of
the
Smith
­
Krutilla
[
1982]
framework
for
classifying
the
different
approaches
and
summarizing
their
conceptual
bases.
Section
2.4
describes
the
welfare
measurement
bases
underlying
the
two
benefit
estimation
approaches
compared
in
this
study
­­
travel
cost
and
contingent
valuation
(
including
the
contingent
ranking
format).
Finally,
Section
2.5
prvides
a
brief
summary.

2.2
A
BRIEF
REVIEW
OF
THE
CONVENTIONAL
THEORY
OF
BENEFITS
MEASUREMENT
The
primary
emphasis
for
this
study
of
recreation
and
related
benefits
of
water
quality
improvements
focuses
on
the
measurement
of
benefits
that
accrue
to
individual
households.
Fortunately,
the
theory
of
consumer
behavior
provides
a
framework
for
measuring
these
benefits.
This
section
briefly
reviews
this
framework
to
set
the
stage
for
the
comparison
of
approaches
that
follows.

The
first
guidepost
for
the
definition
and
measurement
of
economic
benefits
that
the
theory
of
consumer
behavior
provides
is
the
individual
demand
function,
shown
in
Figure
2­
1.
This
function
describes
for
any
good,
X,
the
maximum
amount
an
individual
would
be
willing
to
pay
for
each
quantity
of
X.
The
downward
slope
of
the
curve
indicates
that
individuals
are
willing
to
buy
more
of
X
at
lower
prices
than
they
are
at
higher
prices.
The
simple
twodimensional
diagram
in
Figure
2­
1
assumes
all
other
factors
that
might
influence
demand
­­
income,
the
prices
of
related
goods,
etc.
­­
do
not
change.
Thus,
according
to
the
demand
function,
if
the
market
leads
to
a
price
of
P
o,
the
individual
will
purchase
Q.
of
X
and
make
a
total
expenditure
equal
to
POAQOO.
Since
the
demand
curve
measures
the
individual's
maximum
willingness
to
pay
for
each
level
of
consumption,
the
total
willingness
to
pay
for
Q.
can
be
derived­­
total
expenditures
~
the
triangle
PoPiA.
This
difference
between
the
amount
they
are
willing
to
pay
and
what
individuals
actually
pay
with
a
constant
price
per
unit
is
defined
as
the
consumer
surplus­­
the
conventional
dollar
measure
of
the
satisfaction
individuals
derive
from
consuming
a
good
or
service,
exclusive
of
what
they
pay
for
it.

As
a
dol~
r
measure
of
individual
welfare,
however,
consumer
surplus
is
not
ideal.
The
most
direct
way
of
understanding
its
limitations
is
to
consider
2
­
2
Price
($/
unit)
t
Quantity/
time
Figure
2­
1.
The
demand
function
and
the
consumer
surplus
welfare
measure,

the
measurements
underlying
a
conventional
demand
function.
An
individual's
demand
function
describes
the
maximum
an
individual
with
a
given
nominal
income
would
be
willing
to
pay
for
each
level
of
consumption
of
a
particular
good
.
Specifically,
if
the
price
paid
changes,
it
will
affect
not
only
what
the
individual
can
purchase
of
this
good,
but
also
the
purchases
of
all
other
commodities
through
its
effect
on
the
remaining
disposable
income.
Thus,
movement
along
a
conventional
demand
function
affects
the
level
of
satisfaction
an
individual
will
be
able
to
achieve
with
a
given
income.
For
example,
suppose
the
price
of
hypothetical
good
X
declines
to
P
1
.
The
individual
can
purchase
the
same
quantity
of
X
at
its
new
price
as
indicated
in
Figure
2­
1
by
the
area
OPIBQO
and
have
income
remaining,
as
given
by
PIPOAB,
to
purchase
more
X
or
more
of
other
goods
and
services.
The
movement
to
a
consumption
level
of
OQ1
describes
the
increased
selection
of
X
under
the
new
price.
This
change
leads
to
a
higher
utility
level
because
more
goods
and
services
can
be
consumed
with
the
same
income.
For
consumer
surplus
to
provide
an
"
ideal"
dollar
measure
of
individual
well­
being,
however,
the
conversion
between
dollars
and
individual
utility
levels
must
be
constant
for
every
point
on
the
demand
curve.
According
to
this
example,
then
,
each
point
on
a
conventional
demand
function
in
principle
corresponds
to
a
different
level
of
utility.
Thus,
no
single
conversion
factor
links
consumer
surplus
and
utility.

In
his
seminal
work
on
consumer
demand
theory,
Hicks
[
1943]
noted
that
an
ideal
measure
would
require
that
utility
be
held
constant
at
all
points
along
the
demand
curve.
As
a
practical
matter,
however,
the
difference
between
the
area
under
such
an
ideal,
Hicksian­
based
demand
curve
and
that
under
a
conventional
demand
curve
depends
on
the
size
of
the
income
effects
accompanying
the
price
changes
associated
with
movements
along
the
ordinary
demand
curve.
As
suggested
earlier,
price
reductions
lead
to
more
disposable
income.
To
judge
the
association
between
the
two
measures
of
welfare
change,
all
=
pects
of
the
choice
process
that
affect
the
size
of
the
change
in
disposable
income
must
be
considered.
For
example,
if
the
price
change
for
X
is
small
and
the
share
of
the
budget
spent
on
the
good
X
is
also
small,
the
2
­
3
r.,
.~!

change
in
disposable
income
is
likely
to
be
small.
Thus,
little
difference
will
exist
between
the
ordinary
measure
of
consumer
surplus
and
the
measure
derived
from
Hicks'
idealized
demand
curve.
However,
the
same
outcome
arises
either
if
income
has
littIe
effect
on
the
demand
for
X
or
if
an
individual's
preferences
are
such
that
the
demand
responsiveness
to
income
is
equal
for
all
goods
(
i.
e.,
unitary
income
elasticities
of
demand).

Of
course,
each
of
the
conditions
described
above
is
a
special
case.
When
ordinary
demand
functions
are
used
to
measure
the
benefits
of
an
action
in
practical
applications,
the
factors
influencing
the
demand
function's
relationship
to
an
ideal
dollar
measure
of
welfare
change
must
be
identified.
.,
Fortu
­
nately,
Will­
ig
[
1976]
and
Randall
and
Stoll
[
1980]
have
derived
such
guidelines
for
cases
involving
price
and
quantity
changes,
respectively.
To
understand
these
guidelines,
the
possible
theoretical
measures
of
individual
welfare
change
must
first
be
defined
in
more
precise
terms.

Hicks'
[
1943]
theoretical
analysis
of
measures
of
welfare
change
provides
the
basis
for
developing
a
set
of
rigorous
measures
and,
with
them,
the
error
bounds
for
ordinary
consumer
surplus.
The
four
Hicksian
welfare
measures
for
a
price
decrease
are
summarized
below:

.
Compensating
variation
(
CV)
is
the
amount
of
compensation
that
must
be
taken
from
an
individual
to
leave
him
at
the
same
level
of
satisfaction
as
before
the
change.

l
Equivalent
variation
(
EV)
is
the
amount
of
compensation
that
must
be
given
to
an
individual,
in
the
absence
of
the
change,
to
enable
him
to
realize
the
same
level
of
satisfaction
he
would
have
with
the
price
change.

.
Compensating
Q
surplus
(
CS)
is
the
amount
of
compensation
that
must
be
taken
from
an
individual,
leaving
him
just
as
well
off
as
before
the
change
if
he
were
constrain­
cd
to
buy
at
the
new
price
the
quantity
of
the
commodity
he
would
buy
in
the
absence
of
compensation.

l
Equivalent
surplus
(
ES)
is
the
amount
of
compensation
that
must
be
given
to
an
individual
,
in
the
absence
of
the
change,
to
make
him
as
well
off
as
he
would
be
with
the
change
if
he
were
constrained
to
buy
at
the
old
price
the
quantity
of
the
commodity
he
would
actually
buy
with
the
new
price
in
the
absence
of
compensation

As
a
simplified
comparison,
Figure
2­
2
highlights
the
essential
differences
between
the
Hicksian
variation
measures
and
the
ordinary
consumer
surplus
measures
when
the
price
of
a
good
decreases.
The
two
Hicksian
demand
c
u
r
v
e
s
hold­
ing
utility
constant
(
at
levels
U.
and
U
1
with
U
1
>
Uo)
are
shown
a
s
H(
uo)
and
H(
U
1),
the
prechange
and
postchange
Ieveis
of
utilitY/
respect
i
v
e
l
y
.
The
ordinary
demand
curve­­
also
known
as
the
Marshal
lian
demand
curve
­­
is
shown
as
D,
where
income,
and
not
utility,
is
held
constant.
Ths
2
­
4
Qumltity
H(
U1)
NOTE:
D
­
ordinary
dommd
curve
H(
IJO),
H(
U1)
­
Hidcsian
dmand
ww.
s
Figure
2­
2.
A
comparison
of
alternative
welfare
measures.

compensating
variation
measure,
labeled
as
area
a,
uses
the
original
level
of
utility
as
its
reference
point
and
indicates
the
amount
of
compensation
that
must
be
taken
from
an
individual
to
leave
him
at
the
original
level
of
utility
when
the
price
changes
from
P.
to
PI.
The
equivalent
variation
measure
is
represented
by
area
a+
b+
c.
It
measures
the
change
in
income
equivalent
to
the
change
in
prices
and
thereby
permits
an
individual
to
realize
the
new
level
of
utility
with
old
price
P
o
.
The
change
in
ordinary
consumer
surplus
is
the
area
under
the
ordinary
demand
curve,
D,
between
P.
and
PI.
In
Figure
2­
2
it
is
shown
as
areas
a+
b.

The
concepts
of
compensating
surplus
and
equivalent
surplus
were
originally
defined
as
measures
of
the
welfare
change
resulting
from
a
price
change,
given
that
the
quantity
of
the
good
whose
price
has
changed
is
not
allowed
to
adjust.
However,
it
is
also
possible
to
interpret
these
concepts
as
measures
of
the
welfare
change
associated
with
a
quantity
change
(
see
Randall
and
Stoll
[
1980]
).
Just,
Hueth,
and
Schmitz
[
1982]
have
recently
offered
a
diagrammatic
illustration
of
compensating
and
equivalent
surplus
in
a
format
similar
to
that
used
above
to
describe
compensating
and
equivalent
variation.
However,
in
the
present
example,
the
price
is
assumed
constant
at
some
arbitrarily
low
value
(
effectively
zero
for
Figure
2­
3),
and
D
is
interpreted
as
an
ordinary
demand
curve
(
i.
e.,
as
if
the
quantities
consumed
could
be
realized
only
at
the
corresponding
prices
and
not
the
constant
price).
In
Figure
2­
3
a
change
in
the
quantity
of
the
good
available
from
Q.
to
Q
1
leads
to
a
compensating
surplus
of
c+
f
and
an
equivalent
surplus
of
a+
e+
c+
d+
f+
g.
The
ordinary
consumer
surplus­
is
c+
d+
f+
g,
which
is
d+
g
more
than
the
compensating
surplus
measure
and
a+
e
less
than
the
equivalent
surplus.

2
­
5
Priti
PO
b
c
P,

f
\
H(
lfo)

`%
al
*
Quantity
.
Sowco:
Just,
Hucth,
and
%
hmitz
[
1982].

Nota:
Ordinary
consumer
turplus
­
c
+
d
+
f
+
g
Compmmting
wrplus
=
c
+
f
Equivalent
surplus
=
a
+
e
+
d
+
c
+
f
+
g
Figure
2­
3,
Surplus
measures
for
a
change
in
quantity.

Table
2­
1
relates
the
welfare
measures
under
different
conditions
of
willingness
to
pay/
accept,
showing
quite
clearly
that
no
one
unique
measure
exists.
Rather,
the
appropriate
measure
is
determined
by
the
particular
situation
Table
2­
1
reinforces
this
point
by
presenting
the
types
of
welfare
measure
in
relation
to
different
situations.
For
a
price
decrease,
for
example,
the
following
relationship
holds
between
the
alternative
welfare
measures:

ES>
EV>
CV>
CS.

For
a
quantity
increase,
the
equivalent
surplus
measure
will
be
greater
than
the
compensating
surplus
measure.
The
primary
reason
for
the
differences
between
welfare
measures
is
that
the
equivalent
surplus
and
equivalent
variation
are
not
bounded
by'
an
individual's
income
constraint,
while
the
compensating
variation
and
compensating
surplus
measures
are.
It
should
also
be
noted
that
the
measures
are
symmetrical
and
that,
for
a
price
increase
or
quantity
decrease
the
relationship
between
the
measures
is
exactly
the
reverse.

It
is
important
to
recognize
that
the
compensating
and
equivalent
measures
of
welfare
changes
differ
because
they
imply
a
different
assignment
of
property
rights
to
the
individual
and
therefore
are
based
on
different
corresponding
f
rame~
of
reference.
For
example,
with
a
price
decrease,
the
compensating
variation
measure
takes
the
initial
price
set
as
an
individual's
frame
of
reference
and
asks,
in
effect,
"
What
is
the
maximum
amount
he
would
be
willing
to
pay
to
have
access
to
the
lower
prices?"
By
contrast,
equivalent
variation
takes
the
new,
lower
price
set
as
an
individual's
frame
of
reference
and
2
­
6
Table
2­
1.
Alternative
Welfare
Measures
and
Types
of
Consumer
Surplus
Measures
for
Contingent
Valuation
Studies
Price
Price
Quantity
Quantity
decrease
increase
increase
decrease
WT
P
C
v
;
Cs
EV;
ES
Cs
ES
WTA
EV;
ES
Cv;
Cs
ES
Cs
NOTE:

Cs
is
the
amount
of
compensation
that
must
be
taken
from
an
individual,
leaving
him
just
as
well
off
as
before
the
change
if
he
were
constrained
to
buy
at
the
new
price
the
quantity
of
the
commodity
he
would
buy
in
the
absence
of
compensation.

Cv
is
the
amount
of
compensation
that
must
be
taken
from
an
individual
to
leave
him
at
the
same
level
of
satisfaction
as
before
the
change.

ES
is
the
amount
of
compensation
that
must
be
given
to
an
individual,
in
the
absence
of
the
change,
to
make
him
as
well
off
as
he
would
be
with
the
change
if
he
were
constrained
to
buy
at
the
old
price
the
quantity
of
the
commodity
he
would
buy
in
the
absence
of
compensation.

EV
is
the
amount
of
compensation
that
must
be
given
to
an
individual,
in
the
absence
of
the
change,
to
enable
him
to
realize
the
same
level
of
satisfaction
he
would
have
with
the
price
change.

WTA
is
the
amount
of
money
that
would
have
to
be
paid
to
an
individual
to
forego
the
change
and
leave
him
as
well
off
as
if
the
change
occurred.

WTP
is
the
amount
of
money
an
individual
will
pay
to
obtain
the
change
and
still
be
as
well
off
as
before.

describes
the
minimum
amount
an
individual
would
be
willing
to
accept
to
relinquish
his
right
to
the
lower
price.
These
measures
bound
the
range
of
dollar
values
for
the
welfare
changes
because
they
describe
the
results
obtained
from
the
perspectives
of
the
initial
utility
level
and
the
final
utility
level.
Consequently
Willig
[
1976]
uses
this
feature
to
establish
conditions
under
which
conventional
consumer
surplus
would
approximate
"
ideal"
measures
for
the
welfare
change
associated
with
a
price
change.
Moreover,
Randall
and
Stoll
[
1980]
follow
essentially
the
same
logic
to
gauge
the
relationship
between
ordinary
consumer
surplus
measures
for
a
quantity
change
and
the
corresponding
compensating
and
equivalent
surplus
measures.

Equations
(
2.1
)
and
(
2.2)
provide
the
basis
for
the
Willig
bounds
for
the
difference
between
the
ordinary
consumer
surplus
measure
and
the
equivalent
2­
7
variation
and
compensating
variation
measures
of
a
change
in
welfare
due
to
a
price
change:
*

C
v
­
OCS
~
IOCSI
N
IOCSI
­
2
M
`

OCS
­
EV
~
!
OCSI
N
,
IOCSI
­
2
M
(
2
.
1
)

(
2
.
2
)

where
O
c
s
=
ordinary
consumer
surplus
measure
of
welfare
change
N
=
income
elasticity
of
demand
=
4/
y
Q
M
=
initial
level
of
income.

These
relationships
can
be
evaluated
at
different
values
for
ticity
of
demand
over
the
region
for
the
price
change
a
n
d
the
income
elasthereby
~
rovide
bounds
for
the
magnitude
of
~
he
discrepanc­
y
between
`
ordinary
consumer
surplus
and
the
equivalent
and
compensating
variation
welfare
measures.
Equations
(
2.1)
and
(
2.2)
assume
that
the
income
elasticity
of
demand
(
N)
is
approximately
constant
over
the
region
for
the
price
change
(
see
Willig
[
1976],
pp.
592­
593,
for
a
discussion).
If
this
assumption
is
relaxed,
the
bounds
can
be
stated
as
inequalities
for
the
percentage
difference
between
ordinary
consumer
surplus
and
the
corresponding
measures
of
welfare,
as
in
Equations
(
2.3)
and
(
2.4):

IOCSI
`
s
~
cv
­
OCS
~
Iocst
`
L
,

2M
­
IOCSI
­
2
M
(
2
.
3
)

(
2
.
4
)

where
`
s
=
the
smallest
value
of
the
income
elasticity
of
demand
over
the
region
for
the
price
change
*
lt
is
important
to
note
that
the
direction
of
the
price
change
affects
the
sign
of
ordinary
consumer
surplus,
compensating
variation,
and
equivalent
variation
and,
thus,
the
interpretation
of
Equations
(
2.1)
through
(
2.4).
This
formulation
"
adopts
Willig's
convention
that
ordinary
consumer
surplus
is
posi
­
tive
for
a
price
increase
and
negative
for
a
price
decrease
so
that
it
corresponds
to
the
interpretation
of
compensating
variation
or
equivalent
variation.
~

See
Willig
[
1976],
p.
589.
`
L
=
the
largest
value
of
the
income
elasticity
of
demand
over
the
region
for
the
price
change.

The
Willig
approximation
is
reasonable
if
the
value
of
­$%&.
S
~.~~
If
t
h
i
s
.

value
is
greater
than
0.05,
Willig
has
provided
a
table
of
error
bounds
based
on
the
relationships
used
to
derive
these
approximate
bounds.
*

2.3
A
FRAMEWORK
FOR
COMpARING
ALTERNATIVE
BENEFIT
MEASUREMENT
APPROACHES
Comparing
alternative
approaches
for
estimating
the
recreation
and
related
benefits
of
water
quality
improvements
at
first
seems
formidable
because
of
the
wide
range
of
consumer
behavior
outcomes
described
by
each.
However,
despite
this
diversity,
all
approaches
adhere
to
a
consistent
general
model
of
consumer
behavior:
individuals
allocate
their
monetary
and
time
resources
to
maximize
their
utility
subject
to
budget
and
time
constraints.
As
noted
at
the
outset,
a
complete
comparison
of
the
methods
could
derive
each
method
from
this
common
conceptual
basis.
However,
this
section
simply
provides
a
taxonomic
framework
that
eases
the
comparison
of
approaches
by
drawing
clear
distinctions
between
the
assumptions
underlying
each.

Figure
2­
4
presents
the
Smith­
Krutilla
[
1982]
framework
for
classifying
the
alternative
approaches
for
measuring
the
recreation
and
related
benefits
of
water
quality
improvements.
This
framework
considers
I
inkages
between
changes
in
water
quality
and
observable
actions
of
economic
agents
that
affect
the
information
available
for
measuring
water
quality
benefits.
In
particular,
Smith
and
Krutilla
suggest
that
all
approaches
for
measuring
the
benefits
of
a
change
in
an
environmental
resource
can
be
classified
as
involving
either
physical
or
behavioral
assumptions.

The
category
associated
with
physical
assumptions
in
this
framework
maintains
that
the
association
between
the
environmental
service
of
interest
(
i.
e.
,
water
quality)
and
the
observable
activities
(
or
changes
in
goods
or
services)
is
a
purely
physical
relationship.
The
responses
are
determined
by
either
engineering
or
technological
relationships.
Thus,
the
evaluation
of
water
quality
changes
in
such
a
framework
must
begin
by
identifying
the
activities
affected
by
water
quality.
Analysis
must
then
focus
on
measuring
the
technical
relationships,
sometimes
referred
to
as
damage
functions,
assumed
to
exist
between
water
quality
and
each
activity.
Because
water
quality
improvements
can
be
associated
with
the
support
of
gamefish,
swimming,
and
the
use
of
water
for
human
consumption,
the
physical
approach
seeks
to
specify
the
technical
linkages
between
water
quality
levels
and
permitted
amounts
of
recreation
fishing,
swimming,
and
water
consumption.
Another
example
of
the
*
Two
excellent
discussions
of
the
practical
implications
of
the
Willig
bounds
for
benefit
measurement
are
available
in
Freeman
[
1979a],
pp.
47­
50,
and
in
Just,
Hueth,
and
Schmitz
[
1982],
pp.
97­
103.

2
­
9
No
Role
for
Behavioral
Responses
of
Economic
Agents
Behavioral
Responses
of
Economic
Agents
Are
Essential
Types
of
Linkega
Between
Water
Ouality
Change
and
Observed
Effects
Typas
of
Assumptions
Measurement
Raqui­
Approachaa
I
Responses
are
Physical
determined
by
I
I
Linkages
I
engineering
or
I
Damage
Function
"
technological"
I
Behavioral
Linkages
Indirect
Links
Restrictions
on
the
nature
of
individual
preferences
OR
obaawad
technical
associations
in
the
dolivary
of
goods
or
services
Hedonic
Proparty
Value
Travel
Cost"

Figure
2­
4.
Smith­
Krutilla
framework
for
classifying
the
measurement
bases
and
approaches
of
economic
benefits
resulting
from
improved
water
quality.

physical
approach
to
evaluating
the
effects
(
and,
ultimately,
the
benefits)
of
a
water
quality
change
can
be
found
in
the
dose­
response
models
used
to
evaluate
the
health
risks
associated
with
certain
forms
of
water
pollution
(
see
Page,
Harris,
a
n
d
Bruser
[
1981]
for
a
review
of
these
models).
Although
these
models
ignore
economic
behavior
and
postulate
that
the
relationships
involved
can
be
treated
independently
of
the
motivations
of
economic
agents,
they
may
well
provide
reasonable
approximations
of
the
actual
effects
on
water
quality
for
certain
classes
of
impacts.
However,
these
models
are
unlikely
to
be
adequate
when
economic
agents
can
adjust
their
behavior
in
response
to
the
water
quality
changes
and,
as
a
result
,
are
not
considered
in
this
study.

The
behavioral
category
of
valuation
methodologies
in
the
Smith­
Krutilla
framework
relies
on
the
observed
responses
of
economic
agents
and
on
a
model
describing
their
motivations
to
estimate
the
values
(
or
economic
benefits)
associated
with
a
change
in
a
nonmarketed
good
or
service.
Within
this
class,
direct
or
indirect
links
identify
three
classes
of
assumptions
that
can
be
used
to
develop
measures
of
individual
willingness
to
pay.
T
h
e
f
i
r
s
t
t
y
p
e
o
f
assumption
used
­
within
the
indirect
behavioral
framework
requires
restrictions
on
the
nature­
of
the
individual's
utility
function
and
is
usually
associated
with
Maler's
[
1974]
weak
complementarily.
This
type
of
assumption
maintains
that
an
individual's
utility
function
is
such
that
there
is
a
specific­
a
s
s
o
c
i
a
t
i
o
n
between
the
nonmarketed
good
(
or
service)
and
some
marketed
commodity
such
2­
lC
that
the
mar9inal
utilitY
of
an
increment
to
the
consumption
of
the
nonmarketed
good
is
zero
when
the
individual
is
not
consuming
some
positive
amount
of
the
associated,
marketed
commodity.
This
assumption
maintains
that
a
type
of
lljOintrleSS1l
exists
in
the
formation
of
the
individual's
utility,
which,
in
turn,
constrains
the
feas~
ble
resPonses
to
Price
changes
for
the
marketed
good
(
or
changes
In
the
.
av.
allability
of
the
nonmarketed
good).
Thus,
the
selection
of
the
two
9oods
IS
loint~
and
market
transactions
for
one
good
can
be
used
to
determine
demand
for
the
other.
Of
course,
this
approach
depends
upon
the
plausibility
of
the
restriction
on
an
individual's
utility
function.
Researchers
have
used
this
restriction
to
justify
both
hedonic
property
value
and
travel
cost
studies.

Smith
and
Krutilla
[
1982]
argue
that
the
weak
complementarily
behavioral
restriction
is
.~
necessary
for
these
approaches
and
that
the
observed
technical
associations
between
marketed
and
nonmarketed
goods
are
responsible
for
the
abi
Iity
to
use
these
methods
to
measure
benefits
of
changes
in
a
nonmarketed
good.
In
the
case
of
the
technical
assumptions,
the
availability
of
the
nonmarketed
service
is
tied
to
some
marketed
good
by
the
nature
of
its
natural
delivery
system,
making
the
linkage
an
observable
phenomenon
rather
than
a
feature
of
an
individual's
preferences.
For
example,
water­
based
outdoor
recreation
is
undertaken
using
the
services
of
recreation
sites
on
rivers
or
lakes.
Each
recreationist
is
interested
in
the
water
qualities
only
at
the
sites
considered
for
his
recreation
use.
By
selecting
a
site
for
these
activities,
an
individual
is
also
selecting
a
water
quality,
because
the
two
are
"
technically
linked"
or
jointly
supplied.
Thus,
where
there
is
a
range
of
choice
(
i.
e.,
several
different
combinations
of
recreation
sites
and
water
quality),
how
an
individual
values
the
nonmarketed
good
or
service
can
be
seen
through
his
observable
actions,
including
such
decisions
as
the
selection
of
a
residential
location
or
visits
to
specific
recreation
facilities
(
see
Rosen
[
1974]
and
Freeman
[
1979c]
).
This
study
specifically
considers
the
travel
cost
method,
which
uses
this
technical
association
as
its
basis
for
measuring
water
quality
benefits.

The
last
case
of
behavioral
approaches
to
benefit
estimation
involves
direct
linkages
between
water
quality
and
an
individual's
actions.
The
assumptions
made
to
ensure
these
linkages
are
labeled
institutional,
a
designation
somewhat
more
difficult
to
understand
than
previous
descriptions
because
it
encompasses
the
contingent
valuation
and
contingent
ranking
methods
for
measuring
an
individual's
valuation
of
environmental
amenities.
Specifically,
the
institutional
assumptions
arise
because
the
analyst
assumes
that
individual
responses
to
hypothetical
decisions
(
or
transactions)
are
completely
comparable
to
individual
responses
revealed
in
actual
transactions.
The
term
institutional
is
used
for
this
class
because
the
organized
markets
in
which
goods
and
services
are
exchanged
are
institutions
that
provide
the
information
on
individuals'
marginal
valuations
of
the
commodity
involved.
With
the
survey
approach,
the
interviewer
poses
the
survey
questions
to
construct
an
equivalent
institutional
mechanism
in
the
form
of
a
hypothetical
market.
Both
the
contingent
valuation
and­
the
contingent
ranking
methods
will
be
considered
under
this
approach.

2­
11
2.4
THE
NATURE
OF
THE
BENEFITS
MEASURED
IN
THE
ALTERNATIVE
APPROACHES
This
section
highlights
the
nature
of
the
benefits
measured
in
the
travel
cost
and
contingent
valuation
approaches.

2.4.1
Travel
Cost
Appreach
The
travel
cost
approach
measures
the
change
in
ordinary
consumer
surplus
for
a
water
quality
improvement,
represented
for
an
individual
incurring
travel
costs
per
trip
of
OPI
by
area
ABCD
in
Figure
2­
5.
To
empirically
develop
the
ordinary
consumer
surplus
estimate,
the
travel
cost
approach
assumes
both
that
travel
to
a
recreation
site
reveals
a
respondent's
reservation
price
for
that
site's
services
and
that
water
quality
is
jointly
supplied
along
with
the
other
site
attributes.
If
other
variables
are
held
constant,
and
if
sites
are
placed
on
a
common
measurement
scale,
*
area
ABCD
can
be
measured
by
observing
individuals'
site
selections
across
sites
with
varying
levels
of
water
quality,
thus
revealing
the
effect
of
water
quality
on
site
demand.
Therefore,
while
both
Freeman
[
1979b]
and
Feenberg
and
Mills
[
1980]
maintain
that
conventional
travel
cost
models
cannot
measure
benefits
associated
with
water
quality
change,
?
the
generalized
travel
cost
model
developed
for
this
Travel
costs
($/
ox)

I
PI
o
B
c
Qx/
t
(
visits/
year)

Figure
2­
5.
Travel
cost
demand
function
with
water
quality
improvement.

x~
he
rat~
onale
in
Section
7.3.

~
Their
models
quality
change.
for
this
measurement
approach
is
presented
in
more
detail
do
not
gauge
the
demand
change
that
accompanies
a
water
2­
12
study
(
see
Cha
Pter
7)
uses
the
resPonses
of
individuals
at
different
locations
to
both
travel
cost
and
water
quality
levels
to
infer
benefits
of
water
quality
changes.
The
information
provided
by
these
responses
allows
the
change
from
D(
WQI)
to
D(
WQ2)
in
Figure
2­
5
to
be
distinguished
(
where
WQI
a
n
d
W
Q
2
represent
different
levels
of
water
quality,
with
WQ2
>
WQI).

2.4.2
Contingent
Valuation
Ap
preach
The
contingent
valuation
approach
directly
measures
an
individual's
willingness
to
pay
for
water
quality
in
an
institutional
arrangement
that
approximates
the
market
for
water
quality.
Unlike
the
travel
cost
approach,
contingent
valuation
does
not
require
observations
of
individuals'
decisions
on
use
of
recreation
sites
with
given
"
implicit"
service
prices,
but
it
does
assume
an
individual's
response
in
the
hypothetical
market
is
the
same
as
it
would
be
in
a
real
market.
That
is,
respondents
are
assumed
not
to
behave
strategically,
not
to
give
unrealistic
responses,
and
not
to
be
influenced
by
the
survey
questionnaire
or
the
interviewer
who
administers
the
survey
questionnaire.
Furthermore,
the
contingent
valuation
approach
imposes
an
institution
that
leads
to
a
hypothetical
change
in
an
individual's
budget
constraint
by
requiring
that
the
individual
"
pay"
for
the
specified
water
quality
improvement.
Thus,
the
new
budget
constraint
for
the
utility
maximization
process
includes
both
the
prices
and
quantities
of
market
goods
and
the
hypothetical
price
and
defined
quantity
of
water
quality.

The
institutional
design
underlying
contingent
valuation
surveys
requires
that
ownership
of
the
property
rights
for
water
quality
at
the
recreation
site
be
determined
in
the
specification
of
the
question,
thus
affecting
the
appropriate
measure
of
consumer
welfare.
Specifically,
consumer
ownership
of
property
rights
would
indicate
a
willingness­
to­
accept
measure
as
the
appropriate
valuation
concept,
and
industry
ownership
would
dictate
a
willingnessto
pay
measure.
Although
currently
boatable
throughout,
the
Monongahela
River­­
the
site
used
for
this
study
(
see
Chapter
3)­­
supports
swimming
and
fishing
only
upriver
from
Pittsburgh,
and
property
rights
are
in
a
state
of
flux
with
considerable
confusion
over
ownership
(
see
Feenburg
and
Mills
[
1980]).
Thus,
a
reasonable
allocation
for
this
study's
survey
of
Pittsburgh
residents
is
that
consumers
own
the
rights
to
beatable
water
(
which
suggests
an
equivalent
surplus
measure),
while
no
one
yet
owns
the
rights
to
fishable,
swimmable
water
along
the
entire
river
(
which
indicates
a
compensating
surplus
measure).

While
using
a
willingness­
to­
accept
measure
for
maintaining
a
boatable
water
quality
level
and
a
willingness­
to­
pay
measure
for
the
value
of
moving
to
fishable
and
swimmable
levels
is
consistent
with
current
Monongahela
property
rights,
willingness­
to­
accept
measures
have
proven
difficult
in
hypothetical
market
experiments,
thus
creating
serious
problems
in
the
development
of
a
workable
survey
methodology.
For
example,
respondents
have
either
refused
to
answer,
given
infinite
bids,
or
refused
to
accept
any
compensation
f
o
r
r
e
d
u
c
t
i
o
n
s
i
n
e
n
v
i
r
o
n
m
e
n
t
a
l
q
u
a
l
i
t
y
[
Schulze,
d'Arge,
and
Brookshire
[
1981
]
and
Bishop
and
Heberlein,
1979].
To
cope
with
this
problem,
this
study
employs
a
willingness­
to­
pay
(
equivalent
surplus)
measure
for
the
decrease
from
boatable
water
quality
and
a
compensating
surplus
measure
for
improvements
from
the
same
level.

2­
13
2.4.3
Continent
Ranking
Approach
Like
the
other
contingent
valuation
formats
,
contingent
ranking
relies
on
~
individuals'
responses
in
a
hypothetical
market
situation.
However,
instead
of
requiring
an
individual
to
respond
with
the
maximum
willingness
to
pay
for
a
water
qtiality
improvement,
contingent
outcomes
­­
consisting
of
a
hypothetical
water
quality­­
from
most
preferred
to
underlying
contingent
ranking
is
that
to
the
hypothetical
market
when
both
maximization
framework
underlying
the
ranking
requires
tha~
individuals
rank
p
a
y
m
e
n
t
a
n
d
a
c
o
r
r
e
s
p
o
n
d
i
n
g
l
e
v
e
l
o
f
`
least
preferred.
T
h
e
i
m
p
l
i
c
i
t
a
r
g
u
m
e
n
t
an
individual
is
better
able
to
respond
outcomes
are
specified.
In
the
utility
"
contingent
ranking
approach,
an
indi
­
vidual
ranks
the
alternatives
based
on
their
implications
for
his
ability
to
maximize
utility
with
a
given
income,
the
prices
of
other
goods,
and
the
proposed
combination
of
payment
and
water
quality.
Analytically,
this
choice
can
be
described
by
comparisons
of
the
indirect
utility
functions
arising
from
each
of
these
sets
of
decisions.
An
appropriate
compensating
surplus
measure
can
then
be
derived
from
estimates
of
the
indirect
utility
function.

2.5
Summary
Partly
because
they
are
all
based
utility
maximization,
the
travel
cost,
r
a
n
k
i
n
g
ap~
roaches
can
each
develoD
on
the
common
standard
of
constrained
contingent
valuation,
and
contingent
measurements
of
changes
in
consumer
welfare.
The
travel
cost
approach
measures
the
change
in
ordinary
consumer
surplus,
the
contingent
valuation
approach
measures
equivalent
and
compensating
surpluses,
and
the
contingent
ranking
format
yields
a
compensating
surplus
welfare
measure.
*
The
relationship
between
each
of
these
methods'
measures
of
the
welfare
changes
associated
with
water
quality
changes
is
considered
in
the
comparison
analysis
reported
in
Chapter
8.

*
lt
should
be
noted
that,
for
the
contingent
valuation
approaches,
questions
have
b&
en­
formulated
to
include
both
user
and
nonuser
values.
Strictly
~
speaking,
both
approaches
n)
easure
the
option
price,
but
the
contingent
valua
­
~
tion
approach
permits
the
user
value
component
to
be
identified.
!
$

2­
14
CHAPTER
3
SURVEY
DESIGN
3.1
INTRODUCTION
Estimating
the
recreation
and
related
benefits
of
water
quality
improvement
with
the
contingent
valuation
approach
requires
an
integrated
survey
design.
This
chapter
describes
the
survey
design
for
the
case
study
of
the
Monongahela
River.
Specifically,
Section
3.2
describes
the
general
backg
r
o
u
n
d
o
f
t
h
e
Monongahe!
a
R
i
v
e
r
b
a
s
i
n
a
r
e
a
,
Section
3.3
highlights
the
sampling
plan
for
the
project,
and
Section
3.4,
a
discussion
of
the
survey
plan,
concludes
the
chapter
with
detailed
information
on
the
survey
field
procedure.

3.2
GENERAL
DESCRIPTION
OF
THE
MONONGAHELA
RIVER
BASIN
This
section
describes
the
Monongahela
River
basin,
providing
a
general
description
of
river
geography,
river
uses,
river­
related
recreation
activities,
and
a
socioeconomic
profile.

3.2.1
Geography
Formed
by
the
confluence
of
the
West
Fork
and
Tygart
Rivers
near
Fairmont,
West
Virginia,
the
Monongahela
River
drains
an
area
of
7,386
square
miles
in
southwest
Pennsylvania,
northern
West
Virginia,
and
northwest
Maryland.
(
See
Figure
3­
1
for
a
map
of
the
area.
)
It
flows
northerly
128
miles
to
Pittsburgh,
where
it
forms
the
Ohio
River
headwaters
with
the
Allegheny
River,
and
has
two
major
tributaries,
the
Youghiogheny
and
Cheat
Rivers.
All
128
miles
of
the
Monongahela
are
navigable
year
round
by
motorized
commercial
traffic.

Characterized
by
steep
banks
and
rugged
terrain,
the
Monongahela
River
basin
lies
in
five
Pennsylvania
Counties
(
Allegheny,
Greene,
Fayette,
Westmoreland
and
Washington)
and
two
West
Virginia
counties
(
Monongalia
and
Marion
)
in
the
Appalachian
Plateau
and
the
Allegheny
Mountains.
The
Monongahela
basin
currently
supports
four
major
reservoirs:

.
Deep
Creek
Reservoir­­
A
privately
owned
Maryland
facility
operated
on
a
Youghiogheny
River
tributary
to
generate
51
megawatts
of
electric
power.

.
Lake
Lynn
Reservoir­­
A
privately
owned
West
Virginia
facility
operated
on
the
Cheat
River
to
produce
19
megawatts
of
electric
power.

3­
1
t
3>'
______
i
 
 
­
­­­.,//'
W81un
~

I
it%­­%!!­
P7­
1­­­­­­:
­
1
­

\/'

Figure
3­
1.
Map
of
Monongahela
River
and
other
area
recreation
sites.

.
Tygart
River
Reservoir­­
A
facility
operated
by
the
U.
S.
Army
Corps
of
Engineers
to
provide
flood
control,
recreation,
and
low
flow
augmentation.
This
reservoir
provides
most
of
the
Monongahela's
augmented
flow,
a
minimum
of
340
cfs
in
the
upper
river.

.
Youghiogheny
River
Reservoir­­
A
facility
operated
by
the
U.
S.
Army
Corps
of
Engineers
to
provide
a
minimum
flow
of
200
cfs
for
the
Monongahela
River.

Comprising
nearly
30
percent
of
the
river
basin's
seven­
county
area,
the
following
urban
areas
and
boroughs
(
listed
below
with
1970
census
population)
line
the
Monongahela's
banks:

Pittsburgh
McKeesport
Clairten
Duquesne
Monessen
Monongahela
Morgantown
Fairmont
520,117
37,977
15,051
11,410
17,216
7,113
29,431
26,093
Donora
Charleroi
Brownsville
Braddock
Glassport
Munhall
Port
Vue
West
Miff
in
8,825
6,723
4,856
8,795
7,450
16,574
5,862
28,070
3­
2
3.2.2
Uses
As
part
of
the
Mississippi
River
Waterway
System,
the
Monongahela
has
a
9­
foot­
deep
navigation
channel
from
Pittsburgh
to
Fairmont
to
support
both
commercial
and
recreation
river
traffic.
This
navigation
channel
ranges
in
width
from
a
minimum
of
25~
feet
to
nearly
full
river
width
at
the
river's
mouth
and
is
currently
maintained
by
a
series
of
nine
lock
and
dam
structu
res.
The
heaviest
barge
traffic
occurs
at
Structures
2
and
3.
Use
of
the
locks
and
dams
for
generating
hydroelectric
power
is
currently
under
consideration
and
would
provide
an
estimated
total
capacity
of
96.2
megawatts.
To
support
river
traffic,
the
Monongahela's
banks
have
a
boat
dock
concentration
approaching
one
dock
per
mile.
However,
these
docks
­­
which
numbered
147
in
1979­­
are
mostly
single­
purpose,
single­
user
facilities.

Industrial
activity
along
the
Monongahela
is
dominated
by
the
primary
metals
industry,
which
accounts
for
over
31
percent
of
the
area's
total
manufacturing
employment,
including
29
percent
of
all
Pennsylvania's
steel
industry
employment.
Also
important
with
respect
to
industrial
activity
along
the
Monongahela
are
significant
amounts
of
natural
resources,
including
oil
and
gas,
limestone,
sandstone,
sand
and
gravel,
and
coal.
are
estimated
at
approximately
23
billion
tons,
and
the
region
alone
accounted
for
24
percent
of
total
1977
Pennsylvania
and
West
Virginia.
Underground
mining
in
78
percent
of
this
total,
with
strip
mining
operations
remainder.

3.2.3
Recreation
Area
coal
reserves
Monongahela
R
i
v
e
r
coal
production
in
the
area
produced
accounting
for
the
Because
it
essentially
is
a
series
of
large
pools­­
ranging
from
400
to
1,741
surface
acres­­
created
by
its
nine
lock
and
dam
structures,
the
Monongahela
offers
substantial
opportunities
for
recreation.
I
n
fact,
although
the
lower
20
river
miles,
subjected
to
heavy
industrial
and
urban
development,
offer
limited
recreation
opportunities,
the
remaining
108
miles
have
seen
dramatic
increases
in
recreation
usage
over
the
last
10
years,
partially
because
of
improved
water
quality.
As
a
result
of
this
increased
recreation
usage,
numerous
public
and
private
facilities
have
been
developed
along
the
Monongahela,
ranging
from
single­
lane
boat
launching
ramps
to
boat
club
docks,
commercial
marinas,
and
community
parks.

The
primary
recreation
activities
along
the
river
are
power
boating
and
fishing.
Because
power
boating
is
more
popular,
many
recreation
facilities
have
been
constructed
primarily
to
serve
it.
Partially
as
a
result,
the
Monongahela
River
comprises
a
substantial
portion
of
the
water
acreage
available
in
the
region
for
unlimited
horsepower
boating.

Although
it
is
second
to
power
boating
in
popularity,
fishing
occurs
over
a
greater
number
of
water
acres
in
the
area
when
small
lakes
and
streams
are
considered.
I
n
fact,
fishing
accounts
for
approximately
12
percent
of
all
current
uses
of
the
Monongahela.
Fishing
in
the
river
is
encouraged
by
special
programs
in
both
Pennsylvania
and
West
Virginia
to
stock
warmwater
fish,
and
fish
sampling
has
revealed
the
presence
of
up
to
47
separate
species,

3
­
3
plus
3
hybrids.
Of
special
interest,
the
U.
S.
Environmental
Protection
Agency
(
EPA)
and
the
Pennsylvania
Fish
Commission,
which
have
monitored
fish
population
trends
in
the
Monongahela
since
1967,
have
reported
a
dramatic
increase
over
an
Ii­
year
period
in
species'
diversity
and
biomass,
particularly
in
the
upper
reach.

In
addition
to
power
boating
and
fishing,
the
Monongahela
also
offers
other
recreation
opportunities
at
several
major
facilities,
including
two
constructed
by
the
U
.
S.
Army
Corps
of
Engineers
at
the
Maxwell
and
Opekiska
pools;
the
Tenmile
Creek
Recreational
Area
(
adjacent
to
the
Maxwell
Pool),
which
showed
increased
visitor
days
from
1972
to
1975;
and
the
Prikett
Bay
Recreational
Area
(
at
Opekiska
Pool),
which
has
also
experienced
increased
visitation
from
1972
to
1975.
Recreation
activities
offered
by
these
sites
include
picnicking,
camping,
boating,
and
swimming.
Despite
its
length
and
general
popularity
for
recreation,
t
h
e
Monongahela
nowhere
offers
either
campgrounds
or
State
parks
for
potential
recreational
ists,
who
are
forced
to
the
substitute
sites
offered
by
the
Youghiogheny
River
Reservoir
and
the
Allegheny
River.
Both
of
these
substitutes
offer
better
water
quality
than
the
Monongahela
and,
perhaps,
more
scenic
settings
for
recreation.

3.2.4
Socioeconomic
Profile
In
1977,
population
for
the
seven­
county
area
of
the
Monongahela
River
basin
totaled
2,417,885,
which
results
in
an
average
population
density
of
518
persons
per
square
mile.
Although
density
is
greatest
along
the
river,
there
is
a
recent
trend
to
move
into
other
areas.
However,
population
changes
in
the
basin
vary
according
to
State:
several
Pennsylvania
counties
have
experienced
a
noticeable
population
decrease
in
the
period
from
1960
to
1977,
but
Monongalia
County
in
West
Virginia
experienced
a
dramatic
population
increase
during
the
same
period.
in
general,
the
basin
has
a
greater
percentage
of
urban
population
than
either
the
Pennsylvania
or
West
Virginia
State
averages.

Per
capita
income
within
the
basin
is
Iuwer
than
either
the
Pennsylvania
or
West
Virginia
State
averages,
and
the
basin
in
fact
contains
a
higher
percentage
of
persons
living
below
the
poverty
level
than
does
either
State
generally.
Not
surprisingly,
then,
much
of
the
basin's
housing
stock
is
generally
Considered
substandard,
and,
in
1970,
70
percent
of
it
was
more
than
25
years
old.

The
average
education
level,
which
has
steadily
increased
since
1950,
is
higher
in
the
basin
than
it
is
in
either
Pennsylvania
or
West
Virginia
or
in
the
United
States
generally.
However,
the
difference
between
the
basin
and
the
nation
has
almost
disappeared,
eroded
by
a
steadily
rising
U.
S.
education
level.
Another
steadily
eroding
difference
between
the
basin
and
the
nation
as
a
whole
is
in
the
percentage
of
the
workforce
made
up
of
craftsmen
and
laborers.
Specifically,
due
primarily
to
the
area's
heavy
concentration
of
primary
metals
and
extraction
industry,
the
basin
still
has
a
higher
concentration
of
blue
collar
workers
than
does
the
difference
has
greatly
diminished
during
the
last
20
nation
generally,
but
this
years.
.

3
­
4
 
.
,..
.
..
 
.
 
 
 
.
.
­
 
.
.
._
,.
.
.
.
.
.
­
,,.
3.3
SAMPLING
PLAN
The
following
subsections
describe
the
sampling
plan
implemented
to
accomplish
the
objectives
of
this
study.
A
single­
stage,
area
household
sampling
design
was
used
to
contact
approximately
384
sample
households
in
a
four­
county
area
of
southwest
Pennsylvania.
Appendix
A
contains
additional
detaiis
of
the
survey
design,
sample
selection,
and
weight
calculation.

3.3.1
Target
Population
Five
counties
comprised
the
sample
area
for
this
study
(
outlined
in
Figure
3­
2):
Allegeny,
Fayette,
Greene,
Washington,
and
Westmoreland.
These
counties
were
selected
because
they
contain
the
reach
of
the
Monongahela
River
within
Pennsylvania.
The
random
nature
of
the
sample
resulted
in
no
sample
segments
being
chosen
in
Greene
County.
The
target
population
consisted
of
all
households
in
this
five­
county
area.
Group
quarters
were
not
included,
and
only
adult
(
persons
18
years
and
older)
household
members
were
eligible
for
interview.
One
adult
was
selected
for
the
interview
from
each
household.

E
R
I
E
`"

.
...!.
.
I
"..
­,
0,.
,,.'.
,0."
a
.,­
.­,.

Olu",
c.
m
.
.
.
..­.

0.,,,.
ALIENi
OW~
ULIItIfilEt.
1
fASIUN
.
.
.
..­

"
.
"
.

\
.,
.
.
.
..­

...
­.
G,"

II
TSBURCH
UI,
.

LEGEND:
\

@
Ptacw
of
100,000
or
moro
inhtiitants
l
 
PIs~
of
50,0@
to
100,000
inhabitants
a
Central
citia
of
SMSAS
with
fowor
than
50,000
inhabitwrta
O
pl~
o.
g
of
~,~
to
50,000
in~
bitin~
o~~.
~~
s
Figure
3­
2.
Geographic
location
of
survey
area.

3
­
5
3.3.2
Sample
Selection
and
Survey
Design
The
design
was
a
single­
stage,
stratified
cluster
sample.
The
sampling
units
(
S
US)
were
noncompact
clusters
of
approximately
seven
households
each.
The
clusters
were
developed
by
partitioning
all
the
block
groups
(
BGs)
and
enumeration
districts
(
EDs)
within
the
five­
county
area
into
noncompact
clusters.
The
clusters
were
nonoverlapping
and,
when
aggregated,
completely
accounted
for
all
of
the
households
in
the
five­
county
area.

The
sampling
units
were
stratified
into
three
disjoint
groups:
(
1)
Pittsburgh
(
2)
not
in
a
place,
and
(
3)
a
place
other
than
Pittsburgh.
Fifty­
one
clusters
with
an
average
of
7.78
sample
housing
units
(
SHUS)
each
were
selected,
yielding
397
SHUS.
A
roster
of
all
adults
was
compiled
for
each
SHU.
One
adult
was
randomly
selected
from
each
SHU
for
interview.

3.3.3
Sampling
Weights
The
probability
structure
used
to
select
the
SHUS
and
the
adults
within
each
SHU
allows
calculation
of
the
selection
probability
for
each
individual
interviewed.
The
sampling
weights,
reciprocals
of
the
probability
of
selec
­
tion,
were
then
calculated.
selected
SHUS
(
80.59
percent
for
the
nonresponse.

3.4
SURVEY
PLAN
T
h
i
s
Droiect
recauired
a
Because
interviews
were
response),
the
sampling
detailed
survey
plan
to
not
obtained
from
all
weights
were
adjusted
enable
the
successful
completion
of
a
full
ranae
of
survey
tasks.
The
following
subsections
discuss
the
`­
procedures
and
me~
hods
developed
to
carry
out
these
tasks.
The
major
field
tasks
were
as
follows:

.
To
design
and
perform
a
limited
local
pretest
of
the
survey
questionnaire.

l
To
retain
field
interviewers.

l
To
count
and
list
households
within
the
randomly
selected
area
segments.
(
Two
field
supervisors
and
two
interviewers
performed
this
task.
)

l
To
develop
a
field
procedures
manual
and
interviewer
training
materials.

l
To
conduct
a
field
interviewer
training
session.

l
To
administer
the
benefits
instrument
at
randomly
selected
­­
households
within
the
area
segments.
(
One
questionnaire
was
to
be
administered
by
an
in­
person
interview
at
each
sample
household.
The
desired
number
of
interviews
to
be
conducted
was
305.
)

3
­
6
l
.
To
develop
and
implement
onsite
and
off
site
quality
control
procedures
on
the
work
performed
by
the
field
staff.

.
To
conduct
an
interviewer
debriefing.

.
To
develop
and
implement
data
receipt,
data
editing,
and
keypunch
procedures
for
all
resultant
data.

3.4.1
Questionnaire
Design
and
Limited
Local
Pretest
The
design
of
the
benefits
questionnaire
involved
the
combined
talents
of
RTI
staff
knowledgeable
in
benefits
analysis
and
questionnaire
design,
the
EPA
project
officer,
and
selected
consultants.
Efforts
to
design
the
questionnai
re
centered
on
satisfying
the
two
primary
objectives:

.
To
collect
the
data
required
for
analysis
.
To
collect
the
data
in
such
a
way
that
reliability
and
validity
are
enhanced.

In
meeting
these
objectives,
the
number
and
types
of
questions
included
in
the
instrument
and
the
format
that
those
questions
took
were
determined
by
several
interrelated
factors:
Those
factors
included:

.
The
precise
analytic
goals
of
the
survey.

.
The
adequacy
of
the
project
budget
to
support
the
data
collection
required.

.
The
facility
of
the
interviewers
in
administering
the
instrument.

l
The
tolerance
of
potential
respondents
of
the
time
and
effort
required
to
answer
the
questions.

l
The
ability
of
respondents
to
provide
the
data
requested.

Table
3­
1
outlines
questionnaire
development
activity.
After
the
data
collection
was
completed
and
the
interviewers
debriefed,
it
was
clear
that
the
careful
attention
given
to
questionnaire
design
had
reaped
substantial
rewards.
The
nuances
of
the
questions
and
intricate
skip
patterns
made
necessary
by
anticipated
responses
necessitated
a
considerable
investment
of
time
early
in
the
questionnaire
development.

Another
factor
that
had
a
considerable
effect
on
the
overall
quality
of
the
instrument
was
the
variety
of
skills
brought
to
bear
on
the
wording
of
questions.
The
economic
concepts,
of
course,
resided
with
the
economists.
However,
the
wording
of
questions
was
critiqued
by
survey
specialists
for
sensibility
and
administrative
ease
and
further
reviewed
by
staff
experienced
in
questionnaire
formatting
and
overall
survey
methodology.
The
net
effect
of
these
efforts
was
a
questionnaire
that
was
more
comprehensible
than
the
economists
could
have
ever
produced
themselves
and
more
sophisticated
than
the
survey
specialists
alone
would
have
designed.

3
­
7
9
Table
3­
1.
Questionnaire
Development
Activity
Activity
Date
(
1982)

Review
existing
survey
work:
Resources
for
the
August
5
Future,
Inc.
(
RFF)
(
Mitchell);
Colorado
State;
Wyoming
Develop
first
draft
for
presentation
at
workshop
August
10
Revise
draft
for
review
by
EPA
project
officer,
August
17
consultant,
and
survey
specialist
Incorporate
revisions
from
review
August
20
Review
by
survey
staff
August
22
Send
revisions
to
EPA
project
officer
for
review
by
August
24
EPA
survey
liaison
officer
Perform
limited
pretest
in
Raleigh
area
August
26
Revise
Submit
Revise
Submit
instrument
based
on
pretest
August
28
draft
instrument
to
EPA
for
review
September
2
instrument
based
on
additional
pretest
September
6
Office
of
Management
and
Budget
October
9
(
OMB)
package
Incorporate
OMB
suggestions
October
27
OMB
approval
November
5
After
the
instrument
was
developed,
it
was
administered
on
a
limited
pretest
basis
in
the
Research
Triangle
Park,
North
Carolina,
area.
Further
limited
pretesting
of
the
instrument
was
completed
in
Pittsburgh
after
the
Office
of
Management
and
Budget
(
OMB)
package
was
submitted
for
EPA
review.

The
Research
Triangle
Park
pretest
was
conducted
on
people
from
the
Pittsburgh
area
to
detect
major
faux
~
in
the
instrument
that
Triangle­
area
residents
could
not
perceive.
Am
result
of
this
pretest,
several
recreation
sites
were
added
to
the
site
list,
the
groups
of
activities
were
rearranged,
and
a
better
map
was
developed.
Most
of
the
benefits
from
the
pretest
came
from
find?
ng'
flaws
in
the
logic
of
the
questionnaire.
The
pretest
was
especially
helpful
in
determining
what
subsequent
questions
were
appropriate
for
zero
bidders
and
for
bidders
who
gave
a
zero
to
only
certain
parts
of
tho
questionnaire.

3
­
8
t
A
limited
budget
prevented
extensive
pretesting
in
the
target
area.
I
n
future
surveys
this
activity
should
be
budgeted.
Because
of
the
logical
consistency
desired
across
all
items
in
the
questionnaire,
a
pretest
in
the
survey
area
would
reveal
potential
logical
inconsistencies
only
sample
area
residerltS
could
expose
via
their
responses.
Researching
the
river
and
the
sample
area
was
a
viable
substitute;
but
a
pretest
in
Pittsburgh
would
have
been
a
valuable
complement.

3.4.2
Retaining
Field
Supervisors
and
Hirinq
Interviewers
The
project
used
two
field
supervisors
experienced
in
hiring
and
training
interviewers
and
in
managing
survey
fieldwork
to
supervise
and
carry
out
the
count­
and­
list
task
and
to
recruit
the
field
interviewers
who
performed
the
household
interviewing
task.
Because
much
of
the
cost
of
a
data
collection
effort
is
due
to
count­
and­
list
activities
and
to
interviewer
recruiting,
using
off
site
fieid
supervisors
made
the
project's
field
operations
more
economical.
The
survey
task
leader
closely
monitored
the
field
supervisors
in
the
countand
list
and
recruiting
activities,
which
were
carried
out
during
the
week
of
October
19,
1981.

Project
staff
and
the
field
supervisors
worked
together
to
select
the
interviewers
from
among
experienced
applicants
who
had
previously
performed
well
on
similar
surveys.
Top
prospects
in
the
Pittsburgh
area
were
screened
by
telephone
to
verify
general
qualifications,
availability,
and
interest.
During
the
count­
and­
list
activity,
the
field
supervisors
interviewed
some
of
the
best
qualified
applicants
in
person.
Personal
and
work
references
were
checked
before
final
selections
were
made.
Relevant
selection
criteria
included
interest
in
the
objectives
of
the
study,
availability
of
dependable
transportation
perceived
ability
to
relate
well
to
the
sample
population
of
interest,
input
from
personal
and
work
references,
and
interviewing
skills
(
e.
g.
,
ability
to
read
questions
clearly,
to
follow
instructions,
to
use
nondirectional
probes,
to
record
responses
accurately
and
legibly,
etc.).

The
selected
interviewers
were
nine
professionals
who
had
extensive
experience
in
household
surveys,
focus
groups,
census
work~
and
a
variety
of
other
interviewing
activities.
These
interviewers
performed
admirably
throughout
the
data
collection
process,
overcoming
inclement
weather,
a
few
irate
refusals,
and
an
approaching
holiday
season.
This
was
done
with
a
refreshing
enthusiasm
and
reinforced
the
confidence
of
the
project
team
members.
The
interviewers
were
aware
of
all
the
things
that
can
possibly
bias
a
respondent
and
were
careful
to
follow
the
procedures
outlined
in
the
manual
and
covered
in
the
training
session.
In
summary,
the
importance
of
using
experienced,
professional
interviewers
cannot
be
overstated.

3.4.3
Counting
and
Listing
of
Sample
Segments
Two
field
supervisors
and
two
experienced
interviewers
conducted
z!!
counting
and
listing
of
sample
segments.
This
task
involved:

.
Locating
the
segment
.
Identifying
segment
boundaries
3­
9
.
Counting
the
housing
units
l
Listing
all
eligible
housing
units.

The
count­
and­
list
task
was
completed
in
1
week
and
the
materials
returned
for
an
in­
house
check
and
preparation
of
interviewer
assignments.
Appendix
B
shows
samples
of
the
results
from
the
count­
and­
list
activities.
Details
of
how
these
materials
were
used
by
the
interviewers
are
provided
in
the
Field
Interviewer's
Manual,
available
from
Research
Triangle
Institute.

3.4.4
Developing
Field
Manuals
and
Conducting
Interviewer
Training
Because
the
interviewers
were
supervised
from
the
Research
Triangle
Park
during
the
household
interviewing
phase,
a
high
degree
of
administrative
organization
of
field
personnel
was
required
for
the
project.
Interviewers
were
carefully
informed
of
reporting
and
communications
channels,
procedures,
schedule
requirements,
documentation
of
nonresponse,
reassignments,
quality
control
techniques,
and
other
operating
procedures
required
to
complete
the
project
in
a
timely,
cost­
effective
manner.
The
Field
Interviewer's
Manual
provided
the
details
of
the
organization
of
the
field
procedures
and
covered
the
following
topics:

.
Purposes
and
sponsorship
of
the
project
.
Role
of
the
interviewer
.
Data
collection
schedule
.
Field
sampling
and
locating
procedures
.
Contacting
and
obtaining
cooperation
from
sample
members
l
Reporting
results
of
attempts
to
secure
interviews
.
Documentation
of
nonresponse
l
Validations,
field
edits,
and
other
quality
control
procedures
.
Disposition
of
completed
cases
.
Completion
of
administrative
forms
(
e.
g.
,
field
status
reports,
reassignment
forms,
and
production
and
expense
reports)

.
Communications
with
central
office
staff.

In
addition
to
the
Field
Interviewer's
Manual,
a
series
of
administrative
forms
was
­
developed
including
a
household
control
form
(
see
Appendix
B),
which
served
the
following
functions:

.
Provide
assignment
information
for
the
interviewer
(
i.
e.
,
sample
household
address).

3­
1o
.
Provide
the
interviewer
with
an
introductory
statement
explaining
the
survey.

.
Provide
appropriate
household
enumeration
questions
and
queries
to
obtain
demographic
data
on
persons
in
the
sample
household.

.
Provide
the
interviewer
with
instructions
for
selecting
a
household
member
to
be
interviewed.

.
Require
the
interviewer
to
document
all
attempted
and
successful
contacts
with
the
sample
member.

.
Provide
an
appropriate
set
of
result
codes
for
describing
interim
and
final
results
for
each
case.

.
Require
the
interviewer
to
record
certain
information
required
for
validation
of
completed
interviews
and
noninterviews.

The
training
materials
developed
for
the
project
included
background
on
benefits
analysis
and
administrative
procedures.
The
Interviewer's
Manual
and
a
C
O
P
Y
of
the
questionnaire
were
sent
to
the
interviewers
prior
to
their
classroom
training.
A
specified
amount
of
time
was
authorized
for
advance
study,
and
interviewers
were
expected
to
read
the
manual
and
specifications
prior
to
the
training
session.

3.4.5
Training
Session
The
extensive
experience
of
the
interviewers
enabled
the
project
team
to
focus
on
the
unique
aspects
of
the
project
and
to
highlight
the
technical
details
of
the
interviewing
procedures.
The
agenda,
shown
in
Figure
3­
3,
shows
the
variety
of
topics
covered
in
the
2­
day
session
on
November
11
and
12,
1981.

I
n
addition
to
covering
the
project
objectives,
the
training
session
provided
an
opportunity
for
personal
interaction
with
the
interviewers.
The
session
focused
on
benefits,
EPA
water
policy,
the
water
pollution
basics,
and
mock
interviews
with
all
versions
of
the
questionnaire.
The
mock
interviews
included
zero
bidders,
recalcitrant
and
reluctant
bidders,
use
of
the
payment
card,
and
procedural
problems
that
might
be
encountered.
The
interviewers
were
reminded
not
to
provide
supplemental
information
but
to
reread
an
item
as
many
times
as
necessary.
Each
interviewer
received
a
healthy
dose
of
information
on
benefits
methodology
and
the
important
policy
implications
of
the
project.
The
participation
by
the
project
officer
in
the
training
also
conveyed
the
feeling
that
the
interviewers
were
important
to
the
successful
completion
of
the
survey.

3.4.6
Conducting
Household
Interviews
Face­
to­
face
interviews
were
conducted
between
November
13
and
December
20,
1981.
Conducting
the
interviews
involved
a
series
of
interrelated
operations,
which
included
taking
steps
to
obtain
the
desired
number
3­
11
 
 
Field
Intewiewer
Training
Sassion
Agenda
Study
for
Estimating
Recreation
and
Related
Benefits
of
Water
Quality
November
11,
1981
9:
00
a.
m.
9:
10
a.
m.
9:
15a,
m.

9:
45
a.
m.
10:
15
a.
m.
11:
00
a.
m.
11:
15
a.
m.
12:
OOa.
m.
­
l:
OOp.
m.
l:
OOp.
m.

1:
30
p.
m.

2:
30
p.
m.

2:
45
p.
m.
3:
00
p.
m.
5:
00
p.
m.

November
12,
1981
9:
00
a.
m.

9:
30
a.
m.
10:
OOa.
m.
10:
30
a.
m.
12:
OOa.
m.
­
l:
00
p.
m.
l:
OOp.
m.

2:
00
p.
m.
Introduction
of
RTI
staff
and
field
interviewers
Review
of
training
agenda
Project
administrative
procedures
Break/
picture
taking
and
IDs
Explanation
of
the
Benefits
Study
Overview
of
field
interviewer
responsibilities
Locating
sample
housing
units
Lunch
Completing
household
control
form
and
selecting
sample
individuals
Questionnaire
administration
Demonstration
interview
Break
Mock
interview­
Version
A
Adjourn
Questions
and
answers/
discussion
of
yesterday's
session
Water
pollution:
Dimensions
of
a
problem
The
Benefits
Study
Mock
Interview­
Version
C
Lunch
Questions
and
answers
Distribution
of
assignments
Adjourn
Kirk
Pate
Kirk
Pate
Kirk
Pate
Bill
Desvousges
Kirk
Pate
Kirk
Pate
Kirk
Pate
Kirk
Pate
Kirk
Pate/
Bill
Desvousges
Group
Kirk
Pate/
Bill
Desvouages
Bill
Desvousges
Dr.
Ann
Fisher
Group
Figure
3­
3.
Field
interviewer
training
session
agenda.

of
interviews,
instituting
interviewer
assignment
and
re~
ortin~
PrOCe
dU
r
eS
I
making
initial
household­
contacts
and
obtaining
cooperation
~
enumerating
household
members,
and
administering
the
instrument.

Initial
assignments
of
cases
to
interviewers
were
made
on
the
basis
of
each
interviewer's
location
and
characteristics.
Generally,
assignments
were
made
on
the
basis
of
the
interviewer's
geographic
proximity
to
the
sample
segments.
That
was,
of
course,
a
cost­
effective
practice
and
usually
resulted
in
interviewers
sharing
some
c
h
a
r
a
c
t
e
r
i
s
t
i
c
s
w
i
t
h
t
h
e
people
to
be
interviewed

Efforts
were
made
to
equalize
interviewer
workloads;
however,
individual
assignments
were
made
after
careful
consideration
of
factors
related
to
the
difficulty
of
the
areas
assigned
to
each.
Based
on
an
assumed
equal
distribution
of
cases
per
interviewer,
the
average
number
of
cases
initially
assigned
per
interviewer
for
the
6­
week
data
collection
period
was
40.
Under
Number
of
Cases
Assigned,
Figure
3­
4
shows
the
final
case
load
for
each
interviewer
after
adjustments
in
the
field.

3­
12
f.
,

t
I
Rrl
Project
2222­
2
F
IELO
DATA
Collection
WEEKLY
STATUS
REPORT
ESTINATING
BENEFITS
OF
WATER
QUALITK
lie,
k
{
1
6
Dates
C
o
v
e
r
e
d
:
.~
f
@_/
A
to
L/
4.1
L
Oate
Report
P
r
e
p
a
r
e
d
:
LLl
al
&
.

i
"­
="
'­
"
`­
­­
`"
~
.
.
.

Number
En.
meratxon
No
I
n
t
e
r
v
i
e
w
of
cases
A
c
t
i
o
n
Cases
i
n
F
i
n
a
l
Stat~
s
Code+
Fin.
1
S
t
a
t
u
s
Cod.<+
_

FI
N.,
mr
Ass,
Sned
Taken
_
P:~
gress
02
04
05
06
07
_
o
8
20
22
23
26
_~._
26
 
 
 
.

62
0
42
1
3
4
 
.
x­
 
3
1
1
19
0
>
Q
3
1
28
2
4
1
 
5)
0
57
6
4
2
i
Lo
I
3
_
 
 
 
 
 
­
 
 
.
 
.

36
0
36
1
I
2
31
I
64
0
64
I
1
1
56
1
I
~
 
_
l
3
~~

48
0
48
I
6
3&
1
L
2
_

41
0
d]
I
3
33
 
 
I
_
3
6
0
6
~
L
.
­
 
 
 
.
 
44
0
4h
1
1
36_
1
5
20
0
20
1
2
1
1
13
_
1
1
 
.
 
TOTAI,
397
n
197
9
17
18
3
3
3
0
3
2
14
2k
1
3
 
 
`
:
Status
Codes
:
.:.+
Status
C
o
d
e
s
:
O1stributi.
n
L.
st:

0
2
No
E!
l,
lm.
raLio"
El
,~,
ble
HOIII.
2
0
C
o
m
p
l
e
t
e
d
I"
t.
erviev
B
.
DeSVO"
SReS
O&
E
r
!
,
,
m
.
r
a
t
i
o
n
R
e
f
u
s
e
d
2
2
[
nterv,
ew
f
l
e
e
.
k
.
f
f
K
.
Pate
0
5
[.
aneu.
e.
R.
rr,
?
r
2
3
N
o
t
a
t
H
o
m
e
/
N
o
C
o
n
t
a
c
t
1)
Smith
06
V.+
c.,>
t
Sllu
26
R
e
f
u
s
e
d
SOC
D
e
p
t
.
2
Fil.
s
(
2
2
2
2­
2
]

01
Not
>.
S11(
1
2
5
L.
mg.
age
Barr,
er
0
8
Other
2
6
Other
I
Figure
3­
4.
Summary
of
completed
interviews.

Once
interviewer
assignments
were
identified,
interviewers'
names
were
associated
with
each
household
control
form.
Thus,
manual
control
of
assignments
was
established
and
maintained.
This
control
of
assignments
was
updated
weekly
on
the
basis
of
status
reports
and
receipt
of
completed
work.

Once
assignments
were
issued
at
the
conclusion
of
training,
rigid
reporting
procedures
were
implemented.
At
a
specified
time
each
week,
each
interviewer
telephoned
the
survey
specialist
and
reported
the
status
of
each
as­
`
signed
case,
using
the
current
status
code
from
his
copy
of
the
household
control
form.
The
staff
membe~
entered
the
codes
on
a
field
status
form
for
the
reporting
period
and
discussed
each
active
case
showing
no
progress
or
indicating
a
problem.

3.4.7
Initial
Contacts
and
Obtaining
Cooperation
Obtaining
cooperation
depended
upon
the
persuasiveness
of
interviewers,
who
,
as
a
result
of
training
and
experience,
were
able
to
communicate
to
respondents
their
own
convictions
regarding
the
importance
of
the
study.
There
was
no
major
problem
in
obtaining
respondent
cooperation.
interviewers
indicated
that
people
who
were
uncooperative
for
this
project
were
no
different
from
other
survey
experiences
in
the
Pittsburgh
area.

3­
13
3.4.8
Household
Enumeration
Once
the
interviewer
made
contact
with
an
eligible
household
member,
he
proceeded
to
enumerate
all
individuals
residing
in
the
household.
This
procedure
ensured
that
each
age­
eligible
individual
was
given
a
chance
to
be
selected
for
interviewing.
All
reasonable
field
efforts
were
made
to
interview
all
sample
individuals.
The
following
situations
were
anticipated
and
were
handled
as
indicated
below:

l
Field
efforts
were
discontinued
once
it
was
determined
that
a
sample
member
had
moved
outside
the
sample
counties.

.
'
F
i
e
l
d
e
f
f
o
r
t
s
w
e
r
e
d
i
s
c
o
n
t
i
n
u
e
d
u
p
o
n
l
e
a
r
n
i
n
g
t
h
a
t
sample
members
were
deceased
or
institutionalized.

l
When
non­
English­
speaking
respondents
were
encountered,
an
attempt
to
identify
a
close
relative
to
serve
as
interpreter
was
made
in
an
effort
to
complete
the
interview.
There
was
only
one
interview
with
a
language
barrier,
so
no
special
effort
was
made
in
this
area.

.
An
initial
call
and
at
least
three
additional
callbacks
were
made
at
different
times
of
the
day
and
different
days
of
the
week
in
an
attempt
to
establish
contact
with
sample
individuals
to
com
­
plete
the­
interview.

l
Contacts
with
neighbors
were
made
after
obtain
"
best
time
to
call"
information.

The
enumeration
process
was
facilited
by
the
control
form
(
see
Appendix
B),
which
contained
questions,
and
recording
mechanisms
to
assist
the
the
second
call
to
design
of
the
household
procedural
instructions,
nterviewer
in
identifying
and
listing
household
members
and
determining
sample
status.
Procedures
for
assigning
appropriate
unique
identifiers
were
also
included.

3.4.9
Interviewing
Procedures
Interviewers
were
instructed
to
attempt
to
conduct
interviews
immediately
following
the
enumeration
process
when
the
sample
member
was
identified
and
if
he
were
available.
If
necessary,
appointments
were
made
to
return
at
a
time
convenient
for
the
sample
member.
All
interviews
were
completed
by
means
of
a
face­
to­
face
interview.
The
average
length
of
a
completed
interview
was
approximately
35
minutes.

Table
3­
2
highlights
the
final
tally
from
the
field
data
collection.
The
final
number
of
sample
housing
units
was
397
due
to
the
discovery
by
field
interviewers
of
13
housing
units
not
listed
during
the
listing
phase
of
the
­.
­

3­
14
Table
3­
2.
Final
Distribution
of
Sample
Housing
Units
Result
category
Number
Percentage
of
SHUS
Out­
of
­
Scope
a
SHUS.
.
.
.
.
.
.
.
.

Vacant
Not
an
HU
In
­
Scope
b
SHUS
.
.
.
.
.
.
.
.
.
.
.

No
enumeration
eligible
at
home
Enumeration
refused
Other
enumeration
result
Completed
interviews
Interview
breakoff
Sample
individual
not
at
home
Sample
individual
refused
Language
barrier
Other
interview
result
21
18
4.53
3
.76
 
21
5.29
(
of
397
SHUS)

376
9
17
3
303
2
14
24
1
2.39
4.52
.80
80.59
.53
3.72
6.38
.27
3
.
8
0
376
100.00
aOut­
of­
scope
refers
to
sample
housing
units
not
included
in
response
rate
calculation.
b
In­
scope
refers
to
sample
housing
units
included
in
response
rate
calculation

project.
*
The
interviewers
completed
303
interviews
during
the
data
collection
period
of
November
13
through
December
20,
1981
­­
two
interviews
short
of
the
desired
goal.
The
response
rate
(
80.59
percent)
was
ever
so
slightly
above
the
anticipated
80
percent
rate,
while
the
refusal
rate
equaled
10.90
percent.

*
The
count­
and­
list
process
is
an
imperfect
one
because
interviewers
are
not
required
at
that
stage
to
actually
knock
on
each
door
in
an
effort
to
identify
housing
units
(
H
US).
Procedures
for
discovering
HUS
m
i
s
s
e
d
durjng
the
listing
process
are
implemented
during
the
household
interviewing
stage.
The
inclusion
of
each
missed
HU
in
the
survey
improves
the
statistical
representativeness
of
the
initial
sampling
frame.

3­
15
I
­.
 
 
 
.
.

.
.
.
.
.
.
.
.
.
.
.
,.
.,.
.
.
Twenty­
three
sample
households
either
cllcl
not
complete
tne
!
ntervlew
or
refused
to
cooperate.
These
were
23
cases
in
which
either
no
one
was
at
home
to
provide
the
enumeration
or
the
enumeration
of
the
household
members
was
obtained
but
the
sample
individual
was
never
available
to
complete
the
interview.
The
crush
of
the
Christmas
holidays
and
a
week
of
inclement
weather
conditions
prevented
resolution
of
these
cases.
Without
either
of
these
hindrances,
it
is
not
unreasonable
to
expect
that
an
additional
15
to
20
interviews
could
have
been
obtained
by
the
interviewers.

3.4.10
Interviewer
Debriefing
The
project
staff
and
the
project
officer
conducted
a
l­
day
debriefing
session
in
mid­
December.
This
session
provided
an
opportunity
for
the
interviewers
to
evaluate
survey
procedures
and
the
questionnaire
relative
to
their
other
interviewing
experiences.
The
overall
conclusion
of
the
debriefing
session
was
that
the
questionnaire
was
generally
easy
to
administer
and
that
there
were
few
major
problems.

The
comments
that
follow
represent
general
impressions
and
evaluations
of
the
interviewers.
There
is
no
way
to
validate
them,
but
they
certainly
provided
valuable
insight
for
the
project
staff.
The
debriefing
session
was
highly
valuable
for
project
staff,
both
in
terms
of
current
project
and
ideas
for
handling
problems
in
future
efforts.

Training
Materials
.
More
background
on
water
pol
been
helpful.

.
Background
and
policy
setting
ution
and
recreation
would
have
provided
"
keys"
for
getting
in
doors.
Interviewers
simply
found
it
easier
to
pique
people's
interest
because
they
understood
the
project
objectives
better.

.
More
explanation
of
the
payment
vehicle­­
how
people
are
currently
paying
for
water
pollution
in
higher
prices
and
taxes
­­
would
have
been
helpful
to
the
interviewers.

I
nterviewinq
Process­­
General
Comments
.
Count­
and­
list
maps
and
materials
worked
well.

.
Drinking
water
was
a
major
concern
of
many
people,
especially
the
elderly.
This
was
not
addressed
in
our
instrument
because
of
the
recreation
focus.

.
There
were
occasions
in
which
a
spouse
intervened
or
critiqued
the
interview
responses
of
the
sample
individuals.
The
interviewers
felt,
however,
that
the
respondents
gave
responses
that
reflected
the
households'
views.

3­
16
I
l
Refusals
were
generally
three
types:
busy,
timid,
or
nasty.
This
was
no
different
from
other
household
surveys,
according
to
interviewers.

.
Thirty
minutes
was
the
ideal
length
both
in
terms
of
administration
and
getting
critical
cooperation
of
respondents.

Evaluation
of
Specific
Parts
of
Questionnaire
.
Section
A,
with
activities
listing
and
sites,
worked
very
well.
Easy
to
administer
and
established
interest
of
many
respondents
especial!
y
recreators.

.
Section
B
introduction
is
still
wordy,
especially
B­
1
introduction
"
Season"
ticket
needed
after
advance
in
introduction.

.
B­
2.
needed
a
skip
pattern
for
.
Few
problems
with
B­
3
or
B­
4.

.
There
was
some
confusion
in
non
recreators.

B­
5
as
to
how
to
interpret
zero
response
to
this
question.
Does
it
mean
no
change
or
a
complete
reduction?
This
will
require
careful
attention
in
analysis
There
was
also
some
confusion
over
how
the
water
quality
might
be
bad
sometimes
and
not
at
other
times.

.
Few
problems
with
B­
6.

.
There
was
some
concern
in
B­
7
whether
the
amount
given
was
the
total
amount
already
given,
a
new
amount
independent
of
other
amounts,
or
an
amount
in
addition
to
those
given
earlier.

Visual
Aids
l
Map
and
water
quality
ladder
worked
well.

.
Visual
aid
showing
how
(
but
not
how
much)
people
are
currently
paying
was
needed
to
aid
less
perceptive
respondents.

.
Rank
order
card
design
was
effective.
People
had
little
trouble
connecting
levels
and
dollar
amounts,
but
cards
should
have
been
larger
for
easier
use.

.
Numbers
on
scale
in
water
quality
ladder
were
too
small
elderly
respondents.

.
There
could
have
been
several
more
sites
on
the
site
listing.
for
l
A
better
visual
aid
is
needed
for
"
use­­
might
use,
"
perhaps
with
color
and/
or
larger
print.

3­
17
I
­
 
____

Questionnaires
l
The
direct
question
card
was
the
most
~
eo~
le
often
seemed
of
willingness
to
pay
without
a
payment
difficult
version
to
administer
because
uncomfortable
without
some
aid
(
consistent
with
Mitchell
and
Carson's
[
1981
]
finding).
The
payment
card
was
the
easiest
to
administer.

.
The
bidding
games
usually
reached
an
amount
quickly
as
respondents
supplied
amounts
after
seeing
how
the
process
worked.
The
$
125
starting
point
for
each
level
was
high
relative
to
many
bids
making
this
slightly
embarrassing
for
the
interviewers
to
administer.
Reason
for
high
amount
was
to
test
for
bias
due
to
starting
points.

.
S
p
e
c
i
f
i
c
s
u
g
g
e
s
t
i
o
n
s
f
o
r
r
e
v
i
s
i
n
g
t
h
e
q
u
e
s
t
i
o
n
n
a
i
r
e
a
r
e
presented
in
Appendix
D.

3.4.11
Data
Receipt,
Editing,
and
Keypunchin~

The
last
phase
of
the
survey
process
required
careful
handling
of
the
survey
data,
coding,
editing,
and
keypunching.
Appendix
B
provides
the
details
of
this
process.
In
general,
completed
questionnaires
were
received
from
the
interviewers
on
a
flow
basis
during
the
data
collection
period.
In­
house
editing
was
performed
by
the
survey
specialist
for
the
purpose
of
detecting
any
irregularities.
As
necessary,
irregularities
were
discussed
with
,
the
appropriate
interviewer.
i
The
only
major
coding
of
responses
that
was
required
involved
the
occupation
questions.
The
verbatim
responses
were
coded
into
the
occupation
classes
from
the
Bureau
of
the
Census.
*
Household
control
form
and
questionnaire
data
were
keypunched
on
cards
and
verified
before
analysis
began.

*
March
1971,
publication
from
the
Census
of
Population,
U.
S.
Department
of
Commerce,
Washington,
D.
C.
20233.

3­
18
CHAPTER
4
CONTINGENT
VALUATION
DESIGN
AND
RESULTS:
OPTION
PRICE
AND
USER
VALUES
4.1
INTRODUCTION
Application
of
the
contingent
valuation
approach,
also
referred
to
as
the
direct
survey
approach
in
environmental
economics,
asks
individuals
their
dollar
valuation
of
a
nonmarket
"
commodity"
­­
i.
e.
,
some
good
or
service
not
traded
in
an
actual
market.
*
I
n
environmental
applications,
the
analyst
must
create
a
hypothetical
market
by
describing
how
individuals
would
pay
for
specific
improvements
in
environmental
quality.
For
this
benefits
study
of
the
Monongahela
River
basin,
the
contingent
valuation
design
used
a
household
survey
to
ask
individuals'
valuation
in
terms
they
could
understand­­
terms
t
h
a
t
translate
the
water
quality
improvements
into
a
d
d
i
t
i
o
n
a
l
a
c
t
i
v
i
t
i
e
s
,
s
u
c
h
as
swimming
and
recreation
fishing,
that
individuals
could
undertake
along
the
Monongahela
River.

Contingent
valuation
offers
the
analyst
considerable
flexibility
in
designing
the
"
commodity"
to
be
valued
in
the
hypothetical
market.
At
the
same
time,
however,
it
requires
that
he
take
considerable
care
in
designing
the
market
so
it
is
both
credible
and
understandable
to
the
respondent.
Indeed,
research
suggests
that
contingent
valuation
results
may
be
sensitive
to
the
question
formats
used
to
elicit
an
individual's
valuation,
the
mechanism
used
to
obtain
the
hypothetical
payments
(
payment
vehicle­­
e.
g.,
user
fee
or
utility
bill
increase),
and
the
interviewers
used
to
conduct
the
survey.
To
give
useful
results,
the
survey
design
must
successfully
surmount
these
influences.

The
contingent
valuation
design
for
estimating
the
recreation
and
related
benefits
of
improved
water
quality
in
the
Monongahela
River
used
research
methods
in
fields
ranging
from
survey
and
sample
design
to
resource
economics
This
chapter
traces
the
origins
of
the
design,
describes
the
survey
questionnaire,
characterizes
the
survey
respondents,
and
presents
the
results
on
option
price
and
user
value
for
the
water
quality
improvements.

Section
4.2
reviews
survey
design
issues,
paying
close
attention
to
potential
biases
in
contingent
valuation
research,
and
Section
4.3
describes
major
components
of
the
survey
questionnaire,
including
the
design
for
determining
*
The
interpretation
of
the
valuation
requested
of
respondents
will
depend
upon
the
nature
of
the
question.
For
example,
whether
a
willingness­
to­
pay
or
willingness­
to­
sell
measure
is
elicited
wiil
de"
pend
and
nature
of
the
change
proposed
in
the
question.

4­
1
on
the
property
rights
m.
differences
in
techniques
to
elicit
option
price
responses,
the
selection
of
a
payment
vehicle,
and
the
design
of
tests
for
achieving
plausible
results.
Section
4.4
characterizes
the
survey
respondents
and
the
main
groups
of
interest
among
them
(
users
and
nonusers
of
the
river
and
people
who
refused
to
pay
q
amount
for
improved
water
quality),
Section
4.5
describes
the
estimated
values
for
option
price
and
the
statistical
analyses
of
these
estimates,
and
Section
4.6
provides
the
same
information
for
user
values.
Section
4.7
summarizes
the
chapter's
main
findings.

4,2
A
REVIEW
OF
DESIGN
ISSUES
IN
CONTINGENT
VALUATION
SURVEYS
In
constructing
a
hypothetical
market,
the
contingent
valuation
approach
defines
the
commodity
to
be
valued,
specifies
how
the
exchange
would
occur,
and
describes
the
other
structural
elements
of
the
market.
Brookshire,
Cummings,
et
al.
[
1982]
have
labeled
this
process
as
"
framing
the
question,
"
or
as
simply
setting
the
context
presented
to
respondents
as
part
of
the
contingent
valuation
experiment.
As
with
almost
any
type
of
experimental
design,
the
context
can
influence
the
outcome.
For
example,
within
the
range
of
different
contingent
valuation
contexts
,
an
individual
might
participate
directly
in
a
bidding
procedure
to
elicit
willingness
to
pay
for
the
hypothetical
commodity
might
directly
reveal
this
value
(
with
or
without
the
aid
of
some
type
of
payment
card),
or
simply
might
evaluate
(
rank)
various
outcomes
of
the
hypothetical
market,
as
in
the
case
of
the
contingent
ranking
format.

Partially
because
of
this
range
of
contexts,
the
various
attempts
to
classify
the
methods
for
implementing
the
contingent
valuation
approach­­
and
their
design
features
­­
have
created
considerable
confusion.
Therefore,
to
consider
the
context
of
the
contingent
valuation
approach
used
for
the
Monongahela
River
basin,
this
section
is
organized
according
to
the
approach's
potential
biases.
These
biases
are
not
neatly
compartmentalized;
rather,
they
are
overlapping
and
in
some
cases
interrelated.
(
Indeed,
one
analyst's
strategic
bias
is
another's
hypothetical
bias.
)
At
the
risk
of
blurring
the
boundaries
between
compartments,
the
section
notes
the
most
important
of
these
interrelationships.
The
boundaries
themselves
may,
in
large
part,
be
a
question
of
judgment.

4.2.1
Hypothetical
Bias
Hypothetical
bias
in
contingent
valuation
surveys
is
the
bias
attributable
to
the
use
of
a
hypothetical,
not
an
actual,
market
situation,
and
it
arises
when
individuals
cannot
or
will
not
consider
the
questions
in
a
manner
that
corresponds
to
how
they
would
treat
the
actual
situation.
Consequent
y,
we
can
expect
that
they
provide
inaccurate
answers
to
the
contingent
valuation
questions
about
it.
Mitchell
and
Carson
[
1981
]
argue
that
hypothetical
bias
may
increase
respondents'
uncertainty
and
ambivalence
about
the
hypothetical
experiment
or
induce
them
to
provide
answers
that
they
perceive
are
socially
desirable..
In
general,
hypothetical
bias
may
result
in
respondents
rejecting
or
refusing
to
participate
in
the
contingent
valuation
experiment,
but
the
net
effect
is
to
increase
the
statistical
variance
and
to
lessen
the
reliability
of
the
estimated
willingness­
to­
pay
amounts.

4­
2
r"
?

I
The
empirical
evidence
on
hypothetical
bias
is
somewhat
mixed,
with
some
studies
hindered
by
it
and
others
showing
no
evidence.
To
test
for
several
biases,
Bohm
[
1971]
designed
an
experiment
that
compared
alternative
bidding
and
payment
schemes
for
the
valuation
of
public
television.

tives
were
provided
Several
alternato
respondents,
and,
in
some
cases,
the
respondents
were
~
ctuaiiy
9iven
money
to
spend
on
several
alternatives
to
public
television.
B~
hm
compared
results
from
the
group
that
answered
hypothetical
willingnessto
pay
questions
with
those
from
a
group
that
actually
had
to
pay
for
public
television.
The
willingness­
to­
pay
bids
from
respondents
who
had
to
pay
for
public
television
were
less,
and
significantly
different,
than
those
from
respondents
who
were
simply
asked
how
much
they
were
willing
to
pay.
These
results
imply
that
hypothetical
and
strategic
behavior
were
present
in
the
contingent
valuation
approach.

Mitchell
and
Carson
[
1981
]
question
Bohm's
[
1971]
conclusion
on
hypothetical
bias
based
on
a
reinterpretation
of
his
statistical
evidence.
Bohm's
results
showed
that
onl
Y
one
9rou
P
out
of
six
had
different
mean
values
when
structured
across
different
types
of
information
and
market
actuality.
The
group
that
did
exhibit
higher
willingness­
to­
pay
amounts
was
also
the
group
that
had
higher
incomes,
which,
Mitchell
and
Carson
argue,
may
account
for
t
h
e
size
of
its
mean
w
i
l
l
i
n
g
n
e
s
s
­
t
o
­
p
a
y
b
i
d
.
This
same
group
also
had
one
outlier
that
raised
the
mean
bid
considerably.
If
the
outlier
is
removed,
the
mean
payment
is
reduced
to
a
level
at
which
it
is
no
longer
a
statistically
significant
difference
in
the
means.

Bishop
and
Heberlein
[
1979]
designed
a
mail
survey
that
compared
hypothetical
willingness­
to­
pay
amounts
and
actual
willingness
to
sell.
In
this
study
respondents
were
mailed
checks
in
randomly
selected
amounts
and
requested
to
sell
a
hunting
license
they
had
previously
purchased.
The
authors
found
that
the
amounts
the
respondents
were
willing
to
accept
for
their
hunting
licenses
when
presented
with
an
actual
check
were
considerably
less
than
the
willingness­
to­
pay
amounts
they
gave
in
the
hypothetical
bidding
game
portion
of
the
experiment.
However,
the
results
of
the
hypothetical
and
simulated
market
experiment
suggested
that
the
hypothetical
market
underestimated
willingness
to
pay
relative
to
the
actual
estimates
from
the
simulated
market.
The
Bishop­
Heberlein
findings
suggest
hypothetical
bias
may
be
a
significant
problem
in
contingent
valuation
survey
design,
but
the
implications
of
their
research
may
be
limited
by
their
experimental
design.

Significantly,
the
results
of
several
studies
have
indicated
that
hypothetical
bias
may
contribute
to
the
considerable
variability
in
contingent
valuation
estimates
of
willingness
to
pay.
For
example,
the
Brookshire,
Ives,
and
Schulze
[
1976]
and
Brookshire
et
al.
[
1979]
air
quality
studies
explain
less
than
10
percent
of
their
bid
variation
by
either
socioeconomic
variables
or
changes
in
the
level
of
the
environmental
good
that
the
survey
was
designed
to
measure.

While
not­­
invalidating
the
approach
as
a
means
of
measuring
consumers'
willingness
to
pay,
the
potential
for
hypothetical
bias
in
contingent
valuation
surveys
indicates
the
need
for
considerable
attention
in
the
instrument
design
phase
to
provide
a
credible
survey
questionnaire.
The
respondent
must
be
4
­
3
able
to
perceive
the
experiment
as
a
realistic
approach
to
measuring
the
good
under
consideration.
Aizen
and
Fishbien
[
1977]
have
shown
that
the
more
closely
a
hypothetical
experiment
corresponds
with
actual
situations,
the
greater
the
chance
of
reducing
hypothetical
bias.
Mitchell
and
Carson
[
19811
argue
that
reducing
hypothetical
bias
in
a
contingent
valuation
survey
instrument
does
not
necessarily
lead
to
increased
probability
of
incurring
strategic
bias
(
where
a
respondent
attempts
to
influence
results)
or
other
types
of
biases.
Rather,
they
suggest
that
a
hypothetical
experiment
in
which
the
market
realism
is
high
and
consequence
realism
is
low
will
reduce
or
minimize
each
type
of
bias.
That
is,
respondents
will
perceive
that
a
hypothetical
situation
closely
corresponds
to
a
real
market
situation
(
high
market
realism),
but
they
will
not
perceive
the
nature
of
the
consequences
of
the
hypothetical
experiment
(
to
themselves)
to
the
extent
that
they
will
attempt
to
influence
the
outcome
(
low
consequence
realism).

The
Mitchell
and
Carson
position
differs
considerably
from
that
of
Schulze
et
al.
[
1981],
who
argue
that
the
potential
for
strategic
bias
increases
when
hypothetical
bias
is
reduced.
Mitchell
and
Carson
present
a
viable
alternative
to
the
Schulze
position
in
showing
that
both
biases
can
be
overcome
in
survey
design.
Specifically,
Mitchell
and
Carson
were
able
to
explain
a
considerably
larger
percentage
of
the
variation
in
willingness
to
pay
than
could
authors
of
most
earlier
contingent
valuation
studies
and
did
not
find
evidence
of
strategic
behavior
on
the
part
of
respondents.
Furthermore,
the
Mitchell
and
Carson
results
are
particularly
encouraging
because
their
hypothetical
market
design
offered
national
water
quality
as
a
product,
an
unconventional
situation
that
should
be
particularly
sensitive
to
hypothetical
bias.

4.2.2
Strategic
Bias
The
concern
for
strategic
bias
is
usually
attributed
to
Samuel
son
[
1954],
who
suggested
that
any
attempt
to
value
public
goods
will
be
plagued
by
incentives
on
the
part
of
individuals
or
respondents
to
behave
strategically.
Samuel
son
argued
that,
if
individuals
perceive
they
will
be
able
to
obtain
a
public
good
and
enjoy
its
consumption,
they
may
indeed
try
to
obtain
this
public
good
by
not
revealing
their
true
preferences.
T
h
e
t
h
r
u
s
t
o
f
t
h
e
Samuel
son
argument
for
questionnaire
design
is
that,
depending
on
how
respondents
perceive
the
consequences
of
the
hypothetical
experiment,
they
may
behave
strategically.
For
example,
an
environmentalist
who
thinks
his
bid
might
affect
some
environmental
policy
may
bid
higher
than
his
true
willingness
to
pay
in
order
to
increase
the
average
bid,
provided
he
knows
he
will
not
have
to
pay
based
on
these
bids.
Alternatively,
if
an
individual
believes
his
payment
will
be
based
on
responses
given
to
the
questions,
there
will
be
incentives
to
conceal
true
preferences
provided
the
individual
is
reasonably
sure
the
good
will
be
provided.

The
empirical
evidence
on
strategic
behavior
in
contingent
valuation
surveys
has
generally
found
that
strategic
behavior
is
not
a
major
problem
for
interpreting
willingness­
to­
pay
amounts.
For
example,
Brookshire,
Ives,
and
Schulze
[
1976]
and
Rowe,
d'Arge,
and
Brookshire
[
1980]
attempted
to
design
experiments
that
would
indicate
the
existence
of
strategic
bias.
In
these
experiments,
respondents
were
asked
to
reveal
their
willingness
to
pay
for
I
i
4
­
4
a
~
hange5
in
a
public
good,
which,
if
provided,
would
in
turn
require
them
to
paY
their
share
of
the
mean
of
all
bids.
Brookshire
et
al.
[
1979]
show
that,
for
respondents
to
engage
in
strategic
behavior
in
the
type
of
situation
used
in
the
Brookshire
and
the
Rowe,
d'Arge,
and
Brookshire
studies,
they
would
have
to
know
not
only
the
amounts
that
other
individuals
had
bid,
but
also
the
number
of
bidders
who
had
already
been
asked
and
their
mean
bid.
Both
studies
concluded
that
strategic
bias
was
not
evident
in
the
sample
data
generated
when
respondents
were
told
they
would
have
to
pay
the
mean
of
the
sample.
The
Brookshire
test
for
strategic
bias
examined
the
distribution
of
the
bids,
arguing
that
strategic
bias
leads
to
a
bimodal
distribution
in
which
the
means
for
environmentalists
are
concentrated
in
the
high
values
of
the
distributionT~~
il:
o~
hee
means
for
nonenvironmentalists
fall
primarily
at
the
other
extreme.
d'Arge,
and
Brookshire
test
involved
a
more
rigorous
statistical
analysis
b~
t
found
no
support
for
strategic
bias
after
problem
bids
were
eliminated.
This
study
also
provided
one
group
of
respondents
with
information
on
the
sample
mean
bid
after
it
had
made
its
bid
and
allowed
it
to
change
on
the
basis
of
this
new
information.
The
authors
found
that
only
one
respondent
desired
to
change
an
overall
bid.
The
complexity
of
the
survey
questionnaire
used
in
the
Rowe
study,
as
well
as
the
methods
used
to
screen
observations
omitting
some
bids
from
the
sample,
limits
the
generality
of
the
study
results.
A
study
by
Brookshire,
Ives,
and
Schulze
[
1976]
also
found
no
evidence
of
strategic
bias
in
an
examination
of
the
distribution
of
willingness­
to­
pay
amounts.

Mitchell
and
Carson
[
1981
]
argue
that
the
distribution
test
used
to
indicate
strategic
bias
in
these
earlier
studies
is
inappropriate
because
it
is
impossible
for
most
willingness­
to­
pay
distributions
to
have
standard
normal
distributions.
They
argue
that
the
likely
distribution
is
a
Iognormal
one,
as
shown
in
their
empirical
results.
Unfortunately,
there
are
two
problems
with
the
Mitchell
and
Carson
results
on
strategic
bias.
First,
their
sample
was
subsegmented
into
groups
by
income
levels,
which
could
have
influenced
the
hypothesized
relationship
between
willingness
to
pay
and
income.
Second,
Mitchell
and
Carson's
results
were
limited
by
a
substantial
number
of
zero
bidders
and
protest
bidders
who,
given
the
limitations
of
the
experimental
design,
prevented
them
from
eliciting
additional
information
on
true
preferences.

A
forthcoming
report
by
Cronin
[
1982]
on
willingness
to
pay
for
improved
water
quality
in
the
Potomac
River
suggests
the
existence
of
strategic
bias.
The
design
of
this
study
partitioned
respondents
into
groups
based
on
whether
they
would
actually
have
to
pay
their
bid
through
increased
local
taxes
based
on
the
mean
bid
or
would
have
to
pay
very
little
because
the
Federal
government
would
pay
for
most
of
it.
A
comparison
across
the
two
groups
showed
statistically
significant
differences
in
the
mean
willingness­
to­
pay
amounts
that
are
consistent
with
the
presence
of
strategic
bias.
`
Some
caution
is
interpreting
the
Cronin
finding
because
of
a
poorly
designed
survey
naire
and
specification
problems
in
the
willingness­
to­
pay
equation.

Based
on
the
evidence
that
currently
exists,
strategic
bias
pervasive
problem
that
researchers
oriqinallv
feared.
However,
it
needed
in
questions
not
the
may
be
a
problem
if
the
questionnaire
design
do&
no~
provide
a
Iow­
degree­
of;
con
sequence
realism.
Mitchell
and
Carson
[
1981
]
conclude
that
effectively
designed
survey
questionnaires
can
achieve
the
required
degree
of
realism.

4
­
5
4.2.3
Payment
Vehicle
Bias
Payment
vehicle
bias
occurs
when
a
respondent
is
influenced
by
the
method
of
payment
selected
for
the
contingent
valuation
study.
A
number
of
different
payment
methods
comprise
the
range
of
payment
vehicles:
user
fees,
increases
in
utility
bills,
and
higher
consumer
prices
and
taxes.
To
be
effective
a
payment
vehicle
must
be
realistic
and
familiar
to
respondents
so
they
consider
it
plausible
and
realize
the
implications
of
the
implied
payment
frequency
for
their
total
willingness
to
pay
in
a
given
time
period.
The
ideal
payment
vehicle
would
combine
believability
with
a
wide
range
of
alternative
payment
amounts.

The
contingent
valuation
literature
indicates
very
little
about
the
influence
of
payment
vehicle
bias.
In
the
only
study
that
systematically
examined
this
bias,
Rowe,
d'Arge,
and
Brookshire
[
1980]
found
that
the
type
of
payment
vehicle­­
utility
bill
or
payroll
deduction
­­
had
a
significant
effect
on
willingness
to
pay.
One­
l
ikely
consequence
of
a
particular
p~
yment
vehicle
is
that
ii
may
condition
respondents
to
a
range
of
values
their
responses
are
expected
to
take.
For
instance,
when
a
user
fee
is
selected
as
the
payment
vehicle,
it
is
quite
possible
that
the
respondent
will
think
in
fees.
Thus,
payment
vehicle
bias
may
actually
discussed
below.
On
the
other
hand,
general
to
"
pure"
payment
vehicle
bias,
in
which
the
vehicle
itself.

4
.
2
.
4
Startinq
Point
Bias
The
contingent
valuation
literature
has
question
of
starting
point
bias­­
the
influence
terms
of
a
usual
range
for
user
show
up
as
starting
point
bias,
resentment
of
taxes
could
lead
respondent
rejects
the
payment
devoted
more
attention
to
the
of
the
starting
points
used
in
iterative
bidding
(
or
any
contingent
valuation
procedure
that
uses
starting
point
"
keys,
"
such
as
the
Mitchell­
Carson
[
1981
]
payment
card)
­­
than
it
has
to
the
other
biases.
In
an
evaluation
of
willingness
to
pay
for
air
quality
in
the
Farmington,
New
Mexico,
area,
for
example,
Rowe,
d'Arge,
and
Brookshire
[
1980]
found
strong
evidence
of
the
effects
of
starting
points,
with
a
respondent's
bid
for
improvements
in
visibility
increasing
by
$
0.60
for
every
$
1.00
increase
in
the
starting
point.

Brookshire
et
al.
[
1979]
also
found
starting
point
bias
in
some
of
their
alternative
bidding
situations.
However,
their
starting
point
bias
tests
are
difficult
to
interpret
because
their
study
had
very
small
sample
sizes
across
the
alternative
starting
points,
ranging
from
2
to
16
respondents.
Combined
with
the
substantial
standard
deviation
for
the
mean
responses,
these
small
sample
sizes
make
it
difficult
to
reject
the
null
hypothesis
that
starting
point
has
no
effect.
Mitchell
and
Carson
[
1981
]
argue
that
the
small
sample
size
may
have
had
a
greater
impact
on
the
study's
inability
to
detect
starting
point
bias
in
the
Brookshire
et
al.
[
1979]
study
than
the
researchers
realized.
In
addition,
Mitchell
and
Carson
[
unpublished
1982]
have
also
suggested
that
the
Greenley,
Walsh,
and
Young
[
1981
]
study
was
also
hindered
by
starting
point
bias.
The
payment
vehicles
chosen
by
Greenley,
Walsh,
and
Young
inadvertently
set
two
different
starting
points
for
the
bidding
process.

4
­
6
several
other
studies
­­
including
those
by
Brookshire
and
Randall
[
1978],

and
5Ch
Ulze
[
1977],
Randall
et
al.
[
1978]
,
and
Thayer
[
1981
]­­
have
Thayer
~
l~
o
tested
for
startin9
Point
bias
in
various
degrees.
These
studies
found
of
influence
on
willingness
to
PaY
that
could
be
attributed
to
dif­
"
0
evidence
starting
points.
Unfortunately,
the
research
design
of
some
of
these
ferent
inadequate
to
sufficiently
test
for
starting
point
bias.
~
tudieS
was
The
Randall
study
was
not
aple
tO
differentiate
mean
bids
by
starting
points,
and
other
studies
tested
starting
points
whose
relative
amounts
were
several
of
`
he
~
W
close
to
provide
conclusive
resuits.

in
summary,
the
literature
on
starting
point
bias
indicates
that,
when
a
bidding
9ame
is
used
to
elicit
willingness
to
pay,
the
results
can
be
influenced
by
the
starting
point
used
in
the
bidding
process,
suggesting
that
tests
for
bias
should
be
included
in
the
research
design.
startin9
Point
[
1981]
study
provides
The
Thayer
both
a
simple
test
for
the
existence
of
starting
point
bias
a
n
d
a
n
adjustment
for
willingness­
to­
pay
bids
if
starting
point
bias
exi
StS.
However,
the
assumptions
implicit
in
Thayer's
test
may
limit
its
practical
application,
since
it
assumes
the
respondent
has
a
nonstochastic
honest
bid.

4.2.5
Information
Bias
Information
bias
is
the
influence
on
an
individual's
valuation
that
is
attributable
to
the
amount
of
information
given
to
respondents
in
the
survey
questionnaire.
The
literature
provides
very
little
evidence
on
the
extent
of
information
bias.
Careful
questionnaire
design
and
thorough
interviewer
training
to
provide
consistent
and
equal
information
to
each
respondent
should
minimize
this
bias.
*

4.2.6
Interviewer
Bias
Interviewer
bias
is
attributable
to
the
effect
of
using
different
interviewers
to
elicit
individuals'
valuations.
This
bias
can
stem
from
one
interviewer
being
more
effective
than
another,
either
in
administering
a
bidding
game
or
in
establishing
rapport
with
the
respondent.
In
his
seminal
research
on
wilderness
experiments
in
the
Maine
woods,
Davis
[
1963]
established
a
high
level
of
rapport
with
the
respondents
but
performed
all
of
the
interviews
himself
A
recent
study
by
Cronin
[
1982]
was
able
to
test
for
the
existence
of
interviewer
bias
and
indicates
that
willingness
to
pay
can
be
influenced
by
the
interviewer.
But
the
design
of
the
test
was
not
sufficiently
robust
for
a
conclusive
result.
The
prospects
for
interviewer
bias
can
be
minimized
with
training
sessions
and
by
using
experienced
professional
interviewers.
Nonetheless
even
when
training
is
used,
the
research
should
examine
the
influence
of
using
different
interviewers
because
this
may
serve
to
identify
other
influences
on
the
bids
that
were
not
previously
recognized.

*
This
is
an
from
the
problems
­.
example
of
a
bias
category
that
is
not
easily
distinguished
associated
with
"
framing"
the
experiment.

4­
7
Table
4­
1.
Summary
of
Biases
in
Contingent
Valuation
Experiments
Studies
that
have
Summary
of
Type
of
bias
Definition
tested
for
bias
c
u
r
r
e
n
t
resultsa
General
Hypothetical
Error
introduced
by
posing
hypothetical
One
known
test­­
conditions
rather
than
actual
condi
­
Bishop­
He
berlein
tions
to
an
individual;
response
may
[
1979],
Bohm
not
be
a
good
guide
to
actual
actions
[
1971]
individual
would
take
Some
indication
that
hypothetical
nature
of
question
did
influence
responses,
but
could
not
distinguish
this
effect
from
instrumentrelated
biases
Strategic
Attempt
by
respondents
to
influence
out­
At
least
eight
tests
come
of
study
by
systematically
over­
(
see
Schulze,
or
under­
bidding
so
action
favors
d'Arge,
and
their
true
interests;
strategic
Brookshire
[
1981]
resDonses
deDend
on
how
Davment
scheme
f
o
r
summarv:
Very
little
evidence
of
strategic
bias
except
for
Cronin
[
1982]

is
defined
an"
d
whether
it
is
`
believed
Cronin
[
1
9
8
2
]
)

Instrument
related
Starting
point
Some
differences
in
opinion
over
importance
of
starting
point
bias;
Mitchell­
Carson
[
1981]
feel
starting
point
bias
is
important,
and
Desvousges,
Smith,
and
McGivney
[
1982]
provide
some
support;
Schulze,
d'Arge,
and
Brookshire
[
1981]
feel
it
is
more
limited
Contingent
valuation
experiments
using
bidding
game
format
have
started
with
suggested
payment
and
use
yes
or
no
responses
to
derive
final
willingness
to
pay;
suggestion
may
be
perceived
as
appropriate
bid
At
least
five
tests
(
see
Schulze,
d'Arge,
and
Brookshire
[
1981]
and
Rowe
and
Chestnut
[
1981])

Vehicle
Characteristics
of
proposed
mechanism
for
obtaining
respondent's
willingness
to
pay
may
influence
responses
At
least
four
tests
(
see
Schulze,
d'Arge,
and
Brookshire
[
1981]
and
Mitchell
and
Carson
[
1981])
Some
evidence
of
effects
in
at
least
two
studies
Limited
evidence
of
effects
Information
Effect
of
information
provided
to
respondent
on
costs
of
action
under
stu,
dy
or
other
dimensions
of
problem
may
affect
responses
At
least
four
tests
(
see
Schulze,
d'Arge,
and
Brookshire
[
1981]
and
Mitchell
and
Carson
[
1981])

interviewer
Responses
vary
systematically
according
to
interviewer
Two
tests­­
Desvousges,
Smith,
No
evidence
of
bias
and
McGivney
[
1982]
and
Cronin
[
1
9
8
2
]
)
Bias
present
a
The
definitions
and
results
summarized
in
this
table
are
based
on
Schulze,
d'Arge,
and
Brookshire
[
lgsIl,
Rowe
and
Chestnut
[
1981],
and
Mitchell
and
Carson
[
1981].

4
­
8
~.
2.7
summarY
and
Implications
for
Contingent
Valuation
Research
Design
Table
4­
1
summarizes
the
relevant
research
on
potential
biases
in
contingent
valuation
studies
discussed
above.
Based
on
this
information,
the
MonOngahela
River
contingent
valuation
study
was
designed
to
test
for
starting
point
bias.
In
addition,
after
the
surveys
were
completed,
the
statistical
~
nalYsis
examined
the
prospects
for
interviewer
bias.
The
structure
of
the
survey
attempted
to
control
for
information,
vehicle,
hypothetical
and
strategic
biaseS
in
the
Survey
questionnaire.

4.3
QUEST
l~
NAIRE
DESIGN
questionnaire
design
is
the
most
critical
task
in
a
contingent
valuation
study.
This
section
describes
the
questionnaire
used
to
estimate
the
recreation
and
related
benefits
of
water
quality
improvements
for
the
Monongahela
River
in
Pennsylvania.
Specifically,
building
on
the
sampling
plan
and
survey
procedures
discussion
in
Chapter
3
and
on
the
contingent
valuation
survey
biases
discussion
in
Section
4.2,
this
section
explains
the
treatment
of
potential
biases
either
as
an
element
in
the
questionnaire
design
or
as
an
objective
in
the
analysis
of
the
resulting
data.

4.3.1
Questionnaire
Design:
Part
A
A
key
ingredient
in
successful
contingent
valuation
surveys
is
establishing
credibility
for
the
survey
objectives
(
see
Appendix
D
for
a
complete
copy
of
the
questionnaire).
The
first
component
of
the
questionnaire
has
to
achieve
this
objective
without
biasing
or
offending
the
respondent.
Part
A
in
the
Monongahela
River
questionnaire
attempted
to
achieve
these
goals
by
inquiring
about
recreation
activities
the
respondent
had
engaged
in
during
the
last
year.
The
first
two
questions
dealt
with
boat
ownership
to
determine
if
the
respondent
had
easy
access
to
a
boat
for
recreation
purposes
through
either
ownership
or
"
borrowing"
rights.
Ditton
and
Goodale
[
1973]
found
boat
ownership
to
be
a
significant
factor
in
recreation
attitudes
and
activities
in
Green
Bay,
Wisconsin.
This
suggested
a
question
that
was
unlikely
to
offend
any
respondent.

Following
the
boat
ownership
question,
the
interviewer
prese,
lted
the
list
of
outdoor
recreation
activities
shown
in
Figure
4­
1
and
asked
if
the
respondent
had
participated
in
any
of
the
activities
within
the
past
12
months.
The
list
contains
a
wide
range
of
activities,
including
those
usually
associated
with
water
recreation
­­
boating,
fishing,
and
swimming­­
and
those
that
occur
near
water­
­
picnicking,
biking,
and
sightseeing.
The
list
is
a
subset
of
the
activities
listed
in
the
1977
Federal
Estate
Survey
data
base
used
in
estimating
the
travel
cost
model
in
Chapter
7.
This
activity
matching
was
an
attempt
to
provide
additional
compatibility
between
the
methods.

A
"
no"
answer
to
the
participation
question
on
the
Monongahela
questicn
­
naire
moved
`
the
respondent
into
`
the
benefits
section,
while
"­
"
­

initiated
used
the
and
the
provided
the
site/
activity
matrix,
illustrated
in
Figure
4­
2.
site/
activity
matrix
to
record
the
sites
visited,
the
activities
in
which
the
respondent
participated.
the
respondent
with
two
additional
visual
aids
4
­
9
a
"
yes"
response
The
interviewer
number
of
visits,
The
interviewer
to
facilitate
this
{
01
Canoeing,
kayaking,
or
river
mnning
02
Other
boating
On
or
In
03
Sailing
water
04
Water
skiing
05
Fishing
05
Swimming
outdoors
or
sunbathing
07
!
Camping
in
a
developed
area
08
Picnicking
09
Walking
to
observe
natura
or
bird
watching;
wildlifa
or
bird
photography
10
Other
walking
for
pleasure
or
jogging
11
Bicycling
12
Horsaback
riding
Near
13
Hunting
Water
14
Hiking
or
backpacking
15
Attending
outdoor
sports
avents
(
do
not
include
professional
football
or
baseball)
16
Other
outdoor
sports
or
games
17
Driving
vehicles
or
motorcycle
off­
mad
16
Driving
for
pleasure
19
Sightsming
at
historical
sitea
or
natural
wonders
Figure
4­
1.
Activity
card.

2
g
Q
w
9
9
;
~
!
2
n
~
$

g
z
q
2
$
g
!
j
g
k
?
:
$
~

$
:
9
t!
$
s
2
i!
$
"
~
$
g
g
`
3
~
i
~
;
3
:
$
g
9
g
g
~
!
g
:
~
:
~
2
!!
~
w~
iiit
2
:
i
!!
$
:
ii
$
i
!!
ii
!
3
5
:
!
i!
$
L
:
.

Site
Names
Not
Listed
01
02
03
06
05
06
07
08
09
10
11
12
13
I&
Is
16
17
18
19
01
02
03
04
05
06
07
08
09
10
11
12
13
16
15
16
17
18
19
01
02
03
04
05
06
07
08
09
10
11
12
13
14
1s
16
L7
18
19
01
02
03
04
05
06
07
08
09
10
11
12
13
14
15
16
17
1.9
19
01
02
03
06
05
06
07
08
09
10
11
12
13
16
15
16
17
1.9
19
01
02
03
o&
0s
06
07
08
09
10
11
12
13
III
15
16
17
18
19
01
02
03
04
05
06
07
08
09
10
11
12
13
16
1s
16
17
18
19
01
02
03
Ob
0s
06
07
08
09
10
11
12
13
lk
15
16
17
la
19
01
02
03
04
05
06
07
08
09
10
11
12
13
14
1s
16
17
18
19
01
02
03
04
05
06
07
08
09
10
11
12
13
14
15
16
17
18
)
9
01
02
03
06
0s
06
07
08
09
10
11
12
13
16
IS
16
17
18
19
01
02
03
04
0s
06
07
08
09
10
11
12
13
16
1s
16
17
18
19
01
02
03
04
05
06
07
0s
09
10
11
12
13
14
1s
16
17
18
19
01
02
03
04
05
06
07
08
09
10
II
12
13
16
1s
16
17
18
19
01
02
03
04
05
06
07
08
09
10
11
12
13
14
1s
16
11
la
19
01
02
03
04
05
06
07
08
o~
10
11
12
13
14
1S
16
17
18
19
Figure
4­
2.
Site
activity
matrix.

4­
1o
~
iScuSSiOn
­­
a
colored
pictorial
map
of
the
area
shown
in
Figure
4­
3
and
a
list
sites
(
aiso
shown
on
the
map)
displayed
in
Figure
4­
4.
The
re­
~
f
recreation
described
the
information
requested
for
these
sites
or
any
other
sites
~
pondent
The
data
collected
in
part
A
completed
a
recreation
profile
of
the
visited.
could
be
used
in
the
analysis
phase
and
established
a
rapport
respondent
that
with
him
without
itIflUenCinCJ
the
main
objective­­
benefit
estimation.
Part
A
the
idea
that
a
wide
range
of
recreation
site
services
is
influ
­
~
lso
reinforced
enced
by
water
quality.

4.3.2
Benefits
Measures:
Part
B
Part
B
of
tpe
Mo~
ongahela
River
questionnaire
established
the
hypothetical
market
by
describing
Its
institutional
arrangements.
In
other
words,
this
part
described
the
hypothetical
market,
the
commodity
to
be
valued,
the
payment
vehicle,
and
enacted
the
valuation
experiment.
The
first
section
introduced
the
setting
for
the
hypothetical
market:

The
next
group
of
questions
is
about
the
quality
of
water
in
the
Monongahela
R
i
v
e
r
.
Congress
passed
water
pollution
control
laws
in
1972
and
in
1977
to
improve
the
nation's
water
quality.
The
States
of
Pennsylvania
and
West
Virginia
have
also
been
involved
in
water
quality
improvement
programs
of
their
own.
These
programs
have
resulted
in
cleaner
rivers
that
are
better
places
for
fishing,
boating,
and
other
outdoor
activities
which
people
take
part
in
near
water.
~
all
pay
for
these
water
quality
improvement
programs
both
as
taxpayers
and
as
consumers.

In
this
study
we
are
concerned
with
the
water
quality
of
only
the
Monongahela
R
i
v
e
r
.
Keep
in
mind
that
people
take
part
in
all
of
the
activities
on
Card
1
(
Figure
4­
1
)
both
on
and
near
the
water.

Following
the
introduction,
the
interviewer
handed
the
respondent
the
key
visual
aid
for
the
hypothetical
market­­
the
Resources
for
the
Future
(
RFF)
water
quality
ladder
developed
by
Mitchell
and
Vaughan
at
RFF
and
used
by
Mitchell
and
Carson
[
1981
]
in
their
contingent
valuation
study
of
national
water
quality
(
see
Figure
4­
5).
Appendix
E
provides
details
on
its
construction.
The
ladder's
major
attribute
is
that
it
easily
establishes
linkages
between
recreation
activities
and
water
quality
based
on
an
index
of
technical
water
quality
measures
and
informed
judgment.
This
type
of
linkage
illustrates
a
crucial
distinction
between
the
contingent
valuation
method
and
indirect
techniques
for
measuring
the
benefits
of
water
quality.
Specifically,
rather
than
observing
the
actual
behavior
of
recreational
ists,
who
demand
different
site
services
depending
on
the
level
of
water
quality,
it
directly
introduces
the
relationship
between
activities
and
different
water
quality
levels
into
the
hypothetical
market.

After
showing
the
key
visual
aid,
the
interviewer
read
the
following
text*
to
describe
the
ladder
and
establish
the
desired
linkages:

*
The
words
in
all
capitals
are
instructions
for
the
interviewers
only
were
not
read
to
the
respondent.
They
are
included
in
the
discussion
completeness.
and
for
.

4­
11
.
I
 
 
 
.
.
.
 
:
{
)>'
_______

1­
­­­­_\\,,/'
wntunmNqhuia
,
I
`
4*­

Figure
4­
3.
Map
of
Monongaheia
River
and
other
retreat
ion
sites.

I
Alla@
amy
Rivu:
Monongahala
River
Area:

01
02
03
04
05
06
07
06
OB
10
11
12
13
14
Near
Kittannirrg
Near
Oakmont
Whara
Beaver
River
and
Ohio
River
meat
Crookad
Creak
Park
Loyalhanna
Lake
Kayatone
Dam
Laka
Arthur
in
Moraina
State
Park
Ohiopyle
Stata
Park
North
Park
Lake
(
Near
Allison
Park)
Ramon
Creak
Steta
Park
Youghioghany
River
Lake
Raaarvoir
Cheat
River
Laka
Ryarson
Station
Yellow
Creek
15
16
17
18
19
20
21
22
23
24
25
26
27
28
28
Pittsbur@
(
The
Point,
Smithfield
Brid~,
Braddock)
Where
Monongahela
and
Youghio#
mny
meat
naar
McKeespon
Elrema
The
Town
of
Monongehela
Donors
and
Wbatar
Near
Charlaroi
(
Lock
and
Dam
#
4)
In
tha
California­
Brownwi
lle
Area
Maxwell
Lock
and
Dam
Ten
Mile
Creak
Grays
Landing­
Graanaboro
(
Lock
and
Dam
#
7)

Point
Marion­
cheat
River
Area
(
Lock
and
Dam
#
B)
Morgantown
Hildabrand
Opekiska
Fairmont
Figure
4­
4,
Recreation
sites.

4­
12
I
f
t­­'­
l
H
3
OD
OKAY
FOR
BOATING
E
z
I
OE
 
q
 
!
W!
WZIi%
l
Figure
4­
5.
Water
quality
ladder,

Generally,
the
better
the
water
quality,
the
better
suited
the
water
is
for
recreation
activities
and
the
more
Ii
kely
people
will
take
part
in
outdoor.
recreation
activities
on
or
near
the
water.
Here
is
a
picture
of
a
ladder
that
shows
various
levels
of
water
quality.
GIVE
RESPONDENT
CARD
4,
"
WATER
QUALITY
LADDER.
"

The
top
of
the
ladder
stands
for
the
possible
quality
of
water.
The
bottom
of
the
ladder
stands
for
the
worst
possible
water
quali
t
y
.
On
the
ladder
you
can
see
the
different
levels
of
the
quality
of
the
water.
For
example:
(
POINT
TO
EACH
LEVEL­­
E,
D,
C,
B,
A­­
AS
YOU
READ
THE
STATEMENTS
BELOW.
)

Level
E
(
POINTING)
is
so
polluted
that
it
has
oil,
raw
sewage
and
other
things
~
ke
trash
in
it;
it
has
no
plant
or
animal
life
and
smells
bad.

Water
at
Level
D
is
okay
for
boating
but
not
fishing
or
swimming.

4­
13
Level
C
shows
where
the
water
is
clean
enough
so
that
gamefish
like
bass
can
live
in
it.

Level
B
shows
where
the
water
is
clean
enough
so
that
people
can
swim
in
it
safely.

And
at
Level
A,
the
quality
of
the
water
is
so
good
that
it
would
be
possible
to
drink
directly
fro~
it
if
you
wanted
to.

Following
this
description,
the
interviewer
asked
the
respondent
to
use
the
ladder
to
rate
the
water
quality
in
the
Monongahela
River
on
a
scale
of
O
to
10
and
to
indicate
whether
the
ranking
was
for
a
particular
site,
and,
if
so,
to
name
it.

Question
B­
2
introduced
the
respondent
to
a
key
element
in
the
hypothetical
market:
the
distinction
between
user,
option,
and
existence
values.
Specifically,
the
interviewer
gave
the
respondent
the
value
card
shown
in
Figure
4­
6
and
described
each
type
of
value.
An
attitudinal
question
punctuated
the
descriptions
of
each
type
of
value
by
inquiring
how
important
the
factors
of
actual
use,
potential
use,
and
no
use
were
in
valuing
water
quality.
The
attitudinal
responses
to
these
question
s­­
displayed
on
a
five­
point
scale
ranging
from
very
important
to
not
important
at
all
­­
reinforced
the
concepts,
provided
a
break
in
the
discussion,
and
presented
an
additional
check
for
the
consistency
in
responses.
The
textual
explanations
for
the
three
types
of
values
are:

Why
We
Might
Value
Clean
Water
in
the
Monongahela
River
1.
use
Swimming
Hiking
Fishing
Sitting
by
the
shore
Boating
Hunting
Picnicking
Driving
vehicles
off
road
Birdwatching
Jogging
Il.
Might
Use
To
have
clean
water
in
the
river
to
use
if
you
should
decide
in
the
future
that
you
want
to
use
it.

II
1.
Just
Because
It's
There
Preserve
for
future
generations.
Satisfaction
from
knowing
that
there
is
a
clean
river.
Satisfaction
from
knowing
that
others
can
enjoy
the
river
for
recreation.

Figure
4­
6.
Value
card.

4­
14
Another
important
PUt'pOSe
Of
this
study
iS
to
learn
how
much
the
quality
of
water
of
the
Monongahela
River
is
worth
to
the
people
who
live
in
the
river
basin.
In
answering
this
question,
there
are
three
ways
of
thinking
about
water
quality
that
might
influence
your
decision.
GIVE
RESPONDENT
CARD
5,
"
VALUE
CAR
D."
The
three
Ways
are
shown
on
this
card.

~,
YOU
might
think
about
how
much
water
quality
is
worth
to
you
because
You
use
the
river
for
recreation.
POINT
TO
PART
I
OF
VALUE
CARD
AND
GIVE
RESPONDENT
TIME
TO
READ
THAT
PART.

HOW
important
a
factor
is
your
ZICtUa!
use
of
the
river
in
making
a
decisiofl
about
how
much
clean
water
is
worth
to
you?
CIRCLE
NuMBER.

VERY
IMPORTANT.
.
.
.
.
.
.
.
.
.
01
SOMEWHAT
IMPORTANT
.
.
.
.
.
.
.
02
NEITHER
IMPORTANT
NOR
UNIMPORTANT.
.
.
.
.
.
.
.
.
.
.
03
NOT
VERY
IMPORTANT
.
.
.
.
.
.
.
04
NOT
IMPORTANT
AT
ALL
.
.
.
.
.
.
05
Another
way
you
might
think
about
how
much
clean
water
is
worth
to
you
is
that
it
is
worth
something
to
you
to
know
that
a
clean
water
river
is
being
maintained
for
your
use
if
you
should
decide,
in
the
future,
that
you
want
to
use
it.
POINT
TO
PART
II
OF
VALUE
CARD
AND
GIVE
RESPONDENT
TIME
TO
READ
THAT
PART.
For
example,
you
might
buy
an
advance
ticket
for
the
Steelers
or
Pirates
just
to
be
able
to
go
to
a
home
game
if
you
later
decide
you
want
to
go.
Likewise,
you
might
pay
some
amount
each
year
to
have
a
clean
water
river
available
to
use
if
you
should
decide
to
use
it.

In
deciding
how
much
clean
water
is
worth
to
you,
how
important
a
factor
is
knowing
that
a
clean
water
river
is
being
maintained
for
your
use,
if
you
should
decide
to
use
it?
CIRCLE
NUMBER.

VERY
IMPORTANT.
.
.
.
.
.
.
.
.
.
01
SOMEWHAT
IMPORTANT
.
.
.
.
.
.
.
02
NEITHER
IMPORTANT
NOR
UNIMPORTANT.
.
.
.
.
.
.
.
.
.
.
03
NOT
VERY
IMPORTANT
.
.
.
.
.
.
.
04
NOT
IMPORTANT
AT
ALL
.
.
.
.
.
.
05
4"
15
A
third
thing
you
might
think
about
in
deciding
how
much
water
is
worth
to
you
is
the
satisfaction
of
knowina
that
a
clean
clean
water
river
is
there.
POINT
TO
PART
Ill
OF
VALtiE
CARD
AND
GIVE
RESPONDENT
TIME
TO
READ
THAT
PART.
For
example,
you
might
be
willing
to
pay
something
to
maintain
a
public
park
even
though
you
know
you
won't
use
it.
The
same
thing
could
be
true
for
clean
water
in
the
Monongahela;
that
is,
you
might
pay
something
just
for
the
satisfaction
of
knowing
that
it
is
clean
and
that
others
can
use
it.

In
deciding
how
much
clean
water
is
knowing
that
a
clean
water
river
NUMBER.

VERY
IMPORTANT.
.
.
.
.

SOMEWHAT
IMPORTANT
.
.

NEITHER
IMPORTANT
NOR
UNIMPORTANT.
.
.
.
.
.

NOT
VERY
IMPORTANT
.
.

NOT
IMPORTANT
AT
ALL
.

The
first
paragraph
of
Question
B­
3,
the
respondent,
is
presented
below:
.
.

.
.

.
.

.
.

.
.
worth
to
is
being
you,
how
important
is
maintained?
CIRCLE
.
.
.
01
.
.
.
02
.
.
.
03
.
.
.
04
.
.
.
05
which
introduces
the
payment
vehicle
to
Now,
we
would
like
you
to
think
about
the
relationship
between
improving
the
quality
of
water
in
the
Monongahela
River
and
what
we
all
have
to
pay
each
year
as
taxpayers
and
as
consumers.
We
all
pay
directly
through
our
tax
dollars
eacfiear
for
cleaning
up
all
rivers.
We
also
pay
indirectly
each
year
through
higher
prices
for
the
products
we
buy
because
it
costs
companies
money
to
clean
up
water
they
use
in
making
their
products.
Thus,
each
year,
we
are
paying
directly
and
indirectly
for
improvements
in
the
water
quality
of
the
Monongahela
River.

I
want
to
ask
you
a
few
questions
about
what
amount
of
money
you
would
be
willing
to
pay
each
year
for
different
levels
of
water
quality
in
the
Monongahela
River.
Please
keep
in
mind
that
the
amounts
you
would
pay
each
year
would
be
paid
in
the
form
of
taxes
or
in
the
form
of
higher
prices
for
the
products
that
companies
sell.

This
payment
vehicle
was
selected
because
it
corresponds
with
how
people
actually
pay
for
water
quality,
connotes.
no
implicit
starting
point,
and
provides
a
vehicle
that
will
bias
the
responses
downward,
if
in
any
direction,
because
of
public
attitudes
toward
increased
taxes
and
higher
prices.

The
introduction
continues
with
a
reference
to
the
value
card
(
see
Figure
4­
6)
and
requests
that
initial
amounts
be
based
on
actual
use
and
potential
future
use
­­
user
and
option
values
but
not
existence
values.
The
present
overall
level
of
water
quality
is
described
as
Level
D,
where
it
is
clean
enough
for
boating.

4­
16
Question
B­
3
embodies
the
comparison
of
the
alternative
contingent
valuation
methodologies.
Specifically,
by
dividing
the
sample
of
397
households
into
fourths
and
using
a
diffe~
ent
color
survey
instrument
for
each
quarter,
Quest
i
o
n
B­
3
compares
the
direct
question
method
of
eliciting
willingness­
to­
pay
amounts,
both
with
and
without
a
payment
card
(
illustrated
in
Figure
4­
7),
to
the
iterative
bidding
games
with
$
25
and
$
125
starting
points.
Thus,
the
questionnaire
design
provides
an
explicit
test
for
starting
point
bias
within
the
iterative
bidding
game,
as
well
as
a
test
for
differences
between
direct
questions
and
bidding
games.

o
100
200
300
400
500
600
700
25
125
225
325
425
525
625
725
50
150
250
350
450
550
650
750
75
175
275
375
475
575
675
775
Figure
4­
7.
Payment
card.

The
payment
card
used
in
the
direct
question
method
was
simply
an
array
of
numbers
representing
annual
amounts
from
$
0
to
$
775
per
year.
This
is
in
contrast
with
the
Mitchell
and
Carson
[
1981
]
payment
card,
which
showed
amounts
individuals
paid
for
various
public
goods
adjusted
to
correspond
with
the
respondent's
income
level.
Mitchell
and
Carson
split
their
sample
to
test
for
the
effect
of
the
different
types
of
public
goods
provided,
but
the
sample
size
in
the
Monongahela
study
was
much
smaller
and
already
partitioned
into
four
groups,
so
no
anchoring
amounts
were
listed
on
the
payment
card.
Mitchell
and
Carson
found
no
effect
from
the
anchoring
amounts,
but
this
result
may
have
been
hampered
by
their
adjustment
of
the
amounts
to
correspond
to
the
respondent's
income
level.

The
hypothetical
market
queried
the
respondent
for
willingness­
to­
pay
amounts
for
three
water
quality
levels:

.
Avoiding
a
decrease
in
water
quality
in
the
Monongahela
River
from
D,
boatable,
to
E,
not
suitable
even
for
boating.

l
Raising
the
water
quality
from
D,
beatable,
to
C,
where
gamefish
could
survive.

.
Raising
the
water
quality
from
C,
fishable,
to
B,
where
people
could
swim
in
the
water.
.
Table
4­
2
summarizes
the
formats
for
eliciting
the
option
prices
in
the
contingent
valuation
questionnaire.
(
For
details
on
question
procedures,
see
Appendix
D,
which
contains
a
complete
copy
of
the
survey
questionnaire.
)

4­
17
Table
4­
2.
Summary
of
Option
Price
Question
Formats
by
Interview
Type
[
Interview
type
Question
format
~.

Iterative
bidding,
$
25
To
you
(
and
your
family),
would
it
be
worth
$
25
each
year
in
higher
taxes
and
prices
for
products
that
companies
sell
to
keep
the
water
quality
in
t
h
e
Monongahela
River
from
slipping
back
from
Level
D
to
Level
E?

Iterative
bidding,
$
125
To
you
(
and
your
family),
would
it
be
worth
$
125
each
year
in
higher
taxes
and
prices
for
products
that
companies
sell
to
keep
the
water
quality
in
the
Monongahela
River
from
slipping
back
from
Level
D
to
Level
E?

What
is
the
most
it
is
worth
to
you
(
and
your
family)
on
a
yearly
basis
to
keep
the
water
quality
in
the
Monongahela
River
from
slipping
back
from
Level
D
to
Level
E,
where
it
is
not
even
clean
enough
for
boating?
Direct
question
Payment
card
What
is
the
most
it
is
worth
to
you
(
and
your
family)
on
a
yearly
basis
to
keep
the
water
quality
in
the
Monongahela
River
from
slipping
back
from
Level
D
to
Level
E,
where
it
is
not
even
clean
enough
for
boating?

The
process
for
the
direct
question
is
very
simple,
with
the
interviewer
asking
the
respondent
for
an
amount
for
each
level
and
stressing
that
additional
amounts
are
being
requested.
The
water
quality
ladder
and
the
value
card
are
in
front
of
the
respondent
while
the
market
process
is
initiated.
The
same
procedure
was
used
in
the
payment
card
format,
with
the
only
difference
being
that
the
payment
card
was
given
to
the
respondent.

Table
4­
2
also
summarizes
the
procedure
for
the
bidding
games
with
starting
points.
A
similar
procedure
was
used
for
both
bidding
games,
the
only
difference
being
the
starting
points
used.
In
the
bidding
game,
the
interviewer
initiated
the
market
process
at
the
starting
point
and
increased
or
de­
~
creased
the
requested
amount
until
the
respondent's
maximum
value
was
ob­
,

tained.
This
was
repeated
for
each
of
the
water
quality
levels,
with
emphasis
;
given
to
the
additional
nature
of
the
amounts
for
the
higher
levels
of
water
/
quality.
?

To
conclude
this
part
of
the
hypothetical
market,
the
interviewer
asked
any
respondent
who
gave
a
zero
amount
why
that
amount
was
given,
as
shown
in
the
question
below.
The
purpose
of
this
question
was
to
distinguish
between
a
true
zero
amount
and
a
zero
that
essentially
represented
a
protest
against
either
the
experiment
or
some
part
of
it,
such
as
the
p
a
y
m
e
n
t
vehicle.

T
""­
4­
18
L
.
.
.
.
.
.
.
.
.
.
.
..
 
.
_._
.
_________
,____
we
have
found
in
studies
of
this
type
that
people
have
a
lot
of
different
reasons
for
answering
as
they
do.
Some
people
felt
they
did
not
have
enough
information
to
give
a
dollar
amount,
some
did
not
want
to
put
dollar
values
on
environmental
quality,
and
some
objected
to
the
way
the
question
was
presented.
Others
gave
a
zero
dollar
amount
because
that
was
what
if
was
worth
to
them.

which
of
these
reasons
best
describes
why
you
answered
the
way
you
did?
REPEAT
REASONS
IF
NECESSARY
AND
CIRCLE
NUMBER.

NOT
ENOUGH
INFORMATION.
.
.
.
.
.
.
.
.
.
.
.
01
DID
NOT
WANT
TO
PLACE
DOLLAR
VALUE
.
.
.
.
0
2
OBJECTED
TO
WAY
QUESTION
WAS
PRESENTED.
.
03
THAT
IS
WHAT
IT
IS
WORTH
.
.
.
.
.
.
.
.
.
.
.
04
O
T
H
E
R
(
S
P
E
C
I
F
Y
)
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
0
5
The
next
section
of
the
questionnaire
attempted
to
break
down
the
option
price
into
its
individual
components
of
user
and
option
values.
The
questions
and
results
for
option
value
are
described
in
detail
in
the
following
chapter,

so
no
additional
discussion
is
provided
in
this
chapter.

Part
B
contained
two
additional
plausibility/
consistency
check
questions
that
asked
what
effect
improved
water
quality
in
the
Monongahela
River
would
have
on
visits
to
substitute
sites
and
the
Monongahela
River
sites.
swers
to
these
questions
were
structured
by
choices
ranging
from
(
either
increase
or
decrease)
of
more
than
five
visits
to
no
change
know."*

The
last
auestion
in
Part
B
asked
the
respondent
to
perform
The
ana
change
or
"
don't
a
contin
­
gent
ranking
as
specified
by
the
text
from
the
questionnaire.
Figure
4­
8
depicts
one
of
the
four
combinations
that
the
respondent
was
asked
to
rank.
This
particular
card
shows
the
combination
of
the
lowest
level
of
water
quality
and
the
lowest
payment.
Payment
amounts
of
$
50,
$
100,
and
$
175
were
paired
with
boatable,
fishable,
and
swimmable
levels
of
water
quality,
respectively.
The
survey
design
asked
all
respondents
to
rank
the
cards
after
participating
in
one
of
the
other
valuation
exercises.
This
design
is
a
compromise
resulting
from
the
limited
resources
available
for
sampling
respondents
and
the
objective
to
compare
as
many
methods
as
possible.
A
complete
comparison
would
have
required
an
additional
segmentation
of
the
limited
sample.
Chapter
6
discusses
the
theory
and
results
from
the
contingent
ranking
experiment.

*
These
questions
(
rMB)
in
its
review
of
were
suggested
by
the
Office
of
Management
and
Budget
the
survey
questionnaire.

4­
19
WATER
QUALITY
LADDER
I­
8­
I
t­
g­
i
k%
iiiil
Figure
4­
8.
Rank
order
card.

4.4
PROFILES
OF
SURVEY
RESPONDENTS
Respondents
in
a
contingent
valuatlon
survey
should
re~
resent
the
DoDu­
Iation
of
`
interest
to
provide
~
iausibie
results.
This
section
profiles
the
sample
respondents
from
the
Monongahela
River
basin
area
and
compares
these
profiles
with
Census
data
for
the
area
as
a
check
for
representativeness.
Users,
nonusers,
zero
bidders,
and
protest
bidders
are
also
profiled
to
assess
the
role
of
socioeconomic
and
attitudinal
characteristics
in
influencing
any
of
these
groups.

Tabie
4­
3
presents
the
characteristics
of
key
groups
of
respondents
in
the
Monongaheia
survey.
These
data
are
for
the
301
completed
questionnaires
that
provided
valid
responses.
Two
questionnaires
were
eliminated
because
the
respondents
were
unable
to
complete
the
session.
One
person
was
97
years
oid
and
had
difficulty
seeing
the
cards;
the
other
had
troubie
hearing
the
interviewer.

4­
20
I
 
.
_..
.­..
­.
 
.
.
.
.

Table
4­
3.
Charectechticsof
Key
Respondent
Groups
User
Nonuser
Zero
NOmoro
Protest
Bidsa
Total
Five
Standerd
Standard
Stmderd
Stenderd
Standard
Standard
devl
­
count
y
devl
­
davl­
devi
­
Cherecterlstic
devi
­
Ueen
devi
­
rag
ion
l
 
tion
N
Mean
etion
N
Meen
l
 
tkm
N
Mean
l
 
tion
N
Mean
l
 
tion
N
Mean
ation
N
in
1660
.%
mpie
I=
yea,
O=
no
for
ownership
or
use
of
l
 
boat
I=
yea,
O=
no
for
pwtlclpation
in
l
 
ny
outdoor
recreation
in
the
last
year
Numerlcel
rating
of
the
Wnongahela
River:
O=
loweet,
lo­
highest
l=
yea,
O­
no
If
rating
la
for
l
particular
slta
Lon@
h
of
residence
Yoam
of
education
Rue
(
1
If
whlta)

Age
sex
(
1
if
male)
0.23
0.43
64
0.12
0.32
207
0.11
0.32
106
0.18
0.3S
193
0.15
0.37
56
0.16
0.65
0.23
64
0.36
0.49
207
0.3S
0.49
106
O.
M
0.4S
193
0.50
0.50
56
0.
S6
3.87
1.9
63
3.77
2.01
132
3.51
1.76
61
3.
S2
2.07
180
3.63
1.68
3S
3.81
0.34
0.48
S4
0.06
0.27
207
0.07
0.28
108
0.21
0.41
193
0.10
0.31
56
0.16
6.83
O.%
94
6.
S0
1.02
207
6.62
0.6S
106
6.80
1.02
1S3
6.74
1.18
56
6.61
13.06
1.96
66
12.61
2.12
177
12.36
2.20
66
12.63
1.6S
177
12.77
1.73
47
12.75
0.
s6
0.32
64
0.91
0.2s
m
0.64
0.23
107
0.88
0.33
1s3
0.93
0.28
57
0.90
0.36
301
0.50
301
1.68
221
0.37
301
1.00
301
2.07
2S3
10.36b
12.75
0.30
300
.
Q
.30
20,83313,462
87
18,86713,022
173
17,577
11,500
67
20,53413,876
173
19,66511,464
46
19,
S36
13,164
2@
19,687b
19,536
36.93
16.20
94
51.67
17.85
207
54.55
16.91
106
44.06
10.07
193
52.80
17.27
S6
47.82
18.34
301
45.6
47.6
.31
.46
64
0.38
0.49
207
0.35
0.46
106
0.37
0.46
163
O.
u
0.50
58
0.
s
O.*
301
.47
.36
SOURCE:
U.
S.
BumaJ
of
th8
Cenaua.
lSM
Census
of
the
Population
snd
Mousing.
Washington,
D.
C.
1SS2.

l
 
PXt
bias
@
ro
zero
blda
for
raeuma
other
than
`
all
they
could
 
l
 
fford"
or
`
that
la
what
N
Is
worth.
"
bStatewlde
 
l
 
tatlatlca.

 
.
T
o
d
e
v
e
l
o
p
a
r
e
a
s
o
n
a
b
l
y
c
l
e
a
r
s
n
a
p
s
h
o
t
of
the
r
e
s
p
o
n
d
e
n
t
g
r
o
u
p
i
m
p
o
r
t
a
n
t
for
the
analysis
of
survey
results,
no
adjustments
for
outliers
are
included
in
the
profile
information.
The
first
two
columns
of
Table
4­
3
compare
users
and
nonusers
of
the
Monongahela
River.
The
users
are
broadly
defined
based
on
all
respondents
who
reported
a
user
value
or
visited
one
of
the
13
Monon
­
gahela
River
sites.
This
broader
definition
of
user
can
be
contrasted
with
a
narrow
definition
that
includes
only
those
respondents
who
visited
a
site.
The
broader
definition
is
used
throughout
this
report
because
it
allows
for
the
inclusion
of
some
users
who
may
have
been
prevented
from
visiting
a
Monongahela
site
within
the
12
months
between
November
1981
and
November
1982
for
medical
or
other
personal
reasons
but
still
had
some
user
value
for
the
services
of
the
Monongahela.
Tests
indicated
that
the
differences
between
the
user
definitions
were
insignificant.
This
broad
definition
explains
why
a
few
Monongahela
River
users
had
not
participated
in
an
outdoor
recreation
activity
in
the
second
row
of
Table
4­
3.

Results
of
t­
tests
for
differences
between
the
means
of
users
and
nonusers
(
shown
in
Appendix
C)
highlight
some
important
distinctions
that
continue
throughout
the
survey
results.
Users
of
the
Monongahela
River
are
younger,
are
more
likely
to
own
a
boat,
and
are
more
likely
to
have
rated
a
particular
Monongahela
River
site
than
their
nonuser
counterparts.
The
water
quality
ratings
place
the
Monongahela
above
beatable,
but
a
full
point
below
fishable,
on
the
Water
Quality
Ladder
(
see
Figure
4­
5);
however,
the
ratings
are
not
different
between
the
two
groups.
There
are
no
differences
in
education
income,
race,
sex,
or
length
of
residence
between
users
and
nonusers.
*

For
these
two
groups
t­
tests
for
differences
in
means
between
zero
and
nonzero
bidders
and
a
Iogit
analysis
comprise
the
analysis.
Based
on
these
results,
nonzero
bidders
were
on
average
younger
than
zero
bidders,
earned
higher
annual
family
incomes,
were
more
likely
to
have
rated
the
Monongahela
at
a
particular
site,
and
have
participated
in
outdoor
recreation
during
the
last
year.
These
results
are
consistent
with
the
findings
of
Mitchell
and
Carson
[
1981].
I
n
addition,
no
significant
differences
existed
between
the
groups
in
terms
of
sex,
education,
water
quality
rating
for
the
river,
boat
ownership,
and
length
of
residence
in
the
area.
The
protest
bidders
who
rejected
some
aspect
of
the
contingent
valuation
approach
had
higher
incomes
and
were
more
likely
to
have
participated
in
outdoor
recreation
in
the
last
year
than
were
t
h
o
s
e
w
i
t
h
valid
z
e
r
o
b
i
d
s
.

The
questionnaire
design
also
provided
the
respondent's
reason
for
giving
a
zero
bid.
These
responses
are
shown
in
Table
4­
4
for
the
four
elicitation
methods.
The
direct
question
method
without
the
payment
card
yielded
most
of
the
respondents
who
could
not
place
a
dollar
value
on
water
quality,

*
The
percentage
of
woman
respondents
(
64
percent)
in
the
sample
is
somewhat
higher
than
in
other
studies­­
a
somewhat
surprising
result
since
the
random
procedure
used
to
select
the
respondents
should
have
given
a
m
o
r
e
even
distribution.
The
respondent
was
asked
to
respond
for
the
household,
which
should
reduce
any
potential
bias.

4­
22
T
a
b
l
e
4
­
4
.
Reasons
for
Zero
Bids
by
Elicitation
Method
$
25
$
125
Payment
Direct
iterative
Reason
for
zero
bid
iterative
card
question
bidding
bidding
Total
 
Not
enough
information
1
4
0
1
9
0
0
2
1
2
0
0
4
15
1
Cannot
place
dollar
value
Objected
to
way
question
was
presented
That
is
what
it
is
worth
12
1
1
2
10
5
3
0
7
5
1
2
11
5
5
1
40
16
10
5
Other
All
they
could
afford
Government
waste
or
misuse
of
tax
dollars
Industry
pollutes
so
let
them
clean
it
up
3
2
3
0
8
Taxes
are
too
high
already
2
1
3
0
1
0
2
0
8
1
Desire
no
increase
in
taxes
for
something
that
does
not
affect
respondent
Total
27
33
22
26
108
which
roughly
indicates
the
value
of
either
the
payment
card
or
the
starting
value
in
the
bidding
process.
Approximately
40
percent
of
the
respondents
bid
zero
because
that
is
what
they
felt
the
water
quality
is
worth.
Some
evidence
of
the
consistency
in
the
response
is
indicated
by
the
10
respondents
who
bid
zero
because
that
is
all
they
could
afford.
These
respondents
tended
to
be
elderly
persons
living
on
limited
incomes.

Table
4­
5
shows
the
attitudinal
information
broken
down
for
user,
nonuser
and
zero
bids.
These
responses
on
the
importance
of
water
quality
were
elicited
during
the
discussion
of
the
value
card
(
see
Figure
4­
6)
and
prior
to
the
elicitation
of
the
willingness­
to­
pay
amounts.
These
responses
are
very
consistent
with
the
earlier
characteristics
of
the
groups.
Users
and
nonzero
bidders
were
much
more
likely
to
have
given
very
or
somewhat
important
responses
to
the
questions
than
were
nonusers
and
zero
bidders.
I
I
I
Table
4­
6
completes
the
profiles
of
the
three
groups
by
highlighting
the
respondents'
willingness
to
identify
themselves
by
certain
labels.
Several
interesting
features
are
apparent
from
these
attitudinal
responses.
The
users
and
nonzero
bidders
were
much
more
likely
to
identify
themselves
as
outdoors
persons
than
were
nonusers
and
zero
bidders.
However,
the
differences
between
the
groups
is
much
smaller
for
the
environmentalist
label,
with
26
per­

4­
23
Table
4­
5.
Degree
of
Importance
of
Water
Quality
by
Key
Respondent
Groups
Degree
of
importance
User
Nonuser
Zero
bids
Nonzero
bids
Protest
bidsa
Total
of
water
quality
Frequency
%
Frequency
%
Frequency
%
Frequency
%
Frequency
$
Frequency
%

For
For
own
recreation
Very
important
47
Somewhat
important
28
Neither
important
nor
unimportant
4
Not
very
important
10
Not
important
at
all
Total
9:

possible
future
use
23.3
20
17.5
14
16.0
21
22.3
25
20.9
28
18.5
13.0
19.4
23.1
25.9
75
50
16
31
20
39.1
16
26.0
9
18.3
14
16.1
11
10.4
8
27.6
15.5
24.1
19.0
13.8
95
64
37
56
48
31.7
21.3
12.3
18.7
16.0
50.0
29.7
4.3
10.6
5.3
48
36
33
46
43
300
119
87
31
36
28
301
123
99
28
33
17
300
206
70
53
26
33
25
207
74
70
21
27
15
207
108
192
92
66
13
13
9
193
96
66
13
11
6
192
58
39.5
28.9
10.3
12.0
9.3
41.0
3
3
.
0
9.3
11.0
Very
important
49
Somewhat
important
34
Neither
important
nor
unimportant
5
Not
very
important
3
Not
important
at
all
Total
9:

Even
if
never
use
river
52.1
36.2
5.3
3.2
3.2
33.8
27
25.6
21
25.0
19.4
16.7
21.3
17.6
25.0
30.6
13.9
20.4
10.2
47.7
34.2
16
14
27.6
24.1
22.4
15.5
10.3
32.8
31.0
15.5
15.5
5.2
12.6
18
15.9
23
12.1
19
108
6.7
6.7
4.7
13
96
58
50.0
34.4
19
18
Very
important
49
Somewhat
important
29
Neither
important
nor
unimportant
Not
very
important
:
Not
important
at
all
2
Total
93
52.7
31.2
:::
2.2
35.7
27
33.8
33
6.8
5.7
3.1
993
58
10.1
15
13.0
22
7.2
11
108
aProtest
bids
are
zero
bids
for
reasons
other
than
"
all
they
could
afford"
or
"
that
is
what
it
is
worth.
"

,
,..
 
 
.
.
­
.­
.
 
­
.
 
.
 
.
­.
 
­­
,.
..­.
.
.
.
.
.
.
.
..
 
7
T
a
b
l
e
4
­
6
.
R
e
s
p
o
n
d
e
n
t
A
t
t
i
t
u
d
e
s
A
b
o
u
t
S
e
l
f
b
y
K
e
y
R
e
s
p
o
n
d
e
n
t
G
r
o
u
p
s
 
 
User
Nonuser
 
.
Zero
bids
Nonzero
bids
Protest
bidsa
Total
Attitude
Frequency
%
Frequency
%
Frequency
%
Frequency
`#
j
Frequency
%
Frequency
%

An
outdoors
person
A
lot
42
Somewhat
24
A
little
19
Not
at
all
9
No
opinion
o
Total
94
An
environmentalist
I
A
lot
26
Somewhat
27
A
little
30
Not.
at
all
11
No
opinion
o
Total
94
Against
nuclear
power
electric
plants
A
lot
27
Somewhat
12
A
little
15
Not
at
all
31
No
opinion
Total
9:

Concerned
about
water
pollution
A
lot
45
Somewhat
29
A
little
15
Not
at
all
4
No
opinion
Total
9:

Willing
to
pay
the
cost
required
to
control
water
pollution
A
lot
19
Somewhat
42
A
little
21
Not
at
all
10
No
opinion
1
Total
93
44.7
25.5
20.2
9.6
0
27.7
28.7
31.9
11.7
o
"

28.7
12.8
16.0
33.0
9.6
48.4
31.2
16.1
4.3
0
20.4
45.2
22.6
10.8
1.1
50
56
E
o
207
38
56
51
59
2
206
45
19
23
79
40
206
87
71
29
17
3
207
31
59
50
58
8
206
24.2
27,1
18.4
30.4
0
18.4
27.2
24.8
28.6
1.0
21.8
19.2
11.2
38.4
19.4
42.0
34.3
14.0
8.2
1.4
15.0
28.6
24.3
28.2
3.9
29
23
21
35
0
108
28
14
24
39
2
107
26
13
8
37
24
108
41
31
18
16
2
108
8
18
15
58
8
107
26.9
21.3
19.4
32.4
0
26.2
13.1
22.4
36.4
1.9
24.1
12.0
7.4
34.3
22.2
38.0
28.7
16.7
14.8
1.8
7.5
16<
8
14.0
54.2
7.5
63
57
36
37
0
193
;
57
31
0
193
46
18
30
73
25
192
91
69
26
51
192
42
83
56
10
1
192
32.6
29.5
18.7
19.2
0
18.7
35.8
29.5
16.1
0
24.0
19.4
15.6
38.0
13.0
47.4
35.9
13.5
2.6
0.5
21.9
43.2
29.2
5.2
0.5
20
12
9
17
0
58
20
10
10
15
2
57
15
9
2:
11
58
28
17
85
0
58
6
10
9
28
4
34.5
21.0
15.5
29.3
0
35.1
17.5
17.5
26.3
3.5
25.9
15.5
5.2
34.5
19.0
48.3
29.3
13.8
8.6
0
10.5
17.5
15.8
49.1
7.0
92
80
57
72
0
301
64
83
81
70
2
300
72
31
38
110
49
300
132
100
44
21
3
300
50
101
71
68
9
299
30.6
26.6
18.9
23.9
0
21.3
27.7
27.0
23.3
0.7
24.0
10.3
12.7
36.7
16.3
44.0
33.3
14.7
7.0
1.0
16.7
33.8
23.7
22.7
3.0
 
 
 
 
 
.
 
l
 
Prgt.
st
bids
 
l
 
ro
zero
bids
for
reasons
other
than
"
all
they
could
afford"
or
`(
that
is
what
it
is
worth.
"
 
 
.
.
.
 
.
 
.
 
­
.
 
.
­­­­­
­
 
.
 
cent
of
the
zero
bidders
indicating
the
closest
identity
with
the
label.
This
is
even
more
evident
when
only
the
protest
zero
bids
are
examined.
Thirtyfive
percent
gave
the
strongest
response,
which
is
consistent
with
the
frequency
responses
shown
in
Table
4­
4
for
the
reasons
why
people
bid
zero.
The
most
dramatic
differences
between
respondents
are
evident
in
the
willingness
to
pay
the
cost
required
to
control
water
pollution.
Only
24
percent
of
the
zero
bidders
were
willing
to
identify
with
this
descriptive
statement.
This
consistency
across
different
attitude
responses
suggests
that
the
respondents
correctly
perceived
the
contingent
valuation
experiment
and
gave
careful
responses
that
would
not
have
been
given
if
hypothetical
bias
were
present.
It
is
also
suggestive
of
the
importance
of
attitudinal
questions
in
contingent
valuation
studies
both
for
analysis
purposes
and
as
consistency
checks.

Table
4­
7.
Logit
Estimation
of
Zero
Bids
a
Derivative
of
the
probability
b
evaluated
Independent
variable
Coefficient
t
­
r
a
t
io
at
the
mean
Constant
Sex
Age
Education
Income
Version
B
Version
C
Version
D
Willing
to
pay
cost
of
water
pollution
(
1
if
very
much
or
somewhat)
Interviewer
#
1
Interviewer
#
2
Interviewer
#
3
Interviewer
#
5
Interviewer
#
7
Interviewer
#
8
interviewer
#
9
­
0.435
­
0.522
0.036
­
0.108
6.9
X
10­
9
­
0.319
­
1.728
­
0.665
­
1.622
­
0.625
1.095
­
0.683
­
1.158
­
1.519
0.192
1.099
­
0.251
­
0.924
2.
703=
­
0.867
0.326
­
0.506
­
2.
407
C
­
1.099
­
3.185=

­
0.627
1.318
­
0.807
­
0.913
­
1.175
0.215
0.843
­
0.042
0.003
­
0.009
0
.
5
9
x
10­
6
­
0.026
­
0.113
­
0.050
­
0,169
­
0.044
0.128
­
0.050
­
0.072
­
0.082
0
.
0
1
7
0.141
Note:
Log
of
likelihood
function
=
65.511.
E
s
t
i
m
a
t
e
d
m
a
r
g
i
n
a
l
p
r
o
b
a
b
i
l
i
t
i
e
s
for
mean
value
of
dependent
variables:
Probability
=
1,
0.095;
probability
=
o,
0.905.

a
The
dependent
variable
is
equal
to
1
if
the
individual
bid
zero
dollars
and
~
zero
otherwise.
All
protest
bids
were
eliminated.

bThe
t­
ratio
is
the
ratio
of
the
estimated
parameter
to
the
estimated
standard
error.
Given
the
assumptions
of
the
estimates
are
maintained,
the
maxi
rnurn
l
i
k
e
l
i
h
o
o
d
,
Iogit
p
a
r
a
m
e
t
e
r
e
s
t
i
m
a
t
e
s
a
r
e
a
s
y
m
p
t
o
t
i
c
a
l
l
y
n
o
r
m
a
l
.
W
e
h
a
v
e
u
s
e
d
~
a
t
­
d
i
s
t
r
i
b
u
t
i
o
n
in
j
u
d
g
i
n
g
t
h
e
s
i
g
n
i
f
i
c
a
n
c
e
o
f
t
h
e
s
e
p
a
r
a
m
e
t
e
r
e
s
t
i
m
a
t
e
s
.

`
Significant
at
the
5­
percent
level.
j
4­
26
r
i
I
Additional
insight
into
zero
bidders
issues
can
be
obtained
from
a
logit
analysis
of
valid
zero
bids
(
see
Amemiya
[
1981]).
To
perform
this
analysis
for
the
Monongahela
s
t
u
d
y
,
the
dependent
variable
was
set
equal
to
1
if
a
zero
bid
was
given
and
equal
to
zero
if
a
positive
bid
was
given.
~
onprotf?
st
Consequently/
protest
bids
were
eliminated
from
the
analysis.
F
o
r
consisteflCY
f
the
explanatory
variables
used
are
the
same
as
in
the
option
price
regression
(
as
discussed
in
Section
4.5).
The
binary
variable
to
denote
~
onongahela
users
and
several
interviewer
dummies
were
eliminated
due
to
a
lack
of
variation.

The
results
of
the
Iogit
analysis
of
zero
bidders
are
shown
in
Table
4­
7.
This
model
requires
a
CaUtiOUS
interpretation
of
the
estimated
coefficients.
In
the
Ioglt
procedure,
the
expected
change
in
the
probability
of
bidding
zero
is
derived
from
the
estimated
equation
where
the
probability
of
bidding
zero
d~
pends
on
the
value
of
the
independent
variables.

I
/(

I
The
results
were
encouraging,
with
no
evidence
of
interviewers
significantly
affectin9,
the
odds
of
bidding
zero.
The
performance
of
other
variables
is
consistent
with
previous
results
and
~­
priori
reasoning.
Increases
in
age
significantly
affected
the
likelihood
of
btdding
zero.
Each
year's
increase,
evaluated
at
the
mean,
is
expected
to
change
the
probability
of
bidding
zero
by
0.003.
The
results
also
indicate
a
relationship
between
zero
bids
and
questionnaire
version.
When
the
respondent
was
presented
with
the
$
25
bidding
game
rather
than
the
payment
card,
the
probability
of
bidding
zero
decreased
by
O.
113.
Also,
the
attitude
toward
cost
was
consistent,
because
those
respondents
who
stated
a
willingness
to
pay
a
portion
of
cleanup
cost
had
a
lower
probability
of
bidding
zero.

The
Iogit
model
was
also
used
to
explain
why
individuals
protested
the
option
price
question.
As
shown
in
Appendix
C,
the
results
are
very
weak,
with
only
the
attitude
toward
cost
variable
significant
and
all
other
analysis
variables
insignificant.

4.5
OPTION
PRICE
RESULTS
The
central
element
in
a
contingent
valuation
study
is
the
valuation
responses
revealed
in
the
hypothetical
market
situation.
Much
of
the
analysis
in
the
early
contingent
valuation
experiments
focused
on
the
fitting
of
a
bid
function
to
the
willingness­
to­
pay
bids.
In
this
section,
a
linear
approximation
is
used
in
a
regression
analysis
to
fit
the
bid
function.
However,
the
basic
emphasis
of
the
regression
analysis
is
to
organize
the
information
presented
and
not
to
estimate
the
bid
function.
*

*
The
willingness­
to­
pay
data
contain
no
negative
bids
which
implies
that
they
are
truncated
at
zero.
This
can
lead
to
biased
parameter
estimates
with
regression
analysis,
depending
upon
the
distribution
of
bids.
Since
the
sam
­
Ple
excludes
protest
bidders
,
all
responses
should
fall
in
the
positive
domain.
Negative
responses
would
be
inconsistent
with
the
group
being
described
by
the
model.
The
difficulties
posed
by
truncation
could
be
handled
in
a
variety
of
ways
including:
transforming
the
dependent
variable
(
i.
e.
,
using
the
log
4­
27
Specifically,
this
section
summarizes
the
analytical
basis
of
the
option
price
and
user
amounts,
the
statistical
procedures
employed
to
analyze
these
estimates,
the
comparison
of
estimates
between
elicitation
methods,
and
the
results
on
starting
point
and
interviewer
bias.
In
addition,
it
also
compares
results
with
those
from
previous
studies.

The
amounts
provided
by
the
respondents
represent
their
option
prices
rather
than
user
willingness
to
pay,
as
measured
in
many
previous
contingent
valuation
studies.
That
is,
the
option
price
includes
both
the
expected
consumer
surplus
that
respondents
anticipate
from
future
use
of
the
site's
services
as
well
as
a
premium­­
the
option
value­­
that
they
are
willing
to
pay
to
obtain
these
site
services
should
they
decide
to
use
them.
The
premium
can
be
attributed
to
uncertainty
either
in
the
respondents'
future
demand
for
the
site
and/
or
uncertainty
in
the
supply
of
the
site's
services
at
given
water
quality
levels.
C
h
a
p
t
e
r
5
e
x
p
l
o
r
e
s
t
h
e
s
e
i
s
s
u
e
s
i
n
m
o
r
e
d
e
t
a
i
l
,
but
it
is
i
m
p
o
r
t
a
n
t
t
o
u
n
d
e
r
s
t
a
n
d
t
h
i
s
d
i
s
t
i
n
c
t
i
o
n
t
o
c
o
r
r
e
c
t
l
y
i
n
t
e
r
p
r
e
t
t
h
e
r
e
s
u
l
t
s
.

As
discussed
in
Chapter
2,
the
option
price
amounts
are
based
on
the
Hicksian
surplus
measures,
with
the
equivalent
surplus
measure
used
for
the
loss
of
the
recreation
services
of
the
Monongahela
River
(
Level
D
to
Level
E)
and
the
compensating
surplus
measures
used
in
measuring
the
option
price
for
the
improvements
to
fishable
and
swimmable
water.
The
use
of
these
measures
corresponds
to
the
existing
property
rights
for
the
overall
level
of
Monongahela
recreation
services,
with
the
river
currently
supporting
boating
activities.
It
is
important
to
note
that
several
sections
of
the
Monongahela
have
considerably
higher
water
quality
and
are
capable
of
supporting
sport
fishing
due
to
the
influence
of
tributaries.
However,
the
boatable
designation
is
a
reasonable
description
of
the
overall
water
quality
level.

Determining
the
treatment
of
outlying
responses
is
an
important
step
in
a
contingent
valuation
study.
Randall,
Hoehn,
and
Tolley
[
1981]
suggest
that,
once
the
outliers
are
determined
and
removed,
the
contingent
valuation
method
will
provide
a
"
core"
of
responses
useful
for
analysis.
In
general,
previous
efforts
have
used
subjective
judgment
in
making
this
determination,
with
little
or
no
discussion
provided.
For
example,
Rowe,
d'Arge,
and
Brookshire
[
1980]
follow
the
procedure
mentioned
in
Randall,
Ives,
and
Eastman
[
1974]
of
eliminating
bids
greater
than
10
standard
deviations
from
the
mean.
In
neither
case
is
much
discussion
provided
on
the
judgments
made
in
selecting
this
procedure
While
the
role
of
judgment
will
almost
always
loom
large
in
these
decisions,
it
is
difficult
to
evaluate
and
transfer
the
methods
used
to
evaluate
the
contingent
valuation
results
unless
a
more
systematic
basis
for
the
judgment
is
detailed.

of
the
bids,
"
if
the
zero
bidders
were
dropped)
and
using
an
alternative
estimator
For
the
purposes
of
the
present
analysis,
these
models
are
intended
to
be
used
only
as
a
basis
for
judging
the
factors
likely
to
influence
bids
and
not
necessarily
to
estimate
the
magnitude
of
their
impact.
Past
evidence
on
the
bias
of
ordinary
least
squares
in
presence
of
truncation
effects
indicates
that
it
did
not
greatly
affect
these
judgmental
evaluations
of
specific
variables.

4­
28
our
approach
relies
on
more
formal
use
of
statistical
indexes
of
the
influ
­
.
F
particular
observations
on
a
model's
estimated
parameters.
Belslev,
~
n~
e
uu
r­

~
uh,
and
Welsch
[
19801
suggest
a
 
,
,
number
of
statistical'
procedures
that
can
be
used
in
prescreening
data
for
outliers.
T
h
e
Monongahela
study
used
a
that
follows
their
discussion
to
identify
outlier
candidates.
The
procedure
Belsley­
Kuh­
Welsch
statistic
(
D
FBETA)
measures
the
effect
of
each
individual
on
each
of
the
estimated
coefficients
in
a
regression
model.
It
is
observation
estimated
by
Equation
(
4.
I
)
:

(
XTXJ'
x
i
T
e
i
DFBETA
s
b
­
b(
i)
=
~­
h
i
t
(
4
.
1
)

Where
b
=
the
estimated
coefficient
with
all
observations
included
b
(
i
)
=
the
estimated
coefficient
with
hi
­
1
T
=
xi
(
xTx)
xi
one
less
observation
e
i
=
the
ordinary
least­
squares
residuals.

This
statistic
is
not
a
formal
statistical
test.
It
is
merely
an
index
of
the
extent
of
influence
of
particular
observations.
It
implicitly
assumes
that
option
prices
can
be
related
to
economic
characteristics.
In
this
application,
the
statistics
presented
in
the
first
column
of
Table
4­
8
are
expressed
as
percentage
changes
in
the
income
coefficient
of
the
final
regression
model
discussed
later
in
this
chapter.
The
effect
of
income
was
selected
because
this
variable
is
the
only
variable
we
know,
based
on
economic
theory,
that
should
influence
option
price
bids.
Moreover,
the
relationship
between
option
price
and
user
value
can
be
expected
to
be
influenced
by
the
role
of
income
in
an
individual's
indirect
utility
function.
These
changes
represent
approximations
of
elasticities
described
in
Belsley,
Kuh,
and
Welsch
[
1980].

Rather
than
employ
one
of
the
arbitra~
y
statistical
criteria
suggested
in
Belsley,
Kuh,
and
Welsch,
the
procedure
was
supplemented
in
this
study
with
a
judgment
that
(
t)
30
percent
was
the
cutoff
point
for
outliers.
An
element
of
judgment
is
also
required
in
selecting
the
regression
model
from
which
the
Belsley­
Kuh­
Welsch
statistic
is
calculated.
After
comparing
models,
the
judgment
was
made
to
select
the
general
model
presented
later
in
Table
4­
11.
However
in
comparing
the
results
between
the
models,
the
16
outliers
determined
by
the
same
cutoff
point
for
another
regression
model
(
see
Appendix
G)
were
all
included
in
the
32
outliers
profiled
in
Table
4­
8.

The
results
in
Table
4­
8
are
striking
in
terms
of
the
differences
from
the
Randall,
Ives;
and
Eastman
[
1974]
criteria.
Many
of
the
outliers
or
zero
bids
that
would
have
been
retained
in
their
Drocedure.
In
the
consistency
in
the
characterization
of
the
respondents
classified
as
outliers,
63
percent
outliers"
is
informative.
earned
annual
incomes
are
small
addition,
For
the
of
$
2,500
4­
29
I
Table
4­
8.
Profile
of
Outliers
Option
price:
Belsley
­
Option
price:
avoid
improve
water
User
of
Kuh­
Welsch
loss
of
site
(
D
to
E)
quality
to
swimmable
J
ncome
Age
Education
Monongahela
Boat
statistic
Version
($/
yr)
($/
yr)
$/
Yr
(
yr)
Sex
(
yr)
site
ownership
­
233.12
­
155.99
­
100.04
­
79.83
­
66.19
­
63.25
­
62.95
­
56.70
­
54.98
­
49.68
­
44.62
­
43.80
A
­
43.16
(.
L
­
37.34
0
­
36.46
­
36.03
­
31.40
­
30.43
31.24
33.98
35.39
37.77
41.78
47.15
52.23
52.86
58.18
65.70
69.1S
79.58
82,52
112.04
$
125
bidding
game
$
125
bidding
game
direct
question
$
125
bidding
game
$
125
bidding
game
$
25
bidding
game
payment
card
$
25
bidding
game
direct
question
payment
card
$
125
bidding
game
$
25
bidding
game
$
125
bidding
game
$
25
bidding
game
$
25
bidding
game
$
25
bidding
game
direct
question
$
125
bidding
game
direct
question
$
125
bidding
game
$
125
bidding
game
payment
card
payment
card
$
125
bidding
game
$
125
bidding
game
payment
card
$
125
bidding
game
$
125
bidding
game
direct
question
$
125
bidding
game
payment
card
payment
card
$
125
$
125
$
200
500
$
125
25
450
60
0
50
155
5
155
5
25
0
200
200
500
75
25
5
0000
10
55
00
$
260
200
200
500
220
5
200
85
10
250
250
5
250
500
300
285
300
10
10
130
30
00
10
20
0
0
25
2,500
2,500
7,500
22,500
7,500
2,500
17,500
2,500
2,500
7,500
12,500
2,500
12,500
2,500
2,500
2,500
27,500
22,500
7,500
12,500
2,500
2,500
2,500
2,500
7,500
2,500
2,500
2,500
2,500
2,500
2,500
2,500
25
20
67
39
43
70
37
23
82
40
57
69
44
62
46
76
21
66
34
38
78
59
72
61
50
43
79
66
33
71
53
26
Male
Female
Male
Male
Female
Female
Female
Female
Female
Female
Female
Female
Female
Female
Female
Female
Female
Female
Male
Female
Female
Female
Female
Female
Female
Female
Female
Female
Female
Female
Female
Female
12
12
12
14
10
10
12
12
10
14
12
10
10
10
10
16
12
12
12
12
0
12
12
12
12
10
10
12
12
10
12
12
No
Yes
No
No
Yes
No
Yes
No
No
Yes
No
No
No
No
No
No
Yes
Yes
No
No
No
Yes
No
Yes
Yes
No
No
No
Yes
No
No
Yes
No
No
No
Yes
No
No
No
No
No
No
No
No
No
No
No
No
No
No
No
No
No
No
No
No
No
No
No
No
No
No
No
Yes
­­
 
.
.,?"­­­
.
,..
.
,­
.
.
.
.
.
.
.
,
P,.,,.
.,
.
.
,
.
.,
or
less,
and
78
percent
earned
less
than
$
7,500
a
year.
Female
a
Year
~
omprised
80
percent
of
the
outliers,
while
only
4
respondents
respondentsthan
a
high
school
degree.
The
last
element
of
interest
is
that
14
32
Outliers
had
received
the
$
125
starting
point
bidding
game­­
twice
as
had
more
of
the
as
the
next
version
(
the
payment
card).
This
last
feature
confounds
~
anY
of
starting
point
bias
presented
later
in
this
and
the
followthe
interpretation
ing
chapter.

In
summarY/
the
Belsley­
Kuh­
Welsch
[
1980]
procedure
is
a
systematic
for
identifying
outlying
bids
within
contingent
valuation
studies.
It
approach
does
not
replace
the
need
for
judgment
but
gives
a
basis
for
making
the
judg
­

~
entS.

The
results
presented
­
in
this
chapter
are
all
based
on
two
edits
of
the
301
COrnple@
d
survey
questionnaires.
The
first
edit
removed
the
protest
bids
~
rom
the
calculation
of
means
and
the
regressions.
Protest
zeros
were
bid
zero
for
reasons
other
than
"
that
is
all
they
could
afford"
respondents
who
or
Ilthat
is
what
it
was
worth.
"
This
removal
is
consistent
with
practices
of
R~
ndail,
Ivesl
and
Eastman
[
1974]
and
Rowe,
d'Arge,
and
Brookshire
[
1980]
.

The
second
edit
removed
the
outliers
following
the
Belsley­
Kuh­
Welsch
[
1980]
procedure.
Appendix
C
presents
the
estimated
means
for
both
the
full
~
ample
and
the
sample
w
i
t
h
o
n
l
y
t
h
e
p
r
o
t
e
s
t
b
i
d
s
e
x
c
l
u
d
e
d
.
Calculated
t.~
tatistics
revealed
no
statistically
significant
differences
between
the
means
estimated
from
the
full
sample
and
those
estimated
with
the
protest
bids
excluded.
The
effects
of
omitting
the
outlier
observations
are
discussed
at
the
appropriate
Points
in
this
and
in
the
following
chapter.

The
salient
questions
to
be
answered
from
the
survey
results
center
on
the
comparison
of
the
alternative
methods
used
to
elicit
the
option
price
amounts,
while
the
plausibility
of
the
results
is
substantiated
by
testing
for
potential
biases
in
the
responses.
Table
4­
9
presents
the
estimated
means
grouped
by
questionnaire
version,
with
distinctions
made
between
users
and
nonusers.
The
mean
values
are
provided
for
the
loss
of
the
recreation
services
of
the
site
(
avoiding
a
decrease
from
Level
D
to
Level
E
on
the
water
quality
ladder
in
Figure
4­
5),
for
an
improvement
in
water
quality
from
beatable
to
fishable
(
Level
D
to
Level
C),
and
for
an
improvement
in
water
quality
from
fishable
to
swimmable
(
Level
C
to
Level
B).
Combined
option
prices
are
presented
for
the
improvements
in
the
level
of
water
quality
and
for
the
improvements
plus
the
loss
of
the
services
of
the
site.

One
inference
that
can
be
drawn
from
Table
4­
9
is
that
the
option
prices
are
sizable
for
the
Monongahela
River
but
are
of
the
same
order
of
magnitude
regardless
of
the
method
used
to
elicit
the
amount.
Option
price
amounts
combined
for
all
levels
range
from
a
mean
of
$
54
per
year
for
the
bidding
game
With
a
$
25
starting
bid
to
$
118
for
the
bidding
game
with
a
$
125
starting
bid.
Mean
bids
for
the
combined
amounts
for
the
payment
card
and
direct
question
equaled
$
94
andis
even
narrower
from
$
25
per
year
$
56,
respective
y.
The
ran"
ge­
of
mean
option
price
amounts
when
only
the
bids
for
improvements
are
considered,
varying
to
$
60
per
year
for
the
two
bidding
games.

4­
31
Table
4­
9.
Estimated
Option
Price
for
Changes
in
Water
Quality:
Effects
of
1
nstrument
and
Type
of
Respondent­­
Protest
Bids
and
Outliers
Excluded
User
Nonuser
Combined
Change
in
water
quality
z
s
n
i
s
n
i
s
n
1.
Iterative
bidding
framework­­
starting
point
=
$
25
(
Version
C)

D
to
E
avoid
Dto
C
Cto
B
D
to
B
a
Combined:
all
levels
2.
iterative
bidding
D
to
E
(
avoid)
Dto
C
Cto
B
Dto
B
Combined:
all
levels
27.4
1
6
.
7
1
9
29.7
3
5
.
7
3
9
29.0
18.9
1
6
.
3
1
9
14.5
15.2
39
15.9
11.8
1
4
.
5
1
9
7
.
2
1
1
.
6
3
9
8
.
7
32.1
2
7
.
1
1
9
21.7
2
4
.
0
3
9
25.1
59.5
3
8
.
1
1
9
51.4
5
3
.
1
3
9
54.1
framework­­
starting
point
=
$
125
(
Version
D)

94.7
6
6
.
0
1
6
38.8
5
1
.
3
3
2
57.4
58.1
5
1
.
9
1
6
26.3
4
5
.
4
3
2
36.9
33.1
4
8
.
4
1
6
11.6
3
3
.
1
3
2
18.8
99.7
8
7
.
9
1
6
40.5
69.0
32
`
60.2
194.4
136.5
16
7
9
.
2
1
0
2
.
5
3
2
117.6
30.6
15.5
12.7
25.3
48.5
62.0
49.5
39.7
80.0
126.0
58
58
58
58
58
48
48
48
48
48
3.
Direct
question
framework
(
Version
B)

D
to
E
(
avoid)
45.3
6
5
.
2
1
7
14.2
2
7
.
1
3
4
24.5
4
5
.
4
51
D
t
o
c
31.3
4
4
.
2
1
7
10.8
2
1
.
6
3
4
17.6
32.1
51
Cto
B
20.2
35.5
17
8.5
2
1
.
9
3
4
12.4
2
7
.
4
5
1
i2to
B
52.9
7
2
.
5
1
7
20.3
4
1
.
4
3
4
31.2
5
5
.
2
5
1
Combined:
all
levels
9
8
.
2
1
0
3
.
5
1
7
34.5
6
6
.
4
3
4
55.7
8
5
.
2
5
1
4.
Direct
question
framework:
payment
card
(
Version
A)

D
to
E
(
avoid)
46.8
42.2
17
53.0
7
6
.
3
3
7
51.0
6
7
.
1
5
4
Dto
C
45.3
7
1
.
4
1
7
21.9
3
3
.
8
3
7
29.3
4
9
.
3
5
4
Cto%
22.9
4
8
.
7
1
7
7
.
7
2
0
.
0
3
7
12.5
3
2
.
2
5
4
Dto
B
7
1
.
2
1
1
7
.
7
1
7
29.9
4
7
.
5
3
7
42.9
7
8
.
1
5
4
Combined:
all
levels
1
1
7
.
9
117.0
1
7
8
2
.
8
1
0
4
.
7
3
7
9
3
.
9
1
0
8
.
9
5
4
aD
to
B
are
the
combined
amounts
for
improvements
only.

The
results
of
the
test
for
differences
in
means
between
methods
for
both
users
and
nonusers
are
shown
in
Table
4­
10.
These
results
show
that
the
differences
do
arise
between
the
means
in
the
bidding
games,
suggesting
there
may
be
a
bias
attributable
to
the
difference
in
the
starting
points.
The
combined
and
user
means
are
statistically
different
at
the
5­
percent
level
of
significance
for
users
and
for
the
combined
groups.
However,
the
evidence
is
not
completely
conclusive
because
the
differences
in
nonuser
means
are
not
significant.
In
addition,
the
regression
results
shown
in
Table
4­
11
do
not
conclusively
show
a
starting
point
bias
problem.
The
regression
model
estimated
without
the
outliers
shows
no
statistically
significant
difference
between
the
iterative
bidding
games.
If
the
outliers
are
not
removed,
the
model
suggests
starting
point
bias,
as
indicated
in
Appendix
C.
Thus,
in
the
regression
.

4­
32
I
Table
4­
1
0
.
Student
t­
Test
Results
for
Opti~
n
Price­­
Protests
Bids
and
Outliers
Excluded
User
Nonuser
Combined
Means
~
card
vs.
direct
qUeStiOn
payment
Dto
E
Eto
B
payment
card
vs.
$
25
iterative
bidding
r)
to
E
~
toc
,
Eto
B
payment
card
vs.
$
125
iterative
bidding
Dto
E
Direct
question
vs.
$
25
iterative
bidding
Dto
E
Direct
question
vs.
$
125
iterative
bidding
Dto
E
Dto
C
Eto
B
Dto
B
$
25
iterative
bidding
vs.
$
125
iterative
bidding
Dto
E
Dtoc
Eto
B
Dto
B
­­
2.806
­
­
2.300
.
.
­­
­­
.­

2.061
­­

­
2.499
­­

­­
­
2.074
­
2.161
­
2.453
­
­
­.

­
2.289
­
2.117
­
­
­
­

­
4.294
­­
­
3.119
­­
­
4.131
­­
­
3.183
­­
2.353
1.991
2.263
1.954
2.530
­­

­­

­
3.020
­
2.308
­
2.8786
­
2.109
­
3.072
­
3.046
­
3.539
­
3.159
aonly
cases
With
statistically
significant
differences
in
the
means
at
the
0.05
significance
level
are
reported.

analysis,
differences
attributable
to
starting
point
canhot
be
distinguished
from
the
influence
of
the
outlier
observations.

Some
additional
insights
into
differences
in
the
elicitation
method
can
be
developed
from
the
results
in
Tables
4­
10
and
4­
11.
The
mean
option
price
for
users
of
the
Monongahela
is
significantly
higher
when
the
bidding
game
with
the
$
125
starting
point
is
used
to
elicit
option
price
compared
to
either
direct
question
technique.
The
differences
are
present
for
the
aggregate
option
price
and
for
the
loss
of
site
services,
but
no
differences
are
detected
for
the
incremental
improvements
to
fishable
and
swimmable
water
quality
levels.

The
regression
results
from
Table
4­
11
are
generally
consistent
with
the
means
tests.
Using
the
dummy
variable
technique
to
compare
the
payment
card
with
the
other
three
versions
shows
option
price
is
significantly
higher
`
or
the
payinent
card
than
for
the
direct
question
and
the
$
25
bidding
game,
While
no
differences
exist
between
the
payment
card
results
and
those
for
tho
4­
33
Table
4­
11.
Regression
Results
for
Option
Prise
Estimates­­
Protest
Bids
and
Outliers
Excluded
Water
quality
chanqes
Total
improve­
Independent
variables
D
to
E
(
avoid)
Dto
C
Cto
B
Total,
all
levels
ments
only
Intercept
Sex
(
1
if
male)

Age
Education
Income
.

Direct
question
Iterative
bidding
game
($
25)

Iterative
bidding
game
($
125)

User
(
1
if
user)

Willing
to
pay
cost
of
water
pollution
(
1
if
very
much
or
somewhat)
Interviewer
#
1
Interviewer
#
2
Interviewer
#
3
Interviewer
#
4
Interviewer
#
5
Interviewer
#
6
Interviewer
#
7
Interviewer
#
8
Interviewer
#
9
R
2
F
Degrees
of
f
reeciom
­
34.512
(­
0.973)
8.451
(
0.916)
­
0.292
(­
1.094)
s.
294
(
2.071)
b
0.0006
(
1.6S2)
­
32.311
(­
2.771)
b
­
20.623
(
1.852)
1.7522
(
1.421)
8.840
(
0.919)
17.001
(
1.788)

14.211
(
0.750)
1.723
(
0.099)
­
22.833
(­
1
.344)
­
28.125
(­
0.860)
6.932
(
0.404)
47.012
(
0.887)
27.670
(
1.425)
14.022
(
0.801)
17.874
(
0.454)

0.334
3.78
136
­
29.307
(­
1
.098)
­
0.672
(­
0.097)
0.290
(­
1.440)
2.901
(
1
.508)
0.0003
(
1.1s1)
­
14.372
(­
1.638)
­
12.572
(­
1.
s00)
6.639
(
0.716)
8.083
(
1.117)
21.960
(
3.068)
b
7.090
(
0.497)
12.242
(
0.938)
21.141
(
1
.653)
3.
0s0
(
0.124)
4.996
(
0.387)
95.513
(
2.394)
b
2.470
(
0.169)
29.961
(
2.274)
b
39.586
(
1.336)

0.284
3.00
136
­
5.430
(­
0.257)
­
1.657
(­
O.
302)
­
0.265
(
1
.668)
­
5.27
(
0.347)
0.0003
(
1
.260)
­
3.500
(
0.505)
­
5.657
(
i.
8SJ)

(
0:
101)
6.
839
b
(
1.86)
10.023
(
1.772)

11.334
(
1.006)
16.849
(
1.634)
17.578
(
1
.740)
20­
605
(
1
.059)
2.191
(
0.215)
66.288
(
2.102)
b
4.130
(
0.357)
19.871
(
1.808)
­
7.935
(­
0.339)

0.166
1.50
136
­
56.653
(­
0.916)
6.484
(
0.403)
­
0.854
(­
1.834)
8.066
(
1.810)
0.0012
(
1
.832)
­
50.734
(­
2.495)
b
`
­
39.566
(­
2.037)
b
31.089
(
1.446)
26.026
(
1
.552)
51.326
(
3.095)
b
26.509
(
0.802)
24.719
(
0.817)
9.292
(
0.314)
­
12.334
(­
0.216)
11.435
(
0.382)
198.450
(
2.146)
b
39.645
(
1.170)
58.063
(
1.902)
37.330
(
0.544)

0.366
4.36
136
­
22.141
(­.
517)
1.967
(­
0.177)
­
0.562
(
1.743)
2.773
(
0.899)
0.0006
(
1.278)
­
18.423
(­
1.309)
­
18.943
(
1.409)
13.568
(
0.912)
17.187
(
1,481)
34.326
(
2.990)
b
12.298
(
0.538)
22.996
(
1.099)
32.125
(
1.567)
1s.
791
(
0.400)
4.503
(
0.217)
151.439
(
2.366)
b
11.975
(
o
511)
44.041
(
2.08)
b
19.456
(
0.­)

0.269
0.278
136
aNumbers
in
parentheses
are
asymptotic
t­
ratios
for
the
null
hypothesis
of
no
association.
 
 
b
.
Significant
at
the
0.05
level.

b
i
d
d
i
n
g
game
with
the
$
125
starting
point.
The
differences
are
significant
only
for
the
loss
of
site
services
and
for
the
combined
option
price.
When
other
influences
are
held
constant
in
the
regression
analysis,
respondents
who
received
the
payment
card
expressed
aggregate
option
prices
approximately
$
40
to
$
50
higher
than
those
expressed
by
respondents
in
the
$
25
starting
point
bidding
game
and
the
direct
question.
It
is
possible
to
conclude
that
there
are­
significant
differences
between
methods
but
that
all
methods
estimate
option
price
at
the
same
order
of
magnitude.
The
differences
cannot
be
detected
among
the
bids
for
improvements
in
water
quality
levels,
possibly
because
the
effects
of
the
methods
are
limited
to
the
initial
amounts
given.
This
may
minimize
the
effect
of
the
question
format
when
incremental
amounts
are
elicited.
This
conclusion
should
be
viewed
with
some
caution
sinct
tho
differences
between
methods
could
be
difficult
to
detect
simply
becauso
th.

4­
34
T.
.
.
.
number
of
bids
for
the
improvements
iS
too
Small
to
offset
the
variation
in
the
expressed.
The
consistency
in
the
results
from
the
various
tests,
amounts
however
I
is
particularly
encoura9in9
as
a
plausibility
check
against
the
influ
­
bias
in
the
contingent
valuation
design.
ence
of
hypothetical
An
examination
of
the
regress"
mn
results
for
option
price
combined
over
levels
reinforces
the
plausibility
of
the
results.
The
coeffi
­
all
water
quality
socioeconomic
variables
all
have
the
expected
signs,
and
the
co­
~
ients
of
the
for
age,
education
level,
efficient
and
income
are
significant
at
either
the
0.05
level
or
very
close
to
it.
The
results
indicate
a
strong
role
for
respondent
attitude
toward
paying
the
cost
of
water
pollution.
Persons
who
identified
themselVe5
as
either
very
much
or
somewhat
willing
to
pay
for
water
pollution
control
Were
willing
to
spend
$
50
more
per
year
than
persons
who
were
not
willing
to
pay
the
cost
,
with
all
other
things
held
constant.
This
consistency
of
attitudes~
combined
with
the
performance
of
the
socioeconomic
variables
and
t
h
e
ability
`
f
`
he
`
ode'
`
0
`
Xp'ajn
almost
.37
Percent
of
the
variation
in
option
price,
builds
a
strong
case
against
the
Influence
of
hypothetical
bias
in
the
contingent
valuation
design.

The
regression
results
in
Table
4­
11
also
shed
some
light
on
the
question
.
.
of
a
bias
in
the
Wlllln9ness
tO
PaY
that
could
be
attributable
to
differences
in
interviewers.
Using
the
dummy
variable
technique,
the
results
indicate
that
the
influence
of
interviewer
bias
is
limited.
Only
for
two
interviewers
are
the
coefficients
statistically
significant
at
the
0.05
level
for
some
levels
of
water
quality.
One
of
the
cases
involved
an
interviewer
who
conducted
only
two
interviews
before
being
removed
from
the
interviewing
team.
This
interviewer
did
not
take
part
in
the
training
session
and
also
conducted
interviews
only
in
the
Latrobe
area,
which
is
a
considerable
distance
from
the
Monongahela
River.
The
second
interviewer
also
conducted
interviews
in
the
Latrobe
area
and
in
one
area
very
close
to
the
river.
These
cases
may
simply
reflect
the
model's
inability
to
differentiate
between
an
interviewer
effect
and
some
omitted
variables.
Thus,
the
effect
of
the
interviewer
is
quite
small
and
reinforces
the
importance
of
the
training
sessions
that
were
conducted
in
Pittsburgh
prior
to
the
survey.
*

#
r
i
1
Table
4­
12
presents
the
results
of
student
t­
tests
for
differences
in
means
between
users
of
the
Monongahela
River
and
nonusers
broken
down
by
the
technique
used
to
elicit
option
price.
The
results
show
that
users
who
received
either
the
direct
question
or
the
$
125
starting
point
bidding
game
expressed
bids
that
were
higher
than
those
of
nonusers.
There
were
no
statistically
significant
differences
in
means
for
either
the
payment
card
or
the
$
25
starting
point.
T
h
i
s
s
u
g
g
e
s
t
s
t
h
a
t
u
s
e
r
s
h
a
v
e
s
o
m
e
w
h
a
t
h
i
g
h
e
r
o
p
t
i
o
n
p
r
i
c
e
s
,

*
To
conclusively
design
a
test
for
interviewer
bias
would
require
that
interviewers
be
randomly
assigned
to
different
areas
in
the
survey.
The
practical
issue
is
that
this
could
have
a
significant
impact
on
data
collection
costs
because
of
interviewers
having
to
cover
a
substantial
part
of
the
survey
area.
In
the
Monongahela
survey,
interviewers
were
assigned
areas
based
on
the
lowest
travel
costs
to
obtain
the
interview.

4­
35
Table
4­
12.
Student
t­
Test
Results
for
Option
Price­­
Protest
Bids
and
Outliers
Excluded
Means
compared
User
vs.
nonuser
Means
compared
User
vs.
nonuser
Payment
$
25
iterative
*
­
0.313
bidding
(
C)
Dto
E
­
0.275
Dto
C
1.645
Dto
C
1.026
Cto
B
1.322
Cto
B
1.322
Dto
B
1.103
Dto
B
0.591
Eto
B
1.847
Eto
B
1.488
Direct
$
125
iterative
question
(
B
)
bidding
(
D)
Dto
E
2
.
4
1
4
a
Dto
E
3
.
2
3
1
a
Dto
C
2.
234
a
Dto
C
2
.
1
8
6
a
Cto
B
1.454
Cto
B
1.819
Dto
B
2.
669
a
Dto
B
3.
279
a
Eto
B
2.
049
a
Eto
B
2.
555
a
`
Denotes
significance
at
the
0.05
level.

but
this
difference
is
not
pervasive.
Thus,
a
survey
of
only
the
users
of
Monongahela
River
would
have
substantially
underestimated
the
recreation
and
related
benefits
of
water
quality
improvements.
The
full
extent
of
these
intrinsic
benefits
is
developed
in
the
following
chapter.

4.6
USER
VALUE
RESULTS
Table
4­
13
shows
estimated
user
values,
referring
to
the
value
card
(
see
Figure
4­
6)
component
of
the
option
price.
These
values
which
resulted
from
respondents
and
breaking
out
the
user
v
a
l
u
e
are
comparable
to
those
estimated
in
most
of
the
previous­
contingent
valuation
efforts
and
are
compared
w
i
t
h
the
benefits
estimated
with
the
travel
cost
`
method
in
Chapter
8
.

User
value
means
are
presented
for
users
only
and
the
means
calculated
for
all
respondents.
Tests
to
determine
whether
the
user
values
are
statistically
different
from
zero,
shown
in
Appendix
C,
indicated
that
the
user
values
for
the
D
to
E
levels
and
combined
over
all
levels
are
statistically
different
from
zero
at
the
0.05
level
of
significance.
The
user
values
for
improvements
in
water
quality
are
only
different
from
zero
for
the
$
25
bidding
game
and
not
for
any
other
methods.
Additional
tests
for
differences
in
user
values
between
methods,
also
contained
in
Appendix
C
I
showed
that
means
from
the
$
25
biddina
aames
were
statistically
different
(
lower)
than
those
esti
­

.­
mated
with
the
$
125
bidding
game,
but
only
for
Levels
D
to
E
values
for
all
combined
water
quality
levels.
The
differences
4­
36
and
the
user
for
the
user
Table
4­
13.
Estimated
User
Values­­
Protest
Bids
and
Outliers
Excluded
~
User
only
Combined
i
s
n
2
s
n
Iterative
bidding
f
r
a
m
e
w
o
r
k
$:
5
~~
t~
ting
point
(
C)

I)
toc
Cto
B
Dto
B
Combined:
all
levels
Iterative
bidding
framework
$:
2:
os~
rting
point
(
D)

Dto
C
cto
B
Dto
B
Combined:
all
levels
Direct
question
(
B)
Dto
E
Dto
C
Cto
B
Dto
B
Combined:
all
levels
Direct
question:
payment
card
(
A)
Dto
E
Dto
C
Cto
B
Dto
B
Combined:
all
levels
6.59
4.21
5.00
10.53
17.11
36.25
20.31
20.00
48.75
138.11
19.71
21.18
10.00
31.18
50.88
19.71
30.88
19.71
51.18
70.88
12.59
19
7.68
19
7.99
19
14.43
19
25.13
19
58.98
16
42.67
16
42.82
16
87.87
16
85.00
16
37.85
17
42.22
17
29.10
17
64.63
17
77.46
17
34.30
17
74.57
17
49.42
17
122.88
17
127.61
17
2.16
7.73
58
1.38
4.76
58
1.64
5.08
58
3.45
9.52
58
5.60
16.28
58
12.08
37.52
48
6.77
25.98
48
6.66
25.99
48
16.25
54.81
48
28.33
87.90
48
6.57
23.38
51
7.06
25.93
51
3.33
17.14
51
10.39
39.46
51
16.96
50.07
51
6.20
20.99
54
9.72
43.45
54
6.20
28.68
54
16.11
71.65
54
22.31
77.59
54
values
combined
for
all
respondents
were
the
same
as
those
for
users,
except
for
the
comparison
of
bidding
games
,
where
the
difference
was
significant
only
for
the
Level
Dto
E
change.

Table
4­
14
presents
the
results
for
the
regression
models
with
the
user
values
as
the
dependent
variables.
The
models
generally
have
less
explanatory
power
than
the
option
price
models
but
do
show
some
limited
ability
to
explain
variations
in
user
value.
Age
and
respondent
attitude
toward
paying
the
cost
of
water
pollution
are
the
key
variables
in
the
model,
and
both
have
the
expected
signs.

4­
37
Table
4­
14.
Regression
Results
for
User
Value
Estimates
o
Water
Quality
Changes
­­
Protest
Bids
and
Outliers
Excludedi
Water
quality
changes
Total
com­
Total
improve­
Independent
variable
D
to
E
(
avoid)
Dto
C
cto
B
bined
all
levels
ments
only
Intercept
Sex
(
1
if
male)
(
1
if
male)
Age
Education
Income
Direct
question
Iterative
bidding
game
($
25)

Iterative
bidding
game
($
125)

Willing
to
pay
cost
of
water
pollution
(
1
if
very
much
or
somewhat)
Interviewer
#
1
Interviewer
#
2
Interviewer
#
3
Interviewer
#
4
Interviewer
#
5
Interviewer
#
6
Interviewer
#
7
Interviewer
#
8
Interviewer
#
9
R
2
F
Degrees
of
freedom
10.372
(
0.551)
1.070
(
0.218)
­
0.236
(­
1.761)
0.193
(
0.142)
0.00001
(
0.073)
­
2.842
(­
0.456)
­
4.769
(­
0.803)
6.665
(
1.014)
9.931
(
1.988)
b
­
1.585
(­
0.157)
4.626
(
0.500)
­
3.479
(­
0.395)
­
9.651
(­
0.553)
­
5.724
(­
0.624)
­
6.266
(­
0.221)
12.634
(
1
.225)
­
5.509
(­
0.589)
­
18.707
(­
0.889)

0.13
1.22
137
1.529
(
0.070)
­
1.625
(­
0.285)
­
0.264
(­
1.690)
0.156
(
0.098)
0.0002
(
0.740)
­
5.766
(­
0.796)
­
10.724
(+:?:)

(­
1.119)
10.828
(
1.866)

4.020
(
0.343)
13.666
(
1.270)
27.836
(
2.721)
b
7.079
(
0.349)
1.410
(
0.132)
19.835
(
0.602)
4.664
(
0.389)
11.417
(
1
.050)
­
3.159
(­
0.129)
0.14
1.34
137
­
2.143
(:::;;;)

(­
0.026)
­
0.201
(­
1.817)
0.464
(
0.412)
0.00003
(
0.167)
­
4.300
(­
0.836)
­
5.072
(­
1
.035)
­
3.006
(­:.:;:)

(
l:
969)
b
3.029
(
0.364)
11.118
(
1.455)
19.108
(
2.630)
b
2.996
(
0.208)
­
0.087
(­
0.012)
11.477
(
0.491)
1.177
(
0.138)
3.960
(
0.513)
­
3.381
(­
0.195)

0.14
1.28
137
6.686
(
0.180)
0.121
(
0.013)
­
0.507
(­
1.918)
­
0.063
(­
0.023)
0.0002
(
0.607)
­
11.536
(­
0.940)
­
15.588
(­
1
.333)
­
7.103
(­
0.549)
19.654
(
1.997)
b
8.758
(
0.441)
25.736
(
1.411)
47.530
(
2.740)
9.987
(
0.290)
3.474
(
0.192)
27.795
(
0.498)
16.328
(
0.803)
15.851
(
0.860)
­
8.995
(­
0.217)

0.14
1.34
137
17.058
(
0.363)
1.191
(
0.097)
­
0.743
(­
2.220)
b
0.130
(
0,038)
0.0003
(
0.508)
­
14.378
(­
0.925)
­
20.358
(­
1.374)
­
0.438
(­
0.027)
29.586
(
2.374)
b
7.172
(
0.285)
30.362
(
1.314)
44.051
(
2.005)
b
0.336
(
0.008)
­
2.250
(­
0.098)
21.529
(
0.305)
28.962
(
1.125)
10.342
(
0.528)
­
27.702
(­
0.528)

0.15
1.44
137
a
Numbers
in
parentheses
are
asymptotic
t­
ratios
for
the
null
hypothesis
of
no
association.
b
.
Significant
at
the
0.05
level.

4.7
SUMMARY
The
contingent
valuation
estimates
of
the
option
price
for
quality
improvements
are
consistently
plausible
throughout
the
various
analytical
considerations.
The
empirical
results
indicate
that
the
methods
used
to
elicit
the
bid
do
have
a
statistically
significant
effect
on
the
estimates
ofan
individual's
valuation".
Payment
cards
and
the
bidding
game
with
a
$
125
starting
point
produced
higher
willingness­
to­
pay
estimates
than
either
the
direct
question
or
the
bidding
game
with
a
$
25
starting
point.
There
is
some
evidence
4­
38
of
a
S
t
a
r
t
i
n
g
p
o
i
n
t
bias
in
the
bidding
game,
but
the
statistical
analyses
are
The
results
comparing
bidding
games
with
non
bidding
games
not
~
onclusive"
no
differences
when
these
combined
comparisons
are
made.
In
terms
indicated
contingent
valuation
experiments,
the
results
imply
that
using
bid­
~
f
future
ding
games
to
elicit
willingness
to
pay
requires
a
range
of
starting
points
to
test
for
startin9
point
bias.
No
statistical
or
analytical
differences
are
apparent
when
nonbiddln9
9ames
are
employed
to
elicit
willingness
to
pay.

For
the
continued
use
of
the
contingent
valuation
method
to
estimate
~~
nefitS
of
water
quality
improvements,
the
general
prognosis
from
the
results
River
case
study
is
a
good
one.
The
empirical
models
per­
Of
t
h
e
Monongahela
formed
reasonably
well
in
explaining
variations
in
willingness
to
pay,
with
little
indication
that
individual
interviewers
influenced
the
results.
The
consistently
and
magnitudes
of
key
economic
variables
suggest
that
the
plausible
si9ns
respondents
perceived
the
realism
of
the
survey
and
did
not
experience
problems
with
the
hypothetical
nature.
Moreover,
the
results
came
from
a
random
sample
of
households
from
an
area
whose
socioeconomic
profile
is
not
ideally
suited
for
a
contingent
valuation
survey:
The
respondents
were
older,
less
educated,
and
poorer
than
in
previous
contingent
valuation
studies.

4­
39
"'""­
CHAPTER
5
CONTINGENT
VALUATION
DESIGN
AND
RESULTS:
OPTION
AND
EXISTENCE
VALUES
5.1
INTRODUCTION
Over
a
decade
ago,
Krutilla
[
1967]
emphasized
the
importance
of
nonuser
benefits
to
the
process
of
efficiently
allocating
natural
environments.
In
his
development
of
the
special
problems
associated
with
valuing
the
services
of
natural
environments,
Krutilla
identified
several
types
of
nonuser
values.
The
objective
of
this
chapter
is
to
present
survey
results
that
attempt
to
measure
directly
two
of
the
sources
of
benefits
Krutil!
a
identified­­
option
value
and
existence
value.
It
should
be
acknowledged
at
the
outset
that
the
first
of
these,
option
value,
has
received
the
greatest
attention
in
the
literature
and
is
regarded
as
one
of
the
most
important
components
of
nonuser
values.
As
a
consequence,
the
majority
of
this
chapter
is
devoted
to
the
theoretical
and
empirical
problems
associated
with
modeling
and
measuring
option
value.

The
simplest
approach
to
defining
option
value
is
to
use
an
example.
Consider
an
individual
who
is
uncertain
whether
he
will
visit
a
recreation
site
on
the
Monongahela
River
in
the
future.
Also,
suppose
this
person
is
uncertain
whether
the
facility
will
be
available
in
the
future
should
he
decide
to
use
it.
This
uncertainty
over
availability
may
arise
because
the
individual
either
does
not
know
whether
the
facility
will
permit
any
use
or
does
not
know
the
types
of
uses
that
will
be
permitted.
(
For
example,
a
river
may
not
permit
any
use,
or
it
simply
may
not
be
available
for
swimming.
Of
course,
the
inability
to
support
recreational
swimming
does
not
preclude
the
provision
of
sport
fishing
and
boating
services.
)
What
is
at
issue
is
uncertainty
over
the
character
of
the
supply.
This
uncertainty
can
involve
the
all­
or­
none
case,
a
concept
conventionally
used
in
the
theoretical
literature,
or
simply
a
change
in
the
types
of
uses
that
can
be
supported
in
the
future.
Given
these
condi
­
tions,
a
rational
individual
may
be
willing
to
pay
some
amount
for
the
right
to
use
the
facility's
services
in
the
future.
This
payment
can
be
interpreted
as
a
means
of
insuring
access
to
the
site's
services.
Of
course,
it
does
not
eliminate
the
individual's
uncertainty
over
whether
he
will
actually
decide
to
use
the
site's
services.

In
all
discussions
of
option
value,
the
payment
is
assumed
to
be
constant
regardless
of
whether
or
not
the
individual
visits
the
site.
The
payment
is
usually
described
as
the
option
price.
The
option
value
is
defined
as
the
difference
between
this
payment
and
the
individual's
expected
consumer
surplu,
from
having
access
to
the
site's
services.
1
n
the
extreme
case,
where
the
choice
is
use
or
no
use,
the
expected
consumer
surplus
is
the
weighted
sum
5­
1
 
 
.
(
by
the
relevant
probabilities)
of
the
consumer
surplus
associated
with
access
and
use
of
the
site
plus
that
of
access
and
no
use.
Of
course,
it
must
be
recognized
that
this
discussion
assumes
that
markets
do
not
exist
for
contingent
claims
that
could
handle
the
prospects
for
a
future
demand.
Thus,
there
is
no
alternative
mechanism
(
other
than
purchasing
the
option)
available
to
the
individual
for
diversifying
the
risk
he
experiences.

Researchers
have
generally
agreed
that
this
description
of
behavior
is
plausible.
The
literature,
however,
includes
a
wide
array
of
arguments
concerning
the
relationship
between
the
maximum
willingness
to
pay
for
the
option
and
the
exDected
consumer
surplus.
For
example,
Cicchetti
and
Freeman
[
1971
]
observed
that
option
value
existed
as
a
direct
result
of
risk­
averse
behavior
and
was
therefore
positive.
By
contrast,
using
a
similar
framework,
Schmalensee
[
1972]
concluded
that
option
value
may
be
positive
or
negative
depending
on
the
vantage
point
selected
for
evaluating
the
individual's
choices.
Subsequent
contributions
questioned
Schmalensee's
definition
of
risk
aversion
(
Bohm
[
1975]);
introduced
time
specifically
into
the
analysis
(
Arrow
and
Fisher
[
1974];
Henry
[
1974];
and
Conrad
[
1980]);
and,
more
specifically,
considered
the
mechanisms
available
to
the
individual
for
diversifying
risk
(
Graham
[
1981
]).
The
result
has
been
a
large
and
often
confusing
literature.

Understanding
the
past
contributions
in
this
area
requires
a
clear
description
of
three
aspects
of
the
role
of
uncertainty
in
each
model.
This
characterization
of
uncertainty
is
most
easily
summarized
by
posing
three
questions

.
What
is
the
source
of
the
uncertainty
in
the
individual's
decision
problem?

.
How
will
the
uncertainty
in
this
decision
problem
ultimately
be
resolved?

.
Is
it
possible
to
amend
the
decision
process
to
accommodate
new
information
that
may
resolve
some
of
the
uncertainty?

Each
of
the
past
analyses
of
option
value
provides
implicit
answers
to
these
questions.
Moreover,
the
answers
help
explain
why
these
analyses
yield
such
diverse
conclusions.

Two
recent
papers
have
provided
the
elements
necessary
to
integrate
a
significant
portion
of
the
literature.
The
first
of
these
is
a
review
article
by
Bishop
[
1982]
that
provides
an
excellent
summary
of
past
contributions
and
extends
earlier
work
by
amending
Schmalensee's
framework
to
delete
the
individual's
demand
uncertainty
and
to
explicitly
include
supply
uncertainty.
In
the
second
paper,
Graham
[
1981
]
seeks
to
define
the
appropriate
measure
of
benefits
for
benefit­
cost
analyses
in
the
presence
of
uncertainty.
He
concludes
as
Bohm
[
1975]
did
earlier,
that
option
price
and
not
expected
consumer
surplus
is
the
appropriate
valuation
measure.
 
 
Unfortunately,
his
ev=­
=
of
the
problem
tends
to
focus
on
cases
where
individuals
face
specific
risks
and
have
access
to
ideal
markets
in
which
to
diversify
these
risks.
For
these
cases,
he
quite
correctly
concludes
option
value
is
largely
irrelevant.

5­
2
(

1
I
I
I
I
Of
course,
most
resource
and
environmental
problems
do
not
"
fit"
these
assumptions.
Nonetheless,
his
framework
and
evaluation
of
the
case
of
collective
risk
provide
another
important
insight
into
the
appropriate
treatment
of
option
value.

Section
5.2
of
this
chapter
reviews
the
modeling
of
uncertainty
and,
specifically,
the
use
of
a
contingent
claims
framework.
This
review
is
necessary
to
understand
the
implications
of
alternative
definitions
of
risk
aversion.
With
this
background
it
is
possible
in
Section
5.3
to
describe
the
"
timeless"
analyses
of
option
value
and
to
relate
them
to
the
recent
contributions
of
Bishop
[
1982]
and
Graham
[
1981].
Section
5.4
briefly
discusses
the
relationship
between
option
value
and
quasi­
option
value
introduced
by
Arrow
and
Fisher
[
1974].

Section
5.5
discusses
three
recent
attempts
to
empirical
y
estimate
nonuser
values­­
the
Green
ley,
Walsh
,
and
Young
[
1981
]
estimates
of
option
value
from
potential
water
quality
degradation
in
the
South
Platte
River
basin
in
Colorado;
Mitchell
and
Carson's
[
1981
]
estimates
of
the
total
"
intrinsic"
values
for
improvements
in
national
water
quality;
and
the
Schulze
et
al.
[
1981]
analysis
of
visibility
benefits
for
national
parks
in
the
Southwest.

Sections
5.6
through
5.8
describe
the
survey
results
for
the
Monongahela
River
basin.
Section
5.6
describes
the
questions
used
to
estimate
option
value
and
to
determine
its
sensitivity
to
the
character
of
the
supply
uncertainty
The
survey
has
been
structured
so
that
it
is
possible
to
distinguish
the
estimates
according
to
the
question
used,
the
level
of
supply
uncertainty,
and
the
character
of
the
respondents.
Respondents
are
grouped
according
to
whether
they
have
used
the
river
for
recreation
purposes.
Section
5:
7
presents
a
summary
of
the
empirical
results
and
an
evaluation
of
the
effects
of
the
questioning
mode
(
as
well
as
of
the
starting
point
for
the
iterative
bidding
scheme)
used
for
the.
estimates.
In
addition
to
measuring
option
value,
attempts
were
made
to
measure
existence
values
independently.
Section
5.8
discusses
these
efforts.
Section
5.9
presents
a
summary
of
the
primary
findings
of
this
research.

5.2
CONTINGENT
CLAIMS
MARKETS
AND
THE
MODELING
OF
UNCERTAINTY*

The
traditional
approach
to
dealing
with
production
and
exchange
deci
­
sions
under
uncertainty
involves
a
definition
of
new
commodities
that
specifies
not
only
their
physical
characteristics,
location,
and
date
of
availability,
but
also
a
particular
state
of
the
world
that
must
be
realized
if
the
stipulated
transaction
is
to
take
place.
In
terms
of
the
example
used
in
Section
5.1,
one
state
of
the
world
permits
access
to
the
Monongahela
River
recreation
site
and
another
does
not.
In
this
framework,
uncertainty
has
the
effect
of
expanding
the
commodity
set
available
to
the
individual.
For
example,
if,
in
*
The
theoretical
analysis
in
this
chapter
is
an
expanded
version
of
that
reported
in
Smith
[
1983]
.

.
.
.
.
.
 
.
.".
..
 
.
 
.
.
.
.
.
.
­­­
 
­­­­­
.
 
 
­
­­­­
.

5­
3
the
absence
of
uncertainty,
there
are
N'
commodities,
and
if
uncertainty
introduces
K
states
of
nature,
a
contingent
claims
model
redefines
the
commodity
set
to
be
N
"
K
contingent
claims.
Each
is
a
claim
to
a
good
contingent
upon
the
state
of
nature.
In
this
framework,
the
model
is
describing
how
an
individual's
plans
for
activities
are
made
rather
than
the
actual
activities
themselves
These
plans
involve
the
selection
of
claims
to
goods,
should
the
state
of
the
world
be
realized.
Thus,
the
individual
must
allocate
his
budget
optimally
among
these
claims
before
the
state
of
the
world
is
known.
.
 
 
 
 
.

Of
course,
defining
optimality
in
this
framework
requires
consideration
of
the
rule
that
aggregates
these
claims.
Because
each
of
these
new
commodities
involves
both
a
good
and
a
state
of
world,
each
outcome
needs
an
associated
probability.
This
permits
the
use
of
expected
utility­­
justified
in
the
early
work
of
von
Neumann
and
Morgenstern
[
1947]
­­
as
the
rule
for
aggregating
the
values
associated
with
these
claims.
That
is,
given
the
four
postulates
of
rational
choice,
the
utility
of
any
set
of
contingent
claims
(
e.
g.,
a
commodity
considered
over
all
states
of
nature)
can
be
derived
as
the
expected
utility.
*
The
most
important
of
these
postulates
for
understanding
the
literature
on
option
value
is
the
uniqueness
postulate,
which
requires
the
expected
utility
of
a
set
of
claims
to
be
independent
of
the
"
state
labeling"
of
the
commodities
involved
in
these
claims.
That
is,
these
commodities
could
be
rearranged
over
all
states
of
nature
without
changing
the
expected
utility
as
long
as
each
commodity
is
realized
with
the
same
probability.

Most
analyses
of
option
value
drop
this
postulate
by
assuming
that
the
individual
has
a
different
utility
function
depending
on
whether
the
services
of
a
recreation
site
are
demanded
or
not
demanded.
The
presence
of
a
positive
level
of
demand
for
the
site
is
not
simply
a
reflection
of
a
higher
income
or
a
lower
price.
With
a
given
income
and
prices
of
substitute
goods,
conventional
statements
of
an
individual's
demand
function
often
assume
that
there
is
a
price
at
which
the
services
of
a
site
will
not
be
demanded.
With
a
statedependent
demand
specification
it
is
unlikely
that
the
reasons
why
the
site
will
not
be
demanded
can
be
fully
explained.
Rather,
this
specification
~
used
simply
to
reflect
a
different
set
of
preferences
that
depend
on
the
exist­
.
 
.
 
.
ence
of
demand
for
the
site.
To
emphasize
this
assumption,
the
following
review
summarizes
the
difference
betw­
een
the
consumer's"
allocation
decisions
(
among
contingent
claims)
and
the
definition
of
risk
aversion
under
the
two
frameworks­­
one
that
maintains
the
uniqueness
postulate
and
one
that
does
not.

*
The
four
postulates
are:
(
1)
ordering
and
preference
direction­­
larger
incomes
are
preferred
to
smaller
incomes;
(
2)
certainty
equivalence­­
there
is
an
amount,
the
certainty
equivalent,
that
is
intermediate
in
size
to
the
largest
and
smallest
consequences
of
a
given
prospect;
(
3)
independence­­
a
c!
aim,
designated
as
Z,
can
be
s~
bstituted
for
its
preference
equivalent,
say
Z,
in
any
prospect
into
which
Z
enters
and
vice
versa;
and
(
4)
uniqueness­­
the
certainty
equivalent
of
a
prospect
depends
only
on
the
magnitudes
of
the
probabilities
and
incomes,
not
on
their
state
designations.
See
Hirshleifer
{
1970,
pp.
219­
20]
or
Malinvaud
[
1972,
pp.
285­
90]
for
further
discussion.
Cook
and
Graham
[
1977]
provide
additional
perspective
for
irreplaceable
goods.

5­
4
I
!

II
.
~
on~
idef'
the
case
of
two
continent
commodities
(
or
claims),
XI
and
X
2
,
to
States
1
and
2
and
having
probabilities
of
P
a
n
d
(
l
­
P
)
,
re­
~
orre5pondingIf
the
prices
of
these
claims
are
rl
and
r
2,
and
if
utility
is
de­
~
Pectivel
Y"
the
amount
of
X.,
such
as
u(
X.
),
the
individual's
objective
func­
~
endent
`
n
When
the
uniqueness
~
ostu
late
is
satisfied,
can
be
written
as
Equation
tion
I
~
5.1):

v
=
PP(
X1)

~
here
v
is
the
expected
u
t
i
l
i
t
y
.
I
f
~
z),
the
budget
constraint
limiting
the
+
(
1­
p)
p(
xz)
,
(
5
.
1
)

the
initial
endowment
of
claims
is
(~
1,
individual's
choices
would
be:

Y
=
rlX1
+
r2X2
=
r
l
X
1
+
r
2
X
2
.
(
5
.
2
)

Maximizing
Equation
(
5.1
)
subject
to
Equation
(
5.2)
and
solving
the
first­
~
rder
conditions
yields
the
familiar
equality
of
relative
prices
and
probabilityweighted
marginal
utilities,
as
in
Equation
(
5.3)*:

P
(
1
.5
(
1­~~
uT(
X2)
r
2
"
(
5
.
3
)

is
usually
specialized
further
by
consideration
of
a
"
fair"
T
h
i
s
result
gamble
case
(
i.
e.,
where
`
p
dX
1
+
(
1
­
p)
dX2
=­
O).
This
case
implies
the
equality
of
the
probability
ratio
and
the
price
ratio
for
the
two
contingent
claims
(
i.
fe.
,
p/(
1
­
p)
=
rl/
r2).
t
Using
this
condition,
Equation
(
5.3)
can
be
rewritten
as:
.

The
optimal
allocation
calls
for
equal
claims
in
X
1
and
X
2,
as
point
R
in
Figure
5­
1.
Thus,
the
selection
in
this
case
will
fall
tainty
locus
(
both
income
and
utility
)­­
the
45
°
line
in
Figure
5­
1.
(
5
.
4
)

given
by
the
along
the
cer­

The
traditional
definition
of
risk
aversion
for
this
framework
maintains
that
risk­
averse
individuals
require
better
than
"
fair"
gambles
before
they
will
select
these
alternatives
over
a
certain
claim
with
the
same
expected
income
Under
the
assumption
of
uniqueness
there
are
two
further
implications
*
The
second­
order
conditions
are
d2X2/
dX12
>
0.
This
can
be
shown,
given
uniqueness,
to
be
implied
by
the
assumption
of
concavity
of
U
(.).
That
is:
d2X2/
dX12
=

a/
axl
(
dX2/
dXl)
+
a/
ax2
(
dX2/
dXl)
[
dX2/
dXl]
,
w
h
e
r
e
dX2/
dXl
=
­[
P/(
l­
p)
l
O[
U'(
Xl)
/
U'(
X2)]
h
e
n
c
e
d
2
X
2
/
dX12
=
p
U
"
(
X
l
)
/
(
1­
p)
U'(
X2)
­
p2(
u'(
Xl))
2u''(
X2)
/
(
1­
p)
2(
u'(
X2))
3.
Concavity
of
u(.
)
implies
that
u"(.
)
<
0,
and
thus
dX22/
dX12
is
positive,
because
p,
(
1­
p),
U'(
Xl),
and
u'(
X2)
are
all
positive.

tThis
conclusion
is
derived
by
recognizing
the
implications
of
a
constant
initial
budget
and
the
"
fair"
gamble
for
selections
of
contingent
claims:
A
Constant
budget
implies
rldXl
+
r
2
d
X
2
=
O;
a
fair
gamble
implies
pdXl
+
(
1­
p)
dX2
=
O;
thus,
a
fair
gamble
implies
­
dX2/
dXl
=
p/(
1­
p)
=
rl/
r2.

5
­
5
income
Certainty
Loou
Utility
Certainty
Locus
.

tan
(
a)
=
rllrz
=
pll­
p
x,

Figure
5­
1.
Optimal
allocation
of
choice
with
contingent
claims.

associated
with
risk­
averse
behavior.
They
are
important
because
they
provide
the
means
for
explaining
the
divergence
between
Schmalensee
[
1972]
and
Bohm
[
1975]
in
their
respective
interpretations
of
the
appropriate
definition
of
risk
aversion.
To
understand
these
diverqent
interpretations,
imagine
a
riskaverse
individual
subject
to
the
choice
of
X
with
certainty
versus
th~
prospect
of
Xl
with
probability
p
and
X
2
with
probability
(
1
­
p).
Assume
X
=
pXl
+
(
1
­
p)
x~.
Then
a
risk­
averse
individual's
choice
would
be
consistent
with
a
utility
function
that
ranks
these
prospects
as
follows:

u(
i)
~
pu(
xl)
+
(
1
­
P)
U(
X2)
(
5.5)

Equation
(
5.5)
will
be
realized
if
U
(.
)
is
concave.
Thus,
the
concavity
of
U(.
)
is
usually
taken
to
imply
risk
aversion.
In
this
study's
analysis
of
"
fair"
gambles,
as
given
in
Equation
(
5.4),
the
risk­
averse
individual's
choices
can
also
be
characterized
as
implying
an
allocation
of
resources
among
claims
such
a
s
U'(
Xl)
=
U'(
XZ).
All
individuals
will
allocate
their
resources
among
claims
to
States
1
and
2
so
that
these
marginal
utilities
are
equalized
in
the
case
of
"
fair"
prices.
Since
risk
aversion
is
defined
by
the
concavity
of
U
(
.
),
the
behavioral
responses
of
a
risk­
averse
individual
will
be
determined
by
how
he
responds
to
a
change
in
p.
However,
once
the
assumption
of
uniqueness
is
relaxed
and
state­
specific
utility
functtons
are
permitted,
the
condition
for
fair
gambles
implies
only
that
the
marginal
utilities
will
be
equalized
and
not
that
either
the
total
utilities
or
the
total
monetary
claims
in
each
state
will
be
equalized.
Without
uniqueness
there
will
be
a
distinction
between
the
locus
of
equal
consumption
(
or
income)
over
states
(
i.
e.
,
the
45
°
Ilne
defined
as
the
income
and
utility
"
certainty"
locus
under
the
assumption
of
uniqueness)
and
the
utility
certainty
locus,
where
UI(
XI
)
=
U2(
X2),
as
illustrated
in
Figure
5­
2.
Moreover,
the
optimal
allocation
will
not
necessarily
lie
on
the
utility
certainty
locus
as
it
did
under
the
assumption
of
uniqueness.
Schmalensee
[
1972]
mis­

5­
6
X
2
Income
Cartainw
tan
(
a)
=
r
l
k=
=
PII.
P
x,

Figure
5­
2.
Optimal
allocation
of
choices
of
contingent
claims
without
uniqueness.

interpreted
this
possibility
as
an
indication
that
concavity
was
an
inappropriate
definition
of
risk
aversion
and
selected
the
equality
of
marginal
utilities
as
­­­­
the
characteristic
necessarv
to
define
risk­
averse
behavior
in
the
case
of
.
..­
State­
dependent
utility
functions.
In
provides
an
analytical
vehicle
that
will
ings
of
option
value
that
have
developed
5.3
OPTION
VALUE:
THE
"
TIMELESS"
summary,
the
contingent
claims
model
aid
in
deciphering
the
misunderstanding
the
research
literature.

ANALYSES
The
first
analytical
evaluations
of
option
value
employed
a
"
timeless"
framework
with
the
only
source
of
uncertainty
associated
with
the
state
of
the
individual's
preference
structure
(
see
Cicchetti
and
Freeman
[
1971
],
Bohm
(
1975],
and
Schmalensee
[
1972,
1975]).
To
simplify
the
explanation
of
these
analyses,
assume
that
individual
preferences
can
be
described
by
just
two
states:
State
1,
which
demands
the
services
of
the
asset
with
UI
(.
),
and
State
2,
which
`
does
not
demand
the
services
of
the
asset
with
U2(­.
).
Each
State's
utility
function
will
have
two
arguments
­­
income,
Y,
and
a
variable
indicating
access
to
the
asset's
services
able
and
a
implying
they
are
not.
,
with
d
implying
the
services
are
avail­
This
argument
can
proceed
using
the
compensating
variation
definitions
of
consumer
surplus,
option
price,
and
option
`
aiue?
but
comparable
arguments
can
be
developed
using
equivalent
variation.

5­
7
P
?;
I
!

Equations
(
5.6)
and
(
5.7)
define
consumer
surplus
for
the
I.
th
state
(
SCi)
and
option
price
(
OP),
respectively:

ui(
Yi
­
S
C
i,
d)
=
ui(
Yi,
~),
i=
l,
2
(
5.6)

2
2
~
TCi
Ui(
yi
­
OP,
d)
=
Z
Hi
Ui(
yi,
~)
(
5.7)
i=
l
i=
l
where
Ui(
y,
d)
=
individual
utility
for
State
i
with
income
Y
i
and
with
access
to
the
services
of
the
asset
n.
I
=
probability
of
utility
State
i
(
nz
=
1
­
Kl).

Substituting
Equation
(
5.7)
in
Equation
(
5.6)
and
rearranging
terms
gives:

2Z
n
i
[
ui(
Yi
­
OP,
d)
­
ui(
Yi
­
SC
i,
d)]
=
O
.
(
5.8)
i=
l
Schmalensee
[
1972]
proposed
using
concavity
of
the
state­
specific
utility
functions
to
expand
Equation
(
5.8).
That
is,
the
inequalities
given
in
Equations
(
5.9)
and
(
5.10)
hold
for
concave
ui(.
):*

ui(
Yi
­
O
P
,
d
)
­
Ui(
yi
­
S
C
i,
d
)
>
(
S
C
i
­
Op)
[~
ui/~
Yi
(
Yi
­
Op,
d)]
(
5.9)
 
u
i(
Y
i
­
OP,
d)
­
Ui(
yi
­
S
C
i,
d)
<
(
SCi
­
O
P
)
[
L%
Ji/
ayi
(
Yi
­
Sci,
d)].(
s.
lo)

Substituting
each
into
Equation
(
5.8)
and
rearranging
terms
gives
inequalities
for
option
price
involving
Bohm's
[
1975]
weighted
expected
consumer
surplus
terms
as
Equations
(
5.11)
and
(
5.12):

OP
>
:
71i
Sc
i
[
aui/
aYi
(
Y.
­
OP,
d)]
/
~
ni[~
ui/
8Yi
(
Y.
­
OP,
d)]
.
(
5.11)
 
i=
l
I
i=
l
I
2
OP
<
~
ni
Sci
[
aui/
ayi
(
Yi
­
Sc
i,
d)]
/
Z
ni[~
ui/
tlYi
(
Yi
­
S
C
i,
d
)
]
.
(
5
.
1
2
)
 
i=
l
i=
l
Because
option
value
(
OV)
is
defined
as
the
difference
between
the
option
price
(
OP)
and
the
expected
consumer
surplus
(
SC)
­­
i.
e.
,
OV
=
OP
*
ln
the
analysis
that
follows,
the
point
of
evaluation
of
the
partial
derivatives
will
be
important
to
the
interpretation
given
to
each
relationship.
Therefore
[
8u./~
Y
(
a,
b)]
will
refer
to
the
partial
derivative
of
U
i(.
)
with
respect
to
Y
evaluatkd
at
the
point
(
a,
b).

5­
8
­
f%)
­­
these
inequalities
offer
the
potential
for
determining
the
sign
\
i=
l
/
value
if
it
is
possible
to
relate
the
weighted
consumer
surplus
to
of
the
Option
value
of
the
consumer
surplus.
*
Schmalensee's
definition
of
risk
the
expected
aversion
as
walW
of
­
the
mar9inal
utilities
of
income
across
states
(
i.
e.,

8UJ8YI
=
au2/~
Y2)
provides
the
ability
to
make
this
association
by
making
the
Equations
(
5.11)
or
(
5.12)
unity.
That
is,
depending
upon
whether
~
eightS
in
the
equalitY
is
realized
at
Y.
­
OP
or
Y
.
­
SC.,
option
price
will
be
greater
or
expected
consume}
surplus.
`
Thus,
i
Schmalensee
concludes
that
the
le55
than
~
i~
n
of
option
value
depends
on
the
point
of
evaluation.

AS
observed
earlier,
Bohm
has
correctly
observed
that
this
judgment
is
misleading
for
at
least
two
reasons.
First,
the
interesting
expression
is
Equation
(
5.
II)
because
the
point
of
evaluation
of
the
marginal
utilities
correctly
.
.
.
asSi9ns
`
0
the
`
nd'vldua!
`
he
.
re'evant
income/
access
conditions.
This
expression
describes
the
relationship
between
option
price
and
expected
consumer
*
TO
illustrate
this
pOint
let
WI
=

W2
=
nl
~
(
Y
I­
OP,
d)
aY
~

2
au.
ZnilJ+
(
Yi­
OP,
d
)
i=
l
i
~
2
~
(
Yz­
OP,
d
)

2
au.

This
specification
be
rewritten
as
hti+
(
Yi­
OP,
d)
i=
l
i
will
imply
WI
+
W
2
=
1.
Consequently,
Equation
(
5.11
)
can
2
OP
>
z
W
i
sc
i
 
i=
l
(
2
To
compare
the
specification
with
the
expected
consumer
surplus
2
Xi
Sc
i
i=
l
requires
some
knowledge
about
the
relationship
between
W
i
and
Hi.
For
exam­
Ple,
if
it
is
assumed
that
~
(
Y
l
­
OP,
d)
=
!%
(
y
2
ayl
a~=
­
OP,
d)
(
the
marginal
utilities
of
income
are
equal
in
each
period),
then
W
i
=
ni
and
Equation
(
5.11
)
allows
option
value
to
be
signed.

5­
9
surplus
when
the
individual's
income
is
reduced
suggested:

We
are
asking
him
how
much
he
can
abstain
OP
and
enter
the
future
state,
whatever
it
income
of
Y­
OP
without
being
worse
off.
by
the
option
price.
As
Bohm
from
today
in
terms
of
an
may
be,
with
a
disposable
He
will
be
at
Y­
SC,
oniy
if
he
does
not
pay
an
option
price­­
and
that
is
another
stor~.
We
 
do
not
ask
him
­
about
the
maximum
amount
he
is
willing
to
pay
~­
~
de~
he~
oe~
ot~
t=
amount.
 
.
(
Bohm
[
1975],
p.
735)
 
 
 
.

The
aversion.
aversion,
taneously
.

.
s
e
c
o
n
d
c
o
n
s
i
d
e
r
a
t
i
o
n
i
n
v
o
l
v
e
s
t
h
e
Schmalensee
definition
of
risk
The
previous
section
noted
that
the
conventional
definition
of
risk
with
the
uniqueness
assumption
for
state
utility
functions,
simul
­
impiies
that:

The
utility
function
must
be
concave
to
admit
such
a
response
to
a
"
fair"
gamble.

In
response
to
a
fair
gamble
the
risk­
averse
individual
will
always
"
select
a
point
w­
here
marginal
utilities
of
income
are
equal.

This
latter
point
is
a
result
of
optimizing
behavior
in
the
presence
of
a
fair
gamble
and
concavity
of
the
utility
functions.
Once
the
uniqueness
assumption
is
relaxed
and
state­
specific
utility
functions
are
permitted,
the
only
plausible
definition
for
risk
aversion
is
by
the
concavity
of
the
state­
specific
utility
functions.
Thus,
when
the
correct
point
of
evaluation
(
i.
e.,
the
inequality
given
in
Equation
[
5.11
]
)
and
the
appropriate
definition
of
risk
aversion
are
used,
the
sign
of
option
value
cannot
be
established.
It
may
be
positive,
negative,
or
zero
depending
upon
the
relationship
between
the
marginal
utilities
of
income
at
each
state.

Given
these
conclusions,
how
do
Cicchetti
and
Freeman
[
1971]
establish,
apparently
unambiguously,
a
positive
sign
for
option
value
while
Bohm
does
not?
To
answer
this
question,
return
to
the
example
of
a
"
fair"
gamble
with
state­
specific
utility
functions
that
was
given
in
Figure
5­
2.
Schmalensee
incorrectly
interpreted
this
divergence
to
indicate
the
inadequacy
of
u.(.
)'
s
concavity
as
the
sole
basis
for
defining
risk­
averse
behavior.
How&
er,
Cicchetti
and
Freeman
apparently
intended
to
focus
on
a
comparison
along
the
utility
certainty
locus.
*
As
Anderson
[
1981]
has
recently
observed,
they
as­

*
Cicchetti
and
Freeman
seem
to
have
wanted
to
use
t
h
e
u
t
i
l
i
t
y
certaitlty
locus
to
make
the
state­
specific
actions
commensurate.
This
can
be
seen
in
their
proposal
that:

To
make
the
choice
problem
solvable,
there
must
be
some
way
of
making
the
utilities
of
the
two
alternative
mappings
commensurable
We
have
proceeded
as
follows
to
derive
a
rule
for
comparing
the
utilities
from
the
two
alternative
mappings.
For
any
level
of
disposable
income,
if
the
individual
did
not
demand
the
5­
1o
~
dmed
that
`
h
e
individual's
income
was
equal
across
all
states
and
that,
when
equal,
total
utilities
in
each
state
were
also
equal
at
the
preferred
iflco~
e
was
In
the
present
analysis,
this
would
correspond
to
equal
utility
vector.
~
r;
ce
of
access
to
the
resource
[
i.
e.,
ui(
Yi,
d)
=
Uj(
Yj,
d)
for
Y
=
~
Or
conditions
i
Unfortunately,
the
Cicchetti­
Freeman
analysis
did
not
correctly
de­
Yjl"
an
individual's
choices
of
Y
and
the
services
of
the
asset.
While
they
scribe
proposed
to
consider
a
discrete
choice
similar
to
the
d
versus
6
description,

~
bey
represented
the
services
as
continuously
available,
designed
by
X.
,

figures
5­
3
and
5­
4
reproduce
the
Cicchetti­
Freeman
figures
(
I
I
I
and
IV)
analysi
S.
If
Figure
5­
4
is
interpreted
as
an
illustration
of
the
"
no­
fOr
the
the
assertion
that
us
~
emafld"
case,
=
U
5
at
Y.
is
incorrect.

budget
constraint,
If
the
relevant
Bl,
is
considered,
the
individual
will
not
choose
to
consume
level
of
Y.
In
the
"
no­
demand"
case
(
i.
e.
,
u8)~
the
selected
income
the
same
~
ill
be
Yo,
but
the
"
demand"
case
will
be
Y5
in
Figure
5­
3.
Similar
arguments
can
be
developed
for
the
assumption
that
UI
=
U'
6
at
Y.
­
OPI,
which
indicates
that
the"
construction
of
Figure
5­
4
is
incorrect.
To
adequately
deal
with
the
equivalence
of
state­
specific
utility
functions
at
equal
income
levels,
a
graphical
~
nalYSiS
must
be
in
terms
of
indirect
utility
functions
as
described
by
Bishop
(
1982].
In
this
case
the
ambiguity
in
the
sign
of
option
value
is
clearly
demonstrated.

]
n
Graham's
[
1981
1
recent
attempt
to
use
the
Schmalensee
framework
to
~
omment
on
the
appropriate
treatment
of
option
value,
he
argues
that
the
reasonableness
of
usin9
oPtion
Price
for
benefit­
cost
analyses
will
depend
on
the
nature
of
the
problem
under
study.
More
specifically,
Graham
concluded
that:

.
Option
price
is
the
appropriate
benefit
measure
for
project
analysis
when
one
can
assume
the
individuals
affected
are
similar
and
they
all
experience
the
same
risk.

.
Expected
willingness
to
pay
will
be
the
appropriate
measure
for
those
cases
with
similar
individuals
but
with
risks
specific
to
each.

These
conclusions
are
derived
using
a
generalization
of
the
option
price
definition
(
Equation
(
5.7)).
To
understand
them,
Graham's
arguments
must
be
considered
in
detail.
For
the
case
of
individual
risks,
he
assumes
that
payments
may
be
state
specific.
This
is
equivalent
to
the
assumption
that
a
complete
set
of
markets
for
contingent
claims
exists.
Under
these
assumptions,
the
definition
of
option
price
in
Equation
(
5.7)
would
be
replaced
by
Equation
(
5.13):

good,
he
would
choose
a
consumption
point
on
the
Y
axis
and
experience
a
certain
level
of
utility;
if
he
were
to
demand
the
good
(
assuming
that
it
is
available),
he
would
choose
a
tangency
point
on
the
budget
line
associated
with
that
point,
and
experience
a
given
level
of
utility.
We
assume
that
the
alternative
outcomes
have
the
same
utility.
(
Cicchetti
and
Freeman
[
1971],
p.
534)

5­
11
Y
I
OP*
2
.

.
..
 
 
Y.

I.­
 
­

YO­
OP1
`
6
­­
 
 
 
LL**
`
\
B
1
;;

I
\
*
x
Figure
5­
3.
Option
value
in
Cicchetti­
Freeman's
analysis.

/
l=
f(
Y)
ps
=
~,
 
 
 
 
 
 
 
 
 
.
 
 
PI*
=
P3
­
 
 
 
 
 
 
PI
=
P6
,
 
 
 
YO­
OP1
P
Y.
­
oP**
YO­'
AOP1
Y.
Y
I
I
I
I
I
I
I
i
Figure
5­
4.
Option
value
in
Cicchetti­
Freeman's
analysis
with
"
no
demand."
7
Graham
case
of
2
2
2
fii
Ui(
yi
­
Pi,
d)
=
I
ni
Ui(
yi,
~)
.

i=
l
i=
l
(
5.13)

defines
this
relationship
as
the
willingness­
to­
pay
locus.
The
special
PI
=
P~=
OP
would
yield
the
conventional
definition
of
the
option
price.
The
IOCUS
EdSO
includes
the
point
where
Pi
=
SC
i
(
by
construction),
as
~
ell
as
the
fair­
bet
and
the
utility
certainty
points,
as
illustrated
in
Figure
5­
5
.

To
illustrate
some
of
the
points
on
the
locus;
assume
TT'
corresponds
to
~
he
individual's
bud9et
constraint
where
the
Prices
of
claims
in
States
d
and
d
correspond
to
the
probabilities
of
each
state.
F
will
then
designate
the
fair­
bet
pOint.
When
payments
are
constant,
regardless
of
the
state
of
nature
as
with
point
P
,
the
locus
describes
the
willingness
to
pay
under
institutional
COnd~
tiOnS
consistent
with
an
option
price,
O
P.
Point
S
corresponds
to
the
coordinates
of
the
consumer
surpluses
for
each
state.
To
calculate
the
expected
value
of
the
consumer
surplus,
the
budget
constraint
through
S
parallel
to
TT'
is
used
(
to
reflect
the
state
probabilities).
The
intersection
of
this
new
budget
line,
RR',
with
the
45
°
line
defines
the
expected
consumer
surplus.
For
this
example,
option
value
is
positive.

s
&
It,
d
I
\\
R
\
\
\
s
\
\
\
\

EM
OP
$
In
Smt9a
Figure
5­
5.
Option
value
with
mntingent
claims
in
Graham's
analysis.

5­
13
Aggregating
the
willingness­
to­
pay
loci
across
individuals,
Graham
argued
that:

Justification
of
the
project
hinges
upon
the
question
of
whether
or
not
contingent
prices
exist
at
which
aggregate
willingness
to
pay
in
each
state
exceeds
the
corresponding
resource
cost
of
the
project.
Shouid
such
prices
exist,
that
point
from
an
individual's
locus
which
has
the
greatest
value
at
these
prices
is
the
one
relevant
for
cost­
benefit
analysis,
and
the
corresponding
value
at
these
prices
is
the
appropriate
measure
of
benefit.
(
Graham
[
1981
],
p.
719)

To
apply
this
approach
in
particular
examples
requires
that
one
distinguish:
(
1)
the
benefits
realized
as
a
result
of
moving
from
an
initial
distribution
of
income
to
another
that
assures
an
efficient
distribution
of
risk
and
(
2)
the
benefits
resulting
from
the
project
itself.

Graham's
conclusions
are
based
on
two
rather
special
cases.
The
first
of
these
avoids
the
issue
of
an
inefficient
distribution
of
risk
by
assuming
that
individuals
are
alike
and
that
they
face
identical
risks.
The
second
case
also
skirts
this
issue
by
assuming
the
existence
of
either
complete
contingent
claims
markets
or
an
ideal,
state­
dependent
tax
collection
scheme
(
tied
to
the
project
under
evaluation).
In
either
case,
an
efficient
distribution
of
risk
will
be
realized.
Of
course,
neither
of
these
sets
of
assumptions
is
plausible
in
most
applications,
where
some
attempt
must
be
made
to
include
a
measure
of
the
value
of
an
option
to
use
the
services
of
an
environmental
resource.
Consequently,
as
Graham
acknowledges,
one
is
left
with
option
price
as
the
"
best"
basis
for
measuring
benefits.
Thus,
for
practical
purposes,
Graham's
analysis
has
strengthened
Bohm's
conclusion:
Option
price
is
the
relevant
focus
for
applied
welfare
economics.

Given
these
conclusions,
why
worry
about
the
sign
and
magnitude
of
option
value?
One
pragmatic
reason
arises
with
the
difficulty
in
measuring
each
individual's
option
price.
If
it
is
possible
for
wide
classes
of
assets
and
their
associated
prospective
users
to
demonstrate
that
the
corresponding
option
values
of
the
assets
would
be
positive,
one
would
be
safe
in
assuming
that
measures
of
the
expected
user
benefits
(
i.
e.
,
as
derived
from
an
"
ideal"
consumer
surplus
calculation)
would
understate
the
total
benefits
provided
by
the
asset.
*

5.4
THE
TIME­
SEQUENCED
ANALYSES
Time­
sequenced
evaluations
of
option
value
offer
more
specific
answers
to
the
three
questions
raised
at
the
outset.
That
is,
these
analyses
provide
an
explicit
statement
of
the
relationship
between
decisions
over
time.
In
general
the
uncertainty
is
supply
related.
It
is
resolved
with
the
passage
of
time,
and
decisions
cannot
be
altered.
The
first
of
these
models
was
devel
­

*
This
argument
ignores
the
potential
role
of
existence
values
as
de­
`
scribed
by
Krutilla
[
1967]
and
more
recently
discussed
by
Freeman
[
1981]
.

.

5­
14
 
Oped
by
A
r
r
o
w
a
n
d
F
i
s
h
e
r
[
1
9
7
4
]
,
whose
framework
introduced
a
time­
~
equencin9
of
decisions
and,
as
a
result,
assumed
there
was
a
resolution
of
the
uncertainty
facing
the
decision
process
with
the
passage
of
time.
Their
model
considered
decisions
to
develop
or
preserve
a
fixed
amount
of
land.
~
eci5ions
to
develop
some
fraction
(
or
all)
of
the
land
were
irreversible.
Therefore
l
any
information
acquired
with
the
passage
of
time
could
affect
only
the
decisions
made
on
the
remaining
stock
of
preserved
land.

Arrow
and
Fisher's
quasi­
option
value
can
be
interpreted
as
the
expected
value
of
the
information
obtained
through
delay,
as
has
been
suggested
by
Conrad
[
1980]
and,
indeed,
acknowledged
earlier
by
Krutilia
and
Fisher
[
19751
in
their
overall
evaluation
of
special
problems
associated
with
allocation
decisions
involving
unique
natural
environments.
For
example,
Krutilla
and
Fisher
observed
that:

The
key
new
element
in
Arrow
and
Fisher
is
a
Bayesian
information
structure.
The
passage
of
time
results
in
new
information
about
the
benefits
of
alternative
uses
of
an
environment,
which
can
in
turn
be
taken
into
account
if
a
decision
to
devote
it
to
development
is
deferred.
Since
development
is
not
reversible,
once
a
decision
to
develop
is
made,
it
cannot
be
affected
by
the
presence
of
new
information
which
suggests
that
it
would
be
a
mistake
in
the
future.
The
main
result
of
the
analysis
is
then
that
there
is
an
option
value,
or
quasi­
option
value,
to
refraining
from
development­­
even
on
the
assumption
that
there
is
no
risk
aversion,
and
only
expected
values
matter.
(
Krutilla
and
Fisher
[
1975],
pp.
70­
71)

Conrad
also
suggested
that
option
value
could
be
interpreted
as
the
expected
value
of
perfect
information.
In
so
doing,
he
implicitly
maintains
that
over
time
one
progressively
learns
of
and
resolves
the
uncertainty.
However,
his
conclusion
is
correct
only
if
it
is
regarded
as
the
only
appropriate
translation
of
the
"
timeless"
analysis
of
option
values
into
a
time­
sequenced
decision
process.
Henry
[
1974]
has
drawn
a
similar
conclusion
in
his
evaluation
of
the
importance
of
this
transition,
noting
that:

The
relationship
so
established
between
risk
aversion
and
optionprice
appears
rather
obvious
when
it
is
viewed
as
being
encountered
in
a
`
timeless
world'
where
I
[
the
individual]
has
one
and
only
one
decision
to
take;
in
a
world
of
this
type
any
decision
is
..
 
 
 
@
~
irreversible
~
any
other
[
emphasis
added]
and
it
is
impos:
sible
to
introduce
Krutilla's
option
value
which
is
nothing
but
a
risk
premium
in
favour
of
lirreplaceable
assetsl
.
Krutilla's
idea
can
only
be
examined
in
a
`
sequential
world'
where
~
[
the
individual]
really
has
a
succession
of
decisions
to
take.
(
Henry
[
1974],
p.
92)
 
­
 
Thus,
if
it
is
assumed
that
uncertainty
is
resolved
over
time,
that
the
asset
under
consideration
is
in
some
respect
irreplaceable,
and
that
the
decisions
are
made
sequentially
with
the
benefit
of
the
acquired
information,
there
is
clearly
a
positive
option
value.
If,
on
the
other
hand,
the
resolution
of
t
h
e
uncertainty
is
not
allowed
as
a
part
o
f
a
s
e
t
o
f
d
e
c
i
s
i
o
n
s
,
o
p
t
i
o
n
v
a
l
u
e
will
be
a
reflection
of
risk
aversion,
and
its
sign
will
depend
on
the
nature
of
the
state­
specific
utility
functions.

5­
15
This
distinction
has
estimates
of
option
price.
important
implications
for
any
attempts
to
develop
If
a
direct
survey
or
contingent
valuation
method
is
used
to
obtain
these
estimates,
the
results
will
be
based
on
hypothetical
conditions
in
which
it
is
unlikely
that
respondents
can
be
given
a
means
of
obtaining
information
and
reacting
to
it.
That
is,
as
a
practical
matter,
it
is
probably
safe
to
assume
that
questions
designed
to
elicit
an
individual's
option
price
will
not
be
posed
in
a
way
that
identifies
mechanisms
through
which
the
individual
can
obtain
information
and
alter
decisions
based
on
it.
Thus,
the
timeless
analyses
are
more
likely
to
be
the
relevant
models
for
understanding
the
empirical
measurement
of
option
value.
However,
this
judgment
does
not
imply
that
a
careful
description
of
the
source
of
uncertainty
and
the
means
through
which
it
is
resolved
can
be
ignored
in
question
design.
Rather,
it
simply
recognizes
that
formulating
questions
that
acknowledge
the
prospects
for
learning
and
that
offer
mechanisms
for
enhancing
learning
would
likely
increase
the
complexity
of
the
instrument
to
a
point
where
it
was
not
usable.

Together
with
extensions
of
it
in
Smith
[
1983],
this
analysis
suggests
that
supply
uncertainty
can
be
important
to
the
sign
of
option
value
in
a
timeless
framework.
Accordingly,
supply
uncertainty
should
be
acknowledged
and
explicitly
identified
in
questionnaires
designed
to
measure
option
price.

5.5
RECENT
ESTIMATES
OF
NONUSER
VALUES
There
appears
to
have
been
only
one
published
study
estimating
option
values.
This
study
by
Greenley,
Walsh
,
and
Young
[
1981
]
attempts
to
measure
the
option
value
for
the
recreational
use
of
preserved
water
quality
in
the
South
Platte
River
basin
in
Colorado.
These
authors
used
two
payment
vehicles
­­
an
increment
to
the
sales
tax
and
an
increase
in
the
monthly
watersewer
fee­­
in
a
survey
of
a
random
sample
of
202
residents
of
Denver
and
Fort
Collins.
Their
study
attempted
to
estimate
specific
components
of
the
benefits
of
maintaining
water
quality,
including
option,
user,
existence,
and
bequest
values.
Their
paper
focuses
on
the
results
of
the
question
for
option
value.
Two
aspects
of
their
option
value
question
are
important.
First,
it
seems
to
be
eliciting
an
option
price,
not
option
value,
and
specifies
a
resolution
of
the
supply
uncertainty
associated
with
the
preservation
of
water
quality.
And,
second,
the
question
treats
the
two
payment
vehicles
differently
The
question
is
reproduced
below:

Given
your
chances
of
future
recreational
use,
would
you
be
willing
to
pay
an
additional
cents
on
the
dollar
in
present
sales
taxes
every
year
to
postpone
mining
development?
This
postponement
would
permit
information
to
become
available
enabling
you
to
make
a
decision
with
near
certainty
in
the
future
as
to
which
option
(
recreational
use
or
mining
development)
would
be
most
beneficial
to
you
.
Would
it
be
reasonable
to
add
to
your
water
bill
every
month
for
this
postponement?
(
Greenley,
Walsh,
and
Young
[
1981],
p.
666,
emphasis
added)

.

5­
16
L
AS
discussed
earlier,
option
value
is
the
difference
between
an
individuals
option
price
and
his
expected
consumer
surplus.
It
would
seem
that
this
question
is
soliciting
the
option
price.
Unfortunately,
the
authors
interpreted
the
responses
as
measures
of
the
option
value
and
asked
a
separate
question
to
obtain
user
values.
T
h
e
Greenley,
Walsh,
and
Young
results
with
intended
the
sales
tax
payment
vehicle
indicate
an
average
option
value
of
approximately
~
23.00
per
year
(
with
the
water
fee
payment
vehicle,
it
was
$
8.90
).*

The
interpretation
of
these
results
has
been
somewhat
controversial
.
Both
qUeStions
used
by
Greenley,
Walsh,
and
Young
seem
to
be
asking
for
an
­­
the
first
under
a
timeless
interpretation
and
the
second
under
a
option
prim
time­
sequenced
format.
Greenley,
Walsh,
and
Young
interpret
one
as
a
measure
of
exPected
consumer
surPlus
and
the
other
as
option
value.
Mitchell
and
Carson
[
1981
1
appear
to
have
been
the
first
to
question
the
interpretation
of
the
Greenley,
Walsh,
and
Young
questions.
While
Mitchell
and
Carson
did
not
relate
their
criticisms
to
the
two
conceptions
of
option
value,
they
did
argue
that
both
questions
measure
option
price.
Moreover,
they
suggested
that
the
Greenley,
Walsh,
and
Young
results
indicate
the
possibility
of
a
starting
point
bias,,
based
on
the
differences
in
designated
starting
points
used
for
each
payment
vehicle.
in
a
recent
unpublished
response
to
the
Mitchell­
Carson
comments,
Greenley,
W
a
l
s
h
,
and
Young
[
1983]
argue
that
the
interviewing
process
itself
prevented
interpretation
of
the
questions
as
requesting
option
price.
They
observe
that:

Some
confusion
may
arise
when
expected
consumer
surplus
and
option
value
questions
are
taken
out
of
the
context
in
which
they
are
used
because
they
often
take
the
same
general
form
as
questions
asking
for
option
price.
.
.
.
The
important
distinction
in
this
case
[
their
study]
is
that
a
population
of
users
was
first
asked
to
estimate
their
expected
consumer
surplus,
and
in
addition,
a
separate
estimate
of
option
value.
They
were
informed
that
these
are
separate
and
distinct
values,
and
provided
the
opportunity
to
adjust
values
previously
reported.
The
respondents
provided
well­
focused
estimates
for
each
question.
We
conclude
that
the
procedures
employed
in
our
study
capture,
reasonably
accurately,
the
values
necessary
to
assess
the
recreational
benefits
of
improved
water
quality.
"
(
Greenley,
Walsh,
and
Young
[
1983]).

While
this
may
be
the
case,
no
explanation
is
offered
of
why
the
households
adjust
their
two
bids.
If
each
is
measuring
what
the
authors
intended,
there
would
be
no
basis
for
adjustment.
Equally
important,
one
can
judge
the
responses
to
a
contingent
valuation
experiment
based
only
on
the
questions
posed.
If
they
are
not
clearly
connected
to
the
concept
desired,
there
is
reason
to
question
whether
informal
discussions
between
the
interviewer
and
respondent
will
assure
understanding.
Finally,
our
evaluation
of
the
questions
(
in
contrast
to
Mitchell
and
Carson)
leads
to
the
conclusion
that
two
different
concepts
of
option
price
are
in
fact
asked.

*
lt
should
be
noted
bids­­
both
the
"
true"
zero
refused
to
participate
in
the
that
these
summary
statistics
include
all
zero
bids
and
the
zero
bids
of
those
individuals
who
bidding
game.

5­
17
.
.
.
 
 
 
­
_______
Of
course,
in
fairness
to
all
participants
in
the
exchange,
there
is
no
complete
record
of
exactly
what
the
interviewers
discussed
with
survey
respondents
Greenley,
Walsh,
and
Young's
[
1983]
recent
notes
on
the
Mitchell­
Carson
critique
suggest
that
they
were
aware
of
the
potential
ambiguity
in
their
questions.
What
is
at
issue
is
not
only
how
successful
the
interviewers
were
in
overcoming
it
but
also
that
the
terms
of
the
contingent
market
may
differ
for
each
respondent
(
because
of
the
interviewer
effect)
making
the
results
problematical.

The
second
empirical
study
focusing
on
user
and
nonuser
values
was
conducted
by
Mitchell
and
Carson
[
1981].
It
sought
to
measure
each
individual's
willingness
to
pay
for
cleaning
up
all
rivers
and
lakes
in
the
United
States
to
a
particular
level.
Since
individua~
were
not
classified
according
to
whether
or
not
they
used
these
water
resources,
the
responses
must
be
assumed
to
include
both
use
and
nonuse
values.
*
Indeed,
Mitchell
and
Carson
argue
that
it
is
beyond
the
capability
of
many
respondents
to
reliably
determine
separate
values
for
subcategories
of
water
quality
benefits.
Their
survey
was
based
on
a
national
probability
sample
of
1,576
individuals
and
was
conducted
as
part
of
an
opinion
poll
soliciting
these
individuals'
responses
to
other
questions
associated
with
environmental
attitudes.
This
study
introduced
the
water
quality
ladder
used
in
the
survey
conducted
for
the
present
study.
In
addition
it
assumed
that
the
household
payment
vehicle
was
through
higher
prices
and
taxes
(
the
same
vehicle
used
in
this
survey).
Four
versions
of
an
anchored
payment
card
were
used,
rather
than
an
iterative
bidding
framework.
They
were
differentiated
according
to
the
range
of
values
reported
on
the
cards
and
by
the
anchor
points
reported.
The
cards
were
distinguished
by
income
class
so
that
the
anchored
values
on
the
card
corresponded
to
the
average
of
the
actual
payments
made
by
members
of
each
incom­
e
group.
four
sets
of
anchor
points
used
in
this
study
were:

Version
A
Average
household
expenditures
(
through
taxes)
to
the
space
program,
highways,
public
education,
and
defense.

B
Same
four
public
goods
as
in
Version
A
plus
police
and
fire
protection.

c
The
same
four
public
goods
as
in
Version
A,
but
amounts
increased
by
25
percent
for
each
income
group
over
the
levels
used
with
Version
A.

D
The
same
four
public
goods
and
amounts
as
in
Version
A
plus
the
estimated
amount
for
water
pollution
control.
The
*
since
individuals
do
not
conceive
o
f
using
all
r
i
v
e
r
s
and
lakes
in
t
h
e
U
hi
ted
States,
it
must
be
assumed
that
only
a
sub=
t
of
these
can
be
considered
a
part
of
the
set
actually
used
or
planned
for
future
use.
To
the
extent
that
individuals
express
a
willingness­
to­
pay
bid
for
improved
water
quality
at
all
water
bodies,
they
are
expressing
expected
user
values,
any
option
values
(
associated
with
uncertain
future
use),
and
existence
values.

.

5­
18
A
Table
5­
1.
Summary
of
Mitchell­
Carson
Estimated
Mean
a
Annual
Willingness
to
Pay
by
Version
and
Water
Quality
Version
of
payment
card
~
a~
er
qualitY
A
(
274)
B
(
255)
C
(
244)
Cate90rY
$
168
$
133
$
161
Beatable
Fishable
$
214
$
180
$
198
swimmable
$
247
$
212
$
222
~
afhis
table
was
summarized
from
Mitchell
and
Carson's
[
1981
]
Table
5­
1,
p.

5­
3.
The
numbers
in
parentheses
are
the
numbers
of
respondents
providing
values
to
the
water
quality
questions
for
each
version
in
1980
dollars.

For
three
of
the
four
versions
of
the
payment
card,
Table
5­
1
reports
the
mean
estimates
for
boatable,
fishable,
and
swimmable
water
qualities.
*
While
this
study
provided
detailed
analysis
of
potential
survey
biases,
its
questions
relate
to
an
abstract
conception
of
the
impacts
of
a
water
quality
improvement
for
the
individual.
That
is,
while
the
water
quality
is
described
as
improving
to
levels
defined
by
the
activities­­
swimmable,
fishable,
and
beatable­­
the
quality
of
the
water
already
available
to
the
individual
is
unknown
If
the
water
bodies
available
to
the
individual
have
quality
levels
that
permit
the
full
range
of
his
desired
uses,
the
responses
might
be
expected
to
r
e
f
l
e
c
t
an
existence
value
for
all
other
sites.
By
contrast,
if
the
available
sites
for
water­
based
recreation
do
not
permit
all
or
some
subset
of
these
activities
the
responses
may
reflect
user
values.
Without
knowledge
of
these
sitespecific
features,
Mitchell
and
Carson
must
average
heterogeneous
responses.
That
is,
ideally,
the
responses
based
on
user
values
and
those
associated
with
nonuser
values
should
be
distinguished.
Moreover,
the
analysis
should
control
the
influence
of
the
differential
availability
to
individuals
of
sites
with
the
desired
water
quality.
The
Mitchell­
Carson
method
implicitly
assumes
all
individuals
will
benefit
equally
from
the
uniform
improvement
of
the
water
quality
at
all
sites.
This
may
not
be
correct.
The
benefit
realized
by
each
individual
will
depend
on
his
access
to
sites
with
the
desired
water
quality
before
the
change.

Mitchell
and
Carson
estimate
the
nonuser
benefits
of
water
quality
improvements
by
assuming
that
the
willingness­
to­
pay
responses
of
surveyed
*
The
effects
of
knowing
what
was
actually
paid
for
water
quality
control
(
i.
e.,
version
D)
were
also
reported
by
the
authors.
Forty­
seven
percent
of
t
h
e
3
5
4
r
e
s
p
o
n
d
e
n
t
s
t
o
v
e
r
s
i
o
n
D
said
they
were
willing
to
pay
t
h
e
a
m
o
u
n
t
shown
on
the
card
that
they
were
told
would
raise
water
quality
to
f
i
s
h
a
b
l
e
i
n
the
next
few
years.
For
further
details
on
these
results,
see
Mitchell
and
Carson
[
1981,
pp.
5­
6
to
5­
7].
The
figures
are
not
reported
here
since
they
reflect
only
that
some
people
were
willing
to
pay
at
least
these
amounts.
 
 
5­
19
individuals
who
did
not
engage
in
in­
stream
recreation
will
be
"
almost
purely
intrinsic
in
nature.
"
Even
if
this
reasoning
is
correct,
it
does
not
imply
that
nonuser
willingness
to
pay
will
be
a
reasonable
estimate
of
option
value.
It
may
include
existence
values
as
well.
Nonetheless,
based
on
this
logic,
39
percent
of
the
respondents
with
willingess­
to­
pay
data
reported
they
had
no
in­
stream
use
of
freshwater
in
the
past
2
years.
The
nonusers
mean
bid
for
fishable
water
was
$
111.
The
mean
bid
by
users
for
the
same
water
quality
change
was
$
237.
Hence,
by
these
estimates,
intrinsic
values
were
judged
to
be
approximately
45
percent
of
total
willingness
to
pay
of
users.

Rae
[
1981a,
1981b]
has
also
reported
estimates
of
option
price
for
"
clear"
visibility
conditions
for
future
visits
of
current
users
in
two
separate
onsite
surveys
in
1981
at
the
Mesa
Verda
National
Park
and
Great
Smoky
National
Park.
His
analysis
was
conducted
along
with
a
contingent
ranking
evaluation
of
the
benefits
of
improving
visibility
conditions
(
see
Chapter
6
for
a
more
complete
summary).
A
payment
card
was
used
as
the
instrument,
and
respondents
were
asked
how
much
they
would
pay
for
an
insurance
policy
to
guarantee
clear
visibility
conditions
for
all
visits
to
the
park.
Prices
on
the
card
ranged
from
O
to
$
10
in
increments
of
$
0.25.
The
average
bid
was
$
4.17
for
Mesa
Verda
respondents
and
$
5.96
for
Great
Smoky
respondents
(
estimates
in
1981
dollars).
Rae
interprets
this
as
a
present
value
option
price,
and
uses
estimates
of
current
user
values
for
visibility
improvements
derived
from
the
contingent
ranking
framework
to
estimate
option
value.

To
make
Rae's
interpretation
requires
assumptions
concerning
the
individual's
rate
of
time
preference
and
probabilities
of
future
visits.
Rae
uses
different
assumptions
in
estimating
option
value
in
the
two
studies.
For
the
Mesa
Verda
case,
he
assumed
a
zero
discount
rate
and
one
future
visit
while,
with
the
Great
Smoky
case,
he
postulated
an
8
percent
discount
rate
and
a
0.77
probability
of
one
return
visit
after
5
years.
The
expected
user
values
estimated
for
the
two
cases
were
$
3.00
and
$
5.00,
respectively.
Both
sets
of
assumptions
assure
a
positive
estimate
of
the
option
value.

In
order
to
evaluate
these
estimates,
the
Rae
methodology
for
estimating
user
values
with
the
contingent
ranking
framework
must
be
considered.
I
n
the
next
chapter
we
will
discuss,
in
detail,
the
use
of
the
contingent
ranking
approach
for
benefit
measurement.
Equally
important,
the
formulation
of
the
question
for
option
price
is
somewhat
vague
in
its
specification
of
the
terms
of
payment
for
the
insurance.
It
has
been
interpreted
as
a
one­
time
payment
in
the
analysis.
Given
that
all
the
other
components
of
the
survey
related
to
fees
associated
with
use,
this
distinction
may
not
have
been
appreciated
by
the
survey
respondents.

Finally,
the
estimation
of
option
value
requires
assumptions
on
the
time
horizon,
future
level
of
use,
future
probabilities
of
each
level
of
use,
and
the
individual
rate
of
time
preference.
Rae's
example
calculation
was
intended
to
illustrate
the
required
calculations.
Unfortunately,
there
is
little
basis
for
assuming
values
for
each
of
these
variables
for
his
survey
respondents.

5­
20
.

The
last
empirical
effort
at
measuring
nonuser
benefits
for
an
environis
the
Schulze
et
al.
[
1981]
analysis
of
visibility
at
four
national
~
entaI
amenitY
This
SUrVey
`
as
`
tructured
`
o
distinguish
users
from
nonusers
of
the
parks.
Canyon.
Each
group
was
asked
different
questions.
The
users
were
Grand
the
effects
of
visibility
on
their
user
values,
while
the
nonusers
~~~
ed
about
about
ptWSWVatiOn
values.
The
questions
related
to
visibility
at
were
asked
four
national
parks,
to
the
overall
region,
and
to
an
evaluation
of
the
willingto
avoid
a
visible
plume.
The
respondents
were
drawn
from
four
~
ess
to
Pay
LoS
Angeles,
A
l
b
u
q
u
e
r
q
u
e
,
D
e
n
v
e
r
,
a
n
d
Chicago­
Questionnaires
for
~
itieS:
users
employed
a
park
fee
as
the
payment
vehicle,
while
nonusers
were
queried
about
their
wiilin9ness
to
paY
for
preservation
values
through
electric
utility
~
ill
increaSe~.

Their
results
su99est
a
substantial
preserv?
t~
o~
value
(
in
1980
$)
ranging
~
rom
$
3.72
(
the
average
value
for
preserving
vlslblllty
at
the
Grand
Canyon
by
Denver
respondents)
to
$
9.06
(
the
average
for
Chicago
respondents)
per
~
mth.
These
are
substantially
greater
than
the
estimated
user
values,
which
ranged
from
$
O.'
9
`
0
$
5.40
per
visit
for
a
comparable
visibility
scenario.
t
f
it
is
appropriate
to
comPare
these
r=
ults
across
different
individuals
(
i
.
e.
,
implicitly
assuming
users
would
also
have
a
preservation
value),
the
estimated
~
reservatiOn
values
for
preserving
visibility
conditions
at
unique
natural
env
i
ron
ments,
such
as
the
Grand
Canyon,
may
be
much
greater
than
the
user
values
for
the.
same
visibility
conditions.
Unfortunately,
the
study
does
not
attempt
to
divide.
the
preservation
benefit
into
an
estimate
of
option
price
and
an
estimate
of
existence
value.
Thus,
it
is
not
directly
comparable
to
either
of
the
two
studi=
s
discussed
earlier
in
this
section.
Furthermore,
the
choice
of
two
diff@
rent
PaYm.
ent
vehicles
may
have
introduced
a
starting
point
bias
problem
similar
to
that
In
the
South
Platte
River
study.

Thus,
in
summary,
all
past
efforts
at
measuring
nonuser
values
have
met
with
only
limited
success.
There
has
been
controversy
over
whether
option
values
were
measured
or
it
has
not
been
possible
to
distinguish
option
price
from
other
components
of
intrinsic
values.

5,6
MEASURING
OPTION
VALUE:
SURVEY
DESIGN
As
noted
in
Chapter
1,
an
important
component
of
the
Monongahela
survey
was
the
measurement
of
option
price
and
user
values.
In
addition,
the
question
design
permitted
the
implications
of
supply
uncertainty
for
the
estimates
of
option
value
to
be
examined.
Since
Chapter
3
described
the
sample
survey
design
and
Chapter
4
provided
a
summary
of
the
features
of
the
final
sample,
these
will
not
be
repeated
here.
Rather,
this
section
will
review
the
background
information
provided
to
each
respondent
and
the
form
of
the
ques
­
tions
used
to
derive
estimates
of
the
option
value
associated
with
various
water
quality
changes
in
the
Monongahela
survey.

A
S
noted,
the
payment
vehicle
was
described
to
be
the
taxes
paid
directly
and
the
higher
prices
paid
indirectly
for
imDroved
water
auaiitv.
T
h
i
s
aoproach
fol
10­
ws
t
h
e
portent
additions.
anisms
that
underl
format
used
by
Mi~
chell
and
Carson
[
1981
j
witfi
several
irn­
Each
interviewer
was
trained
to
explain
carefully
the
meche
the
payment
vehicles.
The
objective
of
these
explanations
5­
21
was
to
ensure
that
respondents
understood
the
nature
of
the
payment
vehicle
and
recognized
that
similar
types
of
payments
take
place
in
practice
as
a
result
of
government
and
private
sector
decisions.
Each
respondent
was
shown
a
map
of
the
area
highlighting
the
locations
of
recreation
sites
along
the
river.
This
map
is
reproduced
as
Figure
4­
3.
Before
proceeding
to
the
questions,
the
interviewer
described
the
reasons
why
one
might
be
interested
in
water
quality
for
the
Monongahela
River.
Using
a
value
card
(
i.
e.
,
Figure
4­
6),
actual
use,
potential
future
use,
and
existence
values
were
each
identified
as
separate
reasons
for
interest
in
the
river
water
quality.
Each
was
acknowledged
to
be
a
potential
motivation
for
valuing
water
quality
in
the
Monongahela
River.
The
value
card
was
explained
at
the
outset
of
the
interview
and
then
shown
again
to
each
sample
respondent
as
the
questions
designed
to
separate
option
price,
expected
consumer
surplus,
and
existence
values
were
asked.
Thus,
the
value
card
translated
the
theoretical
relationships
relating
option
value,
user
value,
and
existence
value
into
a
format
that
linked
them
to
respondents'
experiences.

There
are
at
least
two
ways
to
ask
questions
designed
to
measure
the
option
values
associated
with
water
quality.
The
first
of
these
involves
proposing
to
respondents
counterfactual
situations
that
describe,
in
hypothetical
terms,
the
probabilities
and
levels
of
use
of
the
resource
with
different
specified
water
quality
levels.
Each
respondent
is
asked
to
value
these
plans.
A
second
approach
relies
on
the
interviewer's
ability
to
explain
to
the
respondent
why
he
might
value
water
quality
at
a
site,
identifying
the
relationships
between
those
reasons
and
a
benefits
taxonomy
that
isolates
option
value.
With
this
explanation,
the
individual
is
then
asked
to
bid
in
a
way
that
separates
the
individual
components
of
the
values.

These
methods
contrast
with
a
third
approach
employed
by
Mitchell
and
Carson,
where
a
classification
of
individuals
(
as
users
or
nonusers)
assisted
in
decomposing
benefits.
That
is,
their
classification,
together
with
the
assumption
that
nonusers
were
always
nonusers
and
therefore
could
not
have
user
values,
allowed
the
willingness­
to­
pay
estimates
from
nonusers
to
be
interpreted
as
indicative
of
the
intrinsic
benefits
held
by
users.

In
the
absence
of
the
assumption
that
individuals
are
comparable
(
except
in
the
decision
between
use
or
nonuse),
the
first
two
approaches
to
partitioning
the
benefits
of
a
water
quality
improvement
face
problems.
The
first
one
attempts
to
"
second
guess"
plausible
demand
conditions
in
its
specification
of
the
probabilities
and
levels
of
use
that
might
be
associated
with
a
water
quality
level.
Such
specifications
may
actually
bear
little
resemblance
to
what
an
individual
would
select.
Thus,
this
approach
was
not
used
in
this
analysis.

The
second
approach
relies
on
individuals'
ability
to
"
divide
the
benefits
pie"
consistently.
Clearly,
the
estimates
in
this
study
depend
not
only
upon
how
well
each
individual
understood
the
concepts
on
the
value
card,
but
also
upon
how
well
he
was
able
to
(
1
)
use
them
in
classifying
the
contributions
made
to
overall
option
price
by
expected
user
benefits
and
option
values
and
(
2)
separate
existence
values
as
a
distinct
motive
for
valuing
water
quality
improvements.

.
`
fh~
Survey
questionS
elicited
an
option
price­­
the
individually
willingthe
water
quality
change
due
to
actual
and
potential
use
of
~
e5s
to
Pay
`
o
r
Foilowing
this
question,
the
interviewer
asked
each
person
what
the
river"
~
mourlt
of
`
h
e
option
price
was
associated
with
actual
use.
This
response
has
as
an
estimate
of
the
individual's
expected
consumer
surplus.
~
een
interpreted
between
the
reported
option
price
and
the
value
associ
­
Thusl
.
the
difference
@
~
lth
use
corresponds
tO
this
study's
estimate
of
option
value.

The
questionnaire
design
allowed
evaluation
of
two
further
issues
in
the
of
option
value:
(
1)
the
amount
of
the
water
quality
change
and
measurement
mode
of
questioning.
The
design
considered
three
levels
of
change
in
(
z)
the
as
reproduced
in
the
water
quality
ladder
shown
in
Figure
+!
5.
water
qualitY
~
he
first
question
considered
the
willingness
to
pay
to
avoid
having
the
water
deteriorate
from
its
current
level,
Level
D,
acceptable
for
boating,
to
~
UaiitY
Level
El
at
which
no
recreation
activities
would
be
possible.
Individuals
were
their
willingness
to
Pay
for
improvements
from
Level
D
to
Level
C,
~
150
asked
for
sport
fishing,
~
Cceptable
and
improvements
from
Level
C
to
Level
B
,
for
swimming.
As
noted
in
the
previous
chapter,
the
water
quality
~
cceptabie
,
evel~
were
defined
based
on
Resources
for
the
Future's
water
quality
i
n
d
e
x
(
see
Mitchell
and
Carson
[
1981
]).

The
second
asPect
of
the
questionnaire
design
involved
the
mechanism
"~
ed
to
elicit
the
willingness­
to­
pay
response.
To
investigate
the
effects
of
different
questioning
methods,
the
sample
was
divided
into
approximately
four
~
qual
parts,
each
using
a
different
questioning
method­­
two
different
iterative
bidding
9ame
procedures,
a
direct
question
procedure,
and
a
procedure
using
a
direct
question
with
a
payment
card.
Iterative
bidding
games,
practiced
in
most
early
contingent
valuation
experiments
(
see
Schulze,
d'Arge,
and
Brook­
Shire
[
1981]
for
a
review),
involve
a
sequential
process
in
which
an
i
n
t
e
r
­
viewer
proposes
a
value
(
the
starting
point)
to
the
respondent
and
asks
Whether
it
would
be
acceptable
as
a
bid
for
the
conditions
described
in
the
question.
Based
on
the
response,
the
interviewer
raises
or
lowers
the
bid
by
a
fixed
amount
until
there
is
no
change
in
the
bid
with
repetition
of
the
process
Two
subsets
of
the
sample
used
bidding
game
procedures;
the
first
used
a
$
25
starting
point
and
a
$
5
increment,
and
the
second
used
a
$
125
starting
point
and
a
$
10
increment.

The
third
procedure
used
to
elicit
individual
willingness
to
pay
was
a
direct
question
with
no
suggestion
of
an
amount.
In
the
last
component
of
the
sample,
respondents
were
asked
to
look
at
a
payment
card
(
see
Figure
4­
7)
arraying
alternative
dollar
values
and
to
select
one
or
any
other
amount
as
their
willingness
to
pay.
This
last
procedure
is
comparable
to
the
Mitchell­
Carson
[
1981
]
approach,
with
one
modification.
The
values
on
the
card
were
not
identified
as
the
individual's
current
spending
on
specific
public
sector
activities.
This
practice
of
anchoring
the
values
was
not
used
because
it
was
`
elt
it
would
create
the
possibility
of
biased
responses.

Each
subsequent
question
for
user
values,
supply
uncertainty,
and
exist­
`"
ce
values
repeated
the
amount
given
for
willingness
to
pay
and
then
asked
`
he
respondent
to
indicate
what
portion
of
the
reported
willingness
to
pay
is
5­
23
Table
5­
2.
Summary
of
Willingness­
to­
Pay
Questions
by
Type
of
Interview
Type
of
interview
Question
format
Iterative
bidding
$
25
To
you
(
and
your
family),
would
it
be
worth
$
25
each
year
in
higher
taxes
and
prices
for
products
that
companies
sell
to
keep
the
water
quality
in
t
h
e
Monongahela
R
i
v
e
r
f
r
o
m
s
l
i
p
p
i
n
g
b
a
c
k
from
Level
D
to
Level
E?

Iterative
bidding
$
125
To
you
(
and
your
family),
would
it
be
worth
$
125
each
year
in
higher
taxes
and
prices
for
products
that
companies
sell
to
keep
the
water
quality
in
t
h
e
Monongahela
River
from
slipping
back
from
Level
D
to
Level
E?

Direct
question
Payment
card
W
h
a
t
`
is
the
most
it
is
worth
to
you
(
and
your
family)
on
a
yearly
basis
to
keep
the
water
quality
in
the
Monongahela
River
from
slipping
back
from
Level
D
to
Level
E,
where
it
is
not
even
clean
enough
for
boating?

What
is
the
most
it
is
worth
to
you
(
and
your
family)
on
a
yearly
basis
to
keep
the
water
quality
in
the
Monongahela
River
from
slipping
back
from
Level
D
to
Level
E,
where
it
is
not
even
clean
enough
for
boating?

Table
5­
3.
Summary
of
User,
Supply
Uncertainty,
and
Existence
Value
Questions
Tvoe
o
f
resDonse
Question
format
User
value
In
answering
the
next
question(
s),
keep
in
mind
your
actual
and
possible
future
use
of
the
Monon
­

gahela.
You
told
me
in
the
last
saction
that
it
was
worth
$<
AMOUNT~
to
keep
the
water
quality
from
slipping
from
Level
D
to
Level
E.
How
much
of
this
amount
was
based
on
your
actual
use
of
the
river?

Supply
uncertainty
Existence
value
If
the
water
pollution
laws
were
relaxed
to
the
p
o
i
n
t
t
h
a
t
t
h
e
w
a
t
e
r
q
u
a
l
i
t
y
would
d
e
c
r
e
a
s
e
t
o
Level
E
and
the
area
would
be
closed
1/
4
of
the
weekends
of
the
year
for
activities
on
or
in
t
h
e
water
but
would
remain
open
for
activities
near
t
h
e
w
a
t
e
r
,
how
much
would
you
change
this
(
READ
TOTAL
$
AMOUNT
)
to
keep
the
area
open
all
weekends
for
all
activities?

W
h
a
t
i
s
t
h
e
m
o
s
t
t
h
a
t
y
o
u
(
a
n
d
y
o
u
r
f
a
m
i
l
y
)
would
be
willing
to
pay
each
year
in
the
form
of
higher
taxes
and
prices
for
the
goods
Y
O
U
b
u
y
for
keeping
the
river
at
Level
D
where
it
is
okay
for
boating,
even
if
you
would
never
use
the
 
.
river?

Suppose
the
change
could
not
be
reversed
for
an
aven
l
o
n
g
e
r
p
e
r
i
o
d
o
f
t
i
m
e
t
h
a
n
y
o
u
r
l
i
f
e
t
i
m
e
.
How
much
more
than
(
READ
AMOUNT
FROM
a.
)
would
you
(
and
your
family)
be
willing
to
pay
per
year
to
keep
the
river
at
L
e
v
e
l
D,
even
if
You
would
never
use
the
river?

.

5­
24
:

.
 
.
with
each
of
the
components
Of
value
or
complications
to
the
choice
~~~
ociated
Table
5­
2
rePorts
the
form.
of
the
willingness­
to­
pay
questions
used
process.
for
the
case
Of
Preventing
deterioration
from
water
qualitY
Level
D
to
Level
E
for
each
mode"

The
questions
used
to
measure
the
values
associated
with
use,
supply
and
existence
values
did
not
change
with
the
type
of
interview
~
nCertaintYl
and
are
reported
in
Table
5­
3.
The
examples
correspond
to
the
and
Samples
scenario
used
for
the
willingness­
to­
pay
questions
in
Table
5­
2.
T
h
e
reto
these
questions
form
the
basis
for
the
results
reported
in
the
next
spOnses
section
of
this
chapter.

~
07
SURVEY
RESULTS­­
OPTION
VALUE
The
results
for
the
emPirical
estimates
of
option
value
are
divided
into
two
parts.
The
first
considers
the
conventional
treatment
of
option
value
as
~
response
to
demand
uncertainty.
The
second
considers
the
sensitivity
of
these
findings
to
changes.
in
,
the
conditions
of
access
to
the
Monongahela
River
by
varying
the
proposed
likelihood
of
being
able
to
use
the
site.

5.7.
I
Option
Value­­
Demand
Uncertainty
Table
5­
4
presents
a
summary
of
the
sample
mean
estimate
of
option
value
for
each
water
qual,
ity
change
based
on
each
of
the
four
types
of
interview
frameworks.
The
estimates
for
each
water
quality
change
are
the
increments
to
the
reported
willingness
to
pay
to
prevent
the
water
quality
from
deteriorating
to
the
level
given
as
E.
Thus,
each
respondent
was
asked
if
he
Would
be
willing
to
pay
more
than
the
amount
recorded
for
avoiding
a
movement
from
D
to
E.
When
an
affirmative
answer
was
given,
the
interviewer
proceeded
with
the
increments
from
D
to
C
and
from
C
to
B.
Since
some
individuals
were
unwilling
to
pay
for
further
improvements,
the
"
no"
responses
to
subsequent
improvements
were
treated
as
zeros
in
constructing
the
means.

Analysis
of
the
survey
responses
revealed
that
two
definitions
of
"
users"
were
possible.
The
first
of
these
would
classify
individuals
according
to
whether
they
reported
a
user
value
or
indicated
that
they
had
used
the
river
for
recreation
activities
in
the
previous
year.
This
definition
is
the
focus
of
attention
in
this
chapter
and
is
designated
as
the
"
broad
definition"
of
users.
The
second
defines
users
as
only
those
individuals
who
indicated
that
they
had
used
the
Monongahela
sites.
This
narrow
definition
focuses
on
a
subset
of
the
users
under
the
first
definition.
Appendix
C
reports
a
sample
of
the
results
under
the
narrow
definition.

The
analysis
performed
for
this
study
has
considered
both
the
sample
means
and
linear
regression
models
to
summarize
the
survey
reSLJ!
tS.
Table
5­
4
provides
estimates
for
option
value
for
different
levels
of
water
quality
change
according
to
the
survey
instrument
used.
Informal
review
of
these
estimates
seems
to
indicate
that
the
question
format
influences
the
magnitude
of
the
estimates.
Following
the
practices
described
in
Chapter
4,
these
estimates
are
based
on
a
restricted
sample:
Observations
identified
as
either
5­
25
Table
5­
4.
Estimated
Option
Values
for
Water
Quality
Change:
Effects
of
Instrument
and
Type
of
Respondent­­
Protest
Bids
and
Outliers
Excluded
Type
of
respondent
Change
in
User
a
Nonuser
water
quality
2
s
n
i
s
n
1.

2.

3.

4.
Iterative
Bidding
Framework,
Starting
Point
=
$
25
D
to
E
(
avoid)
20.79
16.61
19
29.74
Dto
C
14.74
13.99
19
14.49
Cto
B
6.84
10.70
19
7.18
Dto
B
21.58
22.05
19
21.67
Iterative
Bidding
Framework,
Startinq
Point
=
$
125
D
to
E
(
avoid)
58.44
66.6o
16
38.75
Dto
C
37.81
49.13
16
26.25
Cto
B
13.13
32.65
16
11.56
Dto
B
50.94
71.44
16
40.47
Direct
Question
Framework
D
to
E
(
avoid)
25.59
43.04
17
14.18
Dto
C
10.12
24.45
17
10.82
Cto
B
10.18
24.49
17
8.47
Dto
B
21.77
48.57
17
20.32
Payment
Card
D
to
E
(
avoid)
27.06
33.12
17
52.97
Dto
C
14.41
20.38
17
21,89
Cto
B
3.26
8.28
17
7.70
Dto
B
20.00
25.06
17
29.87
36.69
15.17
11.63
24.04
51.32
45.38
33.06
69.02
27.12
21.56
21.87
41.45
76.31
33.80
19.99
47.54
39
39
39
39
32
32
32
32
34
34
34
34
37
37
37
37
a
These
results
are
based
on
the
broad
definition
of
users.

protest
bids
or
as
rejecting
or
misunderstanding
the
contingent
valuation
experiment
were
deleted.
The
latter
were
initially
identified
as
outlying
observations
using
regression
diagnostics
(
see
Belsley,
Kuh,
and
Welsch
[
1980]).
This
statistical
identification
was
followed
by
an
evaluation
of
the
features
of
the
observations
that
made
them
distinct
(
see
Table
4­
8
and
Section
4.5
for
further
discussion).
To
consider
this
issue,
as
well
as
the
potential
effects
of
being
a
user
of
the
river,
several
null
hypotheses
have
been
chosen
for
testing
using
a
student
t­
test
for
the
difference
of
sample
means.
Equation
(
5.
14)
below
provides
the
test­
statistic
used
for
these
tests:

5­
26
.
.
.
.
.
E
 
.
 
 
.
­.
 
.
.
.
.._
.
 
,
­­­­­
.
­
._.
 
 
.
 
__________
._
where
ii
=

S
i
=

n
i
=
t=
x
,
­
X?

(
n,
­
l)
s12+(
n9­
l)
s22
nl+
n
(
nl
+
n
2
­
2)
n
l
l
 
n
2
.
(
5.14)

sample
mean
for
the
ith
grouping
of
individuals
(
e.
g.
,
users,
nonusers,
respondents
with
a
particular
question
format,
etc.
),

sample
standard
deviation
for
ith
grouping
of
individuals
sample
size
for
the
ith
grouping
of
All
combinations
of
questioning
format
water
quality
were
compared
for
users
and
only
a
few
cases
where
the
estimated
means
individuals.

for
each
type
of
improvement
in
nonusers.
Overall,
there
were
were
significantly
different.
As
a
rule
these
cases
were
associated
with
comparisons
of
the
iterative
bidding
framework
under
the
two
starting
points.
Thus,
there
is
some
evidence
of
Startin9
Po!
nt
bias
with
this
aPProach
to
soliciting
an
individual's
valuation
of
`
Water
q
u
a
l
i
t
y
.
Indeed,
t
h
e
s
e
r
e
s
u
l
t
s
f
o
r
s
t
a
r
t
i
n
g
p
o
i
n
t
b
i
a
s
w
o
u
l
d
b
e
strengthened
­
if
the
observations
that
were
deleted
as
invalid
(
from
the
diagnostic
anal
Ys!
s).
were
Included
in
the
sample.
In
several
cases
it
was
not
possible
to
.
dlstlnguish
the
effect
of
the
higher
starting
point
(
i.
e.,
$
125)
as
an
explanation
of
the
observation's
role
as
an
outlier
from
another
characteristic
of
the
survey
respondent
involved
(
see
Chapter
4).
Table
5­
5
summarizes
the
cases
where
statistically
significant
differences
in
the
mean
values
for
option
value
were
found.

Table
5­
5.
Student
t­
Test
Results
for
Question
Format
a
t­
Ratios
Means
compared
User
Nonuser
Direct
question
vs.
iterative
bidding
with
­
2.069
­
2.452
$
125
starting
point
D
to
C
Iterative
bidding
with
$
25
starting
point
vs.
­
2.384
­
­
iterative
bidding
with
$
125
starting
point
D
to
E
(
avoid)

Iterative
bidding
with
$
25
starting
point
vs.
­
1.960
­
­
iterative
bidding
with
$
125
starting
point
D
to
c
Direct
question
vs.
iterative
bidding
with
­
­
­
2.035
$
125
starting
point
D
to
E
Direct
question
vs.
iterative
bidding
with
­
­
­
2.758
$
125
starting
point
D
to
B
.
`
This
table
reports
only
the
means
were
found
at
the
cases
where
statistically
significant
differences
in
the
.
O.
05
significance
level.

5­
27
­­­­­­
r­
.
 
 
 
 
The
responses
of
users
and
nonusers
were
also
compared
for
each
type
of
question
and
level
of
water
quality
change.
Based
on
observation
of
values
in
Table
5­
4,
none
of
these
cases
indicated
a
significant
difference
in
the
means.
Thus,
despite
the
appearance
of
rather
large
differences
for
a
few
cases
(
e.
g.
,
payment
card
with
Level
D
to
Level
E),
the
estimated
means
are
not
significantly
different.
ti
Table
5­
6
reports
the
findings
of
a
sample
of
the
linear
regression
models
considered
in
attempting
to
explain
the
determinants
of
the
option
value
estimates
using
the
survey
respondents'
economic
and
demographic
characteristics.
These
models
should
not
be
interpreted
as
estimates
of
a
behavioral
model.
Rather,
they
were
estimated
as
summaries
of
the
survey
data
in
an
attempt
to
Table
5­
6.
Regression
Results
for
Option
Valu~
Estimates­­
Protest
Bids
and
Outliers
Excluded
Water
quality
changes
Dto
E
Independent
variables
(
avoid)
Dto
C
cto
B
Dto
B
1
ntercept
Sex
(
1
if
male)

Age
User
(
1
if
user)

Education
Income
Direct
question
Iterative
bidding
game
($
25)
Iterative
bidding
game
($
125)
Willing
to
pay
cost
of
water
pollution
(
1
if
very
much
or
somewhat

R
2
F
Degrees
of
freedom
­
17.014
(­
0.540)
4.121
(
0.484)
­
0.411
(­
1
.637)
­
18.454
(­
2.097)
4.830
(
2.052)
0.0005
(
1
.384)
­
26.128
(
­
2
.
3
5
6
)
­
12.681
(
1.188)
14.638
(
1.245)
16.069
(
1
.842)

0.212
4.34
155
­
7.170
(­
0.380)
­
0.133
(­
0.026)
­
0.216
(­
1
.435)
­
10.609
(­
2.011)
2.084
(
1
.477)
0.00005
(
0.210)
­
7.472
(­
1.124)
­
0.274
(­
0.043)
20.601
(
2.923)
16.611
(
3.176)

0.208
4.23
155
10.149
(
0.692)
­
2.332
(
0.589)
­
0.131
(
­
1
.
1
2
0
)
­
4.518
(­
1.104)
­
0.167
(­
0.152)
0.0002
(
1.035)
3.335
(
0.646)
1.773
(
0.357)
7.575
(
1
.385)
4.510
(
1.111)

0.053
0.90
155
3.635
(
0.126)
­
3.301
(­
0.424)
­
0.350
(­
1
.523)
­
15.761
(­
1
.958)
1.986
(
0.922)
0.0002
(
0.532)
­
3.817
(­
0.376)
0.339
(
0.03s)
29.627
(
2.754)
23.229
(
2.910)

0.170
3.30
155
a
Numbers
in
parentheses
are
asymptotic
t­
ratios
for
the
null
hypothesis
of
no
association.

5­
28
b
~
h~
ability
to
describe
the
attributes
of
individual
respondents
that
improve
ed
to
influence
the
estimates
of
option
vaiue.
Thus,
while
these
results
S@
have
limited
explanatory
power
,
as
measured
by
the
R*
of
each
equation,
they
do
provide
somewhat
different
insights
into
the
role
of
the
type
of
respondent
offered
bY
the
analysis
of
sample
means.
The
independent
variables
~
han
those
model
included
qualitative
variables
for
sex,
question
format
(
with
the
in
the
the
omitted
questioning
mode),
user,
and
the
individualism
ex
­
pay~
~
nt
card
as
attitude
`
o
Ward
.
payin9
for
water
quality
improvements.
The
last
of
pressedwas
coded
as
a
1
If
the
individual
"
strongly"
or
"
somewhat"
considered
these
a
person
willing
to
pay
the
cost
required
to
control
water
pollution.
~
i~
self
the
variable
was
coded
as
zero
(
i.
e.,
for
individuals
who
had
little
~
therwisel
had
no
opinion
on
the
matter
).*
or
no
such
feelings
or
After
the
survey
respondents'
characteristics
were
controlled,
users
have
lower
oPtion
values
than
nonusers.
No
differences
were
found
~
eerned
to
based
on
sample
means.
Since
the
tests
for
the
equality
of
means
~
sin9
tests
did
not
control
for
the
respondents'
characteristics,
the
difference
in
the
two
~
onCiusiOns
is
not
surprising.
The
regression
results
add
further
support
to
tjle,
conclusion
for
a
starting
point
bias.
Two
of
the
four
models
in
Table
5­
6
.
.
indicate
that
the
.
qualltatlv~
var~
able
­
identifying
the
respondents
who
received
~
be
iterative
blddlng
questionnaire
with
a
$
125
starting
point
was
significantly
different
from
zero.
This
implies
that
these
responses
are
significantly
different
than
those
received
using
the
payment
card.
The
two
most
consistent
determinants
`
f
`
he
option
`
a[
ue
results
in
these
models
were
the
qualitative
Variables
fo
r
`
Ser
and
`
or
`
he
Indlvldual's
willingness
to
pay
the
costs
required
for
water
pollutlon
control.

overall,
these
results
indicate
that
it
is
possible
to
estimate
option
value
for
water
quality
changes.
In
general,
the
estimates
are
significantly
different
from
zero.
The
effects
of
payment
vehicle
suggest
that
there
appears
to
be
a
starting
point
bias
with
several
estimates
of
option
value
for
specific
water
quality
changes.
Morever,
with
the
ability
to
control
for
respondents'
characteristics
the
iterative
bidding
approach
with
a
$
125
starting
point
was
found
to
increase
option
value
estimates
over
the
responses
made
using
a
payment
card.

The
results
were
not
especially
successful
in
isolating
the
effects
of
other
individual
characteristics
on
the
option
value
estimates.
Only
the
variable
indicating
the
individual's
attitude
toward
paying
for
water
pollution
control
was
a
consistent
determinant
of
the
option
value
estimates
for
the
water
quality
changes.

These
estimates
are
all
based
on
the
assumption
that
access
to
the
site
is
9Wranteed.
Accordingly,
the
implications
of
supply
uncertainty
for
the
re­
Wondents'
option
prices
are
considered
next.

S.
7.2
Option
Value­­
SupPIy
Uncertainty
Because
the
theoretical
analysis
of
the
sign
of
option
value
and
the
re­
`
U'ts
in
Smith
[
1983]
suggest
that
individuals'
assumptions
regarding
their
*
A
more
detailed
description
of
these
variables
is
provided
in
Chapter
4.

5­
29
.
ability
to
gain
access
to
the
site
­­
i
.
e.
,
the
degree
of
perceived
supply
uncertainty
may
be
important
to
the
magnitude
of
option
value,
severai
qu@
stions
were
incorporated
to
attempt
to
measure
its
effects
on
individual's
responses.
Table
5­
3
reported
the
question
used
to
gauge
the
effects
of
supply
uncertainty
Three
variants
of
the
question
were
posed,
each
of
which
referred
to
the
amount
an
individual
would
be
willing
to
pay
to
prevent
water
quality
in
the
Monongahela
River
from
deteriorating
from
boatable
to
unusable.
supply
uncertainty
was
introduced
by
suggesting
that
the
water
quality
deterioration
would
take
place
and
that
it
would
reduce
the
probability
of
having
access
to
the
river's
recreation
sites.
The
first
question
postulated
that
activities
o
n
or
in
the
water
would
be
precluded
for
one­
fourth
of
the
weekends
in
the
year.
The
respondent
was
informed
that
it
would
not
be
known
in
advance
which
weekends
would
be
involved.
The
fraction
of
weekends
during
which
the
sites
were
closed
was
progressively
increased
through
two
more
steps
to
one­
half
and
three­
fourths
of
the
weekends.
Table
5­
7
reports
the
estimated
mean
adjustments
to
the
original
bids
made
by
users
and
nonusers.
That
is,
each
respondent
was
reminded
of
his
bid
to
prevent
water
quality
from
deteriorating
from
Level
B
to
Level
E
and
then
asked
how
much
this
amount
would
be
altered
to
reflect
the
supply
uncertainty.

These
responses
indicate
that
supply
uncertainty
clearly
affects
the
option
prices
bid
by
users.
The
means
for
users
under
each
of
the
three
conditions
of
supply
uncertainty
are
significantly
different
from
zero
at
the
5­
percent
level.
These
results
suggest
that
the
option
price
would
be
reduced
if
the
water
quality
level
led
to
uncertain
availability
of
the
site.
The
mean
adjustments
to
the
option
prices
reported
by
nonusers
were
not
significantly
different
from
zero.

Table
5­
7.
Effects
of
Supply
Uncertainty
on
Option
Price
a
Summary
Condition
of
water
quality
change
statistics
U
s
e
r
b
Nonuser
b
Avoid
a
certain
change
B
to
E
i
114.710
61.817
s
112.501
85.40
n
69
142
Experience
water
quality
change
i
­
14.552
­
6.354
to
E,
lose
1/
4
weekends
s
52.328
39.891
n
67
96
Experience
water
quality
change
to
i
­
22.537
­
5.833
E,
lose
1/
2
weekends
s
58.331
43.996
n
67
96
Experience
water
quality
change
to
2
­
26.866
­
6.042
E,
lose
3/
4
weekends
s
68.500
46.220
n
67
96
a
These
results
are
based
on
a
sample
that
deletes
protest
bids
and
the
observation
s
identified
as
inconsistent
with
the
contingent
valuation
framework.

bThe
difference
in
the
number
of
observations
between
the
certain
case
and
the
uncertain
cases
reflects
missing
observations.

5­
30
.
Table
5­
8.
Student
t­
Tests
for
the
Effects
of
Supply
Uncertainty
for
Users
Means
t­
Ratio
Water
quality
reduces
access
for:

(
1)
1/
4
weekends
vs.
1/
2
weekends
0.834
(
2)
1/
4
weekends
vs.
3/
4
weekends
1.169
(
3)
1/
2
weekends
vs.
3/
4
weekends
0.394
Table
5­
8.
rePorts
the
results
for
tests
of
the
differences
in
the
mean
~
djustments
with
progressive
increases
in
the
degree
of
supply
uncertainty.
The
results
suggest
that
the
mean
adjustments
are
not
significantly
different
with
increases
in
the
uncertainty
in
the
availability
of
the
site.

In
summary,
these
empirical
findings
confirms
the
theoretical
arguments
developed
earlier.
Supply
uncertainty
can
be
expected
to
affect
option
value.

AVoidin9
supply
`
certainty
and
the
associated
risk
is
further
basis
for
a
positive
option
value.

5.8
EXISTENCE
VALUE
ESTIMATES
Since
they
were
first
introduced
have
been
given
Iittie
attention
within
by
Krutilla
[
1967]
,
existence
values
conventional
models
of
consumer
behavior
*
The
recent
experimental
findings
of
Schulze
et
al.
[
1981],
discussed
earlier
in
this
chapter,
have
changed
this
perspective.
Their
estimates
of
preservation
values
for
the
Grand
Canyon's
visibility
conditions
indicate
that
the
nonuser
values
for
this
unique
natural
environment
are
likely
to
be
several
times
the
magnitude
of
the
user­
associated
benefits.
While
~
t
is
not
unambiguously
clear,
preservation
values
can
be
expected
to
include
option
value,
existence
value,
and,
perhaps,
bequest
values.
Each
of
these
motivations
for
desiring
the
services
of
a
unique
natural
environment
was
identified
by
Kru
­
tilla
as
values
that
would
not
necessarily
be
reflected
in
the
private
market
transactions
for
the
services
of
such
resources.

As
a
result
of
these
empirical
findings,
the
attention
given
to
modeling
and
measuring
existence
values
has
increased.
Freeman's
[
1981
]
recent
notes
on
the
problems
associated
with
defining
and
measuring
existence
values
indicate
at
least
two
interpretations
of
an
individual's
reasons
for
valuing
the
existence
of
a
resource.
In
the
first
note,
Freeman
designates
a
stewardship
value
(
or
motive),
where
the
level
of
use
of
a
resource
affects
the
value
derived
In
this
c
a
s
e
,
one's
existence
value
would
be
reduced
if
the
resource
*
One
notable
exception
is
Miller
and
Menz's
[
1979]
model
for
describing
efficient
allocation
decisions
involving
wildlife
preservation.
These
authors
introduce
species
stock
terms
into
individuals'
utility
functions
as
a
source
of
value,
without
requiring
that
these
values
arise
from
consumptive
uses.
Howeverr
the
authors
do
not
explicitly
identify
the
rationale
for
their
specification
in
terms
of
existence
value.

5­
31
were
not
properly
managed.
Freeman's
second
proposed
reason
for
existence
value
stems
from
a
form
of
vicarious
consumption.
An
individual
derives
benefit
from
the
knowledge
that
other
individuals
can
use
a
resource.

Freeman's
analysis
does
not
develop
either
of
these
frameworks
in
detail.
They
were
suggested
only
as
prospective
explanations
for
values
due
to
the
existence
of
a
resource
and
can
be.
interpreted
as
defining
different
forms
of
consumption.
Thus,
they
do
not
provide
direct
insight
into
how
existence
values
might
be
measured.
However,
Freeman
does
suggest
that
attempts
to
measure
existence
value
should
carefully
identify
the
likelihood
of
future
use
of
the
site
and
elicit
an
individual's
user
and
nonuser
values.
I
n
effect,
he
proposes
that
questions
call
for
the
sum
of
option
price
and
existence
value.

The
design
of
the
existence
value
questions
for
this
survey
attempted
to
use
these
insights.
The
sources
of
site
valuation
(
on
the
value
card
used
in
the
interviews)
were
separated
into
direct
use,
potential
use,
and
existence
motives.
After
reviewing
these
motivations,
the
interviewer
asked
each
respondent
how
much
he
would
be
willing
to
pay
to
prevent
the
deterioration
of
water
quality
from
boatable
conditions
to
an
unusable
state
even
though
he
never
would
plan
to
use
the
river.
Responses
to
these
questions
were
regarded
as
tentative
estimates
of
existence
values.
The
situation
is
a
difficult
one
for
the
respondent
to
conceptualize.
Water
quality
is
to
remain
at
a
beatable
level,
but
the
individual
nonetheless
will
not
use
the
river.

Table
5­
9
presents
these
results
for
users
and
nonusers
with
the
sample
restricted
to
exclude
protest
bids
and
observations
judged
to
be
inconsistent
with
the
contingent
valuation
framework.
Both
estimates
are
significantly
different
from
zero.
Users
do
exhibit
significantly
different
estimated
existence
values
from
nonusers
at
the
5­
percent
level.
These
values
are
quite
comparable
to
the
estimates
for
the
option
price
(
aggregated
over
question
mode),
as
reported
in
Table
4­
9
for
avoiding
the
loss
of
use
of
the
river.
Indeed,
there
is
not
a
significant
difference
between
the
means
for
either
users
or
nonusers.
This
finding,
together
with
the
fact
that
many
respondents
repeated
their
option
price
bids
for
the
existence
value
question,
suggests
that
these
results
should
be
interpreted
with
caution.
Until
the
theoretical
issues
associated
with
describing
the
relationship
between
user
and
existence
values
is
resolved,
it
cannot
be
concluded
that
these
estimates
represent
independent
sources
of
value
for
a
water
quality
improvement.

Table
5­
9.
Estimated
Existence
Values
User
Nonuser
Mean
(<)
65.985
42.115
Standard
deviation(
s)
92.824
64.023
n
66
139
5­
32
~
og
SUMMARY
This
chapter
has
reviewed
~
alue~
summarized
the
results
of
and
presented
the
user
values~
reiate
to
nonuser
values.
the
theory
underlying
the
definition
of
option
past
efforts
to
measure
option
and
other
nonresults
of
the
Monongahela
River
survey
that
The
findings
provide
clear
support
for
a
positive,
statistically
significant,
option
value
for
water
quality
improvements
for
the
Monongahela
and
substantial
River.
The
estimated
option
values
for
loss
of
the
use
of
the
area
in
its
cur­
(
i.
e.
,
providing
boating
recreation
activities)
range
from
ap­
rent
condition
to
$
58
for
users
(
and
$
14
to
$
53
for
nonusers).
The
option
~
roximatelY
$
21
price
for
users
ranges
from
approximately
$
27
to
$
95.
Thus,
option
value
is
a
substantial
fraction
of
the
option
price
of
users
and
generally
exceeds
their
"
se
values
for
a
change
in
water
quality.
The
Monongahela
River
is
not
a
unique
recreation
site.
Thus,
these
estimates
may
well
require
reconsideration
of
the
conventional
assumption
that
oPtion
value
is
small
in
comparison
to
use
Value
for
`
atura'
`
nv'ronments
without
unique
attributes.
Of
course,
it
should
also
be
acknowledged
that.
the
available
estimates
of
option
value
are
quite
limited
Most
can
be
crltlclzed
for
problems
in
the
research
design,
including
possible
flaws
in
the
survey.
The
design
of
the
Monongahela
River
study
places
heavy
reliance
on
the
use
of
a
schematic
classification
of
the
sources
of
an
individual's
valuation
of
the
river
(
i
.
e.
,
the
value
card)
in
eliciting
a
division
of
user
and
nonuse
benefits.
Because
this
is
the
first
application
of
this
device~
It
was
not
possible
to
evaluate
its
effectiveness.

Users
appear
to
have
a
somewhat
lower
option
value
than
nonusers
for
most
levels
of
change
in
water
quality.
For
the
most
part,
the
respondents'
socioeconomic
characteristics
were
not
useful
in
explaining
the
variation
in
estimated
option
values.

The
limited
analysis
of
the
role
of
supply
uncertainty
for
measures
of
option
value
clearly
suggests
it
is
an
important
influence
on
users'
option
price
(
and
therefore
on
the
derived
option
value).
Assurance
of
supply
is
quite
important
to
our
positive
estimates
for
option
value.

Finally,
this
survey
provided
the
ability
to
estimate
existence
values.
While
the
findings
suggest
that
these
values
are
positive
and
statistically
significant
prudence
requires
they
be
interpreted
cautiously.
It
is
not
clear
that
respondents
understood
the
distinction
sought.
Many
bid
the
same
amounts
as
their
earlier
option
prices
for
a
comparable
change
in
water
quality.

5­
33
CHAPTER
6
CONTINGENT
RANKING
DESIGN
AND
RESULTS:
OPTION
PRICES*

6.1
INTRODUCTION
T
h
e
purPose
of
this
chapter
is
to
report
a
set
of
water
quality
benefit
~~
timates
based
on
an
analysis
of
the
Monongahela
survey
respondents'
ranking5
of
four
hypothetical
combinations
of
water
quality
levels
and
amounts
paid
in
the
form
of
higher
taxes
and
prices.
The
use
of
data
including
individuals
rankings
of
goods
or
services
described
in
terms
of
the
features
of
each
of
a
set
of
possible
alternatives
together
with
an
extension
of
the
McFadden
[
1974]
random
utility
model
was
first
proposed
by
Beggs,
Cardell,
and
Hausman
[
1981
1
=
a
method
for
measuring
the
potential
demand
for
new
goods.
Rae
[
1981a,
1981b]
has
subsequently
used
this
approach
as'
an
alternative
means
of
estimating
individuals'
valuation
of
air
quality
improvements.
The
implicit
assumption
of
the
contingent
ranking
approach
is
that
individuals
are
more
likely
to
be
capable
of
ordering
hypothetical
combinations
of
environmental
amenities
and
fees
than
to
directly
reveal
their
willingness
to
pay
for
any
specific
change
in
these
amenities.
Unfortunately,
past
studies
have
tended
to
adopt
only
one
or
the
other
of
these
two
approaches,
and
there
has
been
little
basis
for
comparing
their
respective
estimates.
As
a
resu{
t,
t
h
e
survey
instrument
for
the
Monongahela
study
was
designed
explicitly
to
include
the
use
of
contingent
ranking
as
a
method
for
measuring
individuals'
valuation
of
water
quality
improvements.
All
survey
respondents
were
asked
to
rank
four
hypothetical
combinations
of
water
quality
and
payments
to
permit
a
comparison
of
contingent
valuation
and
contingent
ranking
methods
within
the
context
of
a
common
application.

To
understand
the
economic
basis
for
modeling
consumer
behavior
using
contingent
rankings,
the
random
utility
model
­­
widely
applied
to
model
consumer
behavior
that
involves
discrete
choices­­
must
first
be
considered.
Section
6.2
provides
some
of
this
background
by
describing
the
features
of
the
random
utility
model,
and
Section
6.3
discusses
two
possible
methods
for
implementing
the
model.
The
first,
an
adaptation
of
the
conditional
Iogit
model,
can
be
derived
under
the
assumption
that
the
errors
associated
with
the
random
utility
function
are
additive
and
follow
an
extreme
value
distribution
(
i.
e.
,
the
Weibull
d
i
s
t
r
i
b
u
t
i
o
n
)
.
T
h
e
s
e
c
o
n
d
,
a
normal
counterpart
to
the
*
Special
acknowledgment
is
due
Donald
Waldman
of
the
Department
of
Economics
University
of
North
Carolina
at
Chapel
Hill,
who
helped
develop
the
maximum
likelihood
program
for
ordered
Iogit
analysis
and
provided
a
general
program
for
estimating
the
Keener­
Waldman
ordered
normal
estimates.
He
also
assisted
in
the
estimation
and
discussed
several
aspects
of
these
models
with
the
authors.

6­
1
ordered
Iogit,
was
recently
developed
by
Keener
and
Waldman
[
1981]
,
who
used
numerical
procedures
to
approximate
the
likelihood
function
associated
with
a
random
utility
function
having
additive
normal
errors.
With
this
background
Section
6.4
summarizes
the
results
of
Rae's
survey
applications
of
the
contingent
ranking
approach
to
benefit
estimation
for
visibility
change;
Section
6.5
discusses
the
question
used
for
contingent
ranking
and
the
empirical
estimates
of
random
utility
models;
and
Section
6.6
considers
some
of
the
theoretical
issues
associated
with
Rae's
proposed
approach
for
benefit
estimation
with
the
model
and
reports
the
results
derived
by
applying
it
directly
with
the
Monongahela
survey
data.
Finally,
Section
6.7
summarizes
the
chapter
and
proposes
an
alternative
application
of
the
random
utility
model.

6.2
CONSUMER
BEHAVIOR
AND
THE
CONTINGENT
RANKING
FRAMEWORK
The
conventional
economic
description
of
consumer
behavior
general
I
y
maintains
that
each
individual
consumes
some
amount
of
every
good
or
service
that
enters
his
utility
function.
The
objective
of
these
models
is
to
describe
the
choices
individuals
make
for
marginal
increments
to
their
consumption
levels.
That
is,
individuals
are
usually
portrayed
as
adding
to
previous
consumption
of
goods
or
services
from
which
they
derive
utility.
*
Of
course,
many
consumer
choices
involve
major
purchases.
In
the
purchase
of
an
automobile
or
a
house,
the
selection
of
an
occupation,
or
the
choice
of
an
appliance
the
consumer's
decisions
all
require
discrete
choices.
In
these
cases,
the
commodity
often
is
durable
and
provides
a
stream
of
services
over
some
time
period
or
involves
some
commitment
of
the
individual's
time.
Thus,
the
assumption
of
continuous
incremental
adjustment
in
the
levels
of
consumption
of
each
good
or
service
that
is
implied
in
the
conventional
model
of
consumer
behavior
is
not
plausible
for
describing
individuals'
choices
when
they
involve
discrete
selections.

Several
types
of
modifications
to
conventional
models
have
been
proposed
to
make
them
more
amenable
to
explaining
such
discrete
choice
problems.
One
involves
an
extension
of
the
time
horizon
in
the
conventional
model
of
consumer
behavior.
For
example,
on
any
particular
day
a
commuter
will
select
a
travel
mode
to
reach
his
job.
Viewed
on
a
daily
basis,
modal
choice
is
discrete
since
fractions
of
the
available
travel
modes
cannot,
as
a
rule,
be
consumed
in
a
single
trip
to
the
workplace.
However,
over
the
course
of
a
month
or
a
year,
the
individual
may
well
select
a
varied
menu
of
transport
modes.
Thus,
with
this
adaptation
of
lengthening
the
time
horizon,
the
conventional
model
of
consumer
behavior
may
be
more
relevant
to
explaining
these
decisions.

A
second
proposed
adaptation
for
dealing
with
discrete
choices
involves
modeling
consumer
decisions
as
service
flows
rather
than
as
the
choice
of
any
particular
asset.
For
example,
an
individual
purchases
an
auto
for
transportation
services.
These
service
decisions
may
be
more
amenable,
under
this
interpretation
of
conventional
theory,
to
modeling
than
the
discrete
choices
of
*
Conventional
models
of
consumer
behavior
assume
positive
levels
of
consumption
of
all
gmds
and
services
to
avoid
dealing
with
corner
solutions.

6
­
2
L
I
I
I
durable
goods
themselves.
As
a
practical
matter,
however,
most
of
the
modifications
to
the
conventional
theory
have
enjoyed
limited
success.
Information
on
the
consumption
rates
for
the
services
of
durables
is
virtually
nonexistent.

Forecasts
of
the
rates
of
use
of
travel
modes
based
on
aggregate
information
over
long
time
spans
cannot
take
account
of
the
specific
constraints
facing
individuals
in
making
these
decisions
and,
as
a
result,
may
be
inadequate
for
many
problems.

The
random
utility
model
has
been
proposed
as
one
approach
for
dealing
with
discrete
consumer
choices.
It
generally
replaces
the
assumption
of
a
common
behavioral
objective
function
across
individuals
with
the
assumption
of
a
distribution
of
objective
fUnCtiOn
S.
Attention
is
shifted
from
the
intensive
choice
margin
and
the
associated
incremental
analysis
to
individual
decisionmaking
at
an
extensive
margin
with
discrete
selections.
As
a
result,
random
utility
models
are
often
quite
simple
in
their
description
of
the
choice
procesS
Individuals
are
assumed
to
have
utility
functions
affected
by
(
1)
the
objects
of
choice
and
their
features
and
(
2)
the
characteristics
of
the
individuals
making
the
decisions.
The
analyst
is
assumed
to
be
capable
of
observing
the
distribution
of
individuals
and
their
respective
choices
but
does
so
without
complete
information.
Thus,
the
observed
behavior
is
assumed
to
be
described
as
a
trial­­
the
drawing
of
one
individual
from
a
population;
the
recording
of
his
attributes,
the
alternatives
available,
and
their
features;
and
the
making
of
a
choice.
Because
there
is
a
distribution
of
individuals,
the
model
describes
the
choice
process
using
a
conditional
probability.
Each
alternative
has
some
probability
of
being
selected
based
on
its
characteristics,
the
other
alternatives
available
and
their
features,
and
the
attributes
of
the
individual
selected.
Behavior
is
described
by
modeling
these
probabilities.

The
random
utility
function
provides
the
vehicle
for
modeling
these
conditional
probabilities.
In
a
random
utility
framework,
the
individual
is
assumed
to
select
alternatives
that
provide
the
highest
utility
level.
Thus,
if
Equation
(
6.1
)
describes
a
random
utility
function,
then
individual
j's
probability
of
selecting
alternative
k,
given
j's
attributes,
z.,
and
in
the
presence
of
the
set
of
alternatives
defined
by
A,
is
defined
by'
the
probability
that
j's
utility
of
k
will
exceed
the
utility
of
all
other
alternatives,
as
given
in
Equation
(
6.2)
below:

where
U(
a,
z)
=

z
=

V(
a,
z)
=

e(
a,
z)
=
U(
a,
z)
=
V(
a,
z)
+
s(
a,
z)
,
(
6
.
1
)

utility
provided
by
an
alternative's
vector
of
characteristics,
a;

attributes
of
the
individual;

nonstochastic
component
of
utility,
describing
what
constitutes
representative
tastes
in
the
population;
and
stochastic
effect
reflecting
the
nondeterministic
effects
of
taste
on
decisionmaking
for
an
individual
with
attributes,
z,
facing
an
alternative
with
characteristics,
a.

6­
3
Prob
[
a
k
zj,
A]
=
Prob
[
U
k
>
Ui
for
all
i
#
k]
q
(
6
.
2
)
P
r
o
b
[
V(
ak,
Zj)
,)
­

E(
akl
`
j),
for
all
i
#
k]
­
V(
ai,
zj)
>
s(
ai,
z
.

By
making
distributional
assumptions
to
characterize
the
s's,
the
probability
statement
in
Equation
(
6.2)
can
be
defined
in
terms
of
the
characteristics
of
the
alternatives
and
the
features
of
the
individual.
For
example,
assuming
that
the
c's
are
independently,
identically
distributed
with
the
Weibull
distri
­
bution*
allows
the
probability
to
be
expressed
as
a
logistic,
as
in
Equation
(
6
.
3
)
:

`
Xp(
v
k)
prOb[
Uk
>
Ui
fOr
i
#
k]
=
exp(
Vk)
+
exp
(
V
i)
(
6
.
3
)

Before
the
relationship
of
random
utility
functions
to
contingent
ranking
is
explained,
several
observations
on
the
nature
of
these
functions
should
be
noted.
The
description
in
Equation
(
6.1)
is
a
conventional
treatment
(
see
McFadden
[
1974]
or
[
1981])
that
is
completely
general.
In
this
general
description
there
is
no
explicit
treatment
of
the
constraints
to
choice,
such
as
an
individual's
income
or
market
prices.
To
make
these
constraints
clearer,
it
is
completely
consistent
with
the
random
utility
model
to
view
V(
 
l
 
)
as
the
result
of
a
constrained
optimization
process.
Within
such
a
framework,
V(°)
would
be
an
indirect
utility
function,
reflecting
an
individual's
attributes,
the
characteristics
of
the
choice
alternatives
(
to
the
extent
they
are
not
reflected
in
market
prices),
the
individual's
income,
and
the
prices
of
the
alternatives
available
on
organized
markets.
T
Thus,
a
random
utility
function
framework
does
not
imply
that
the
conventional
economic
view
of
the
consumer
behavior
be
ignored.
Indeed,
as
McFadden
[
1981
]
has
suggested,
V(°)
can
be
regarded
as
an
indirect
utility
function,
even
in
applications
where
it
has
been
specified
as
linear
in
its
parameters.
This
interpretation
is
possible
because
any
continuous
function
can
be
approximated
to
any
desired
degree
of
accuracy
with
a
linear
specification
The
requirement
that
V(°)
be
homogeneous
of
degree
zero
in
income
and
prices
can
be
met
by
requiring
that
the
variables
in
the
linear
approximation
(
in
parameters)
be
homogeneous
of
degree
zero.
(
This
requirement
is
necessary
for
consumers
to
be
free
from
"
money
illusion"
to
changes
in
relative
prices
and
income.
)

*
The
distribution
function
for
the
Weibull
distribution
Prob(
Z
~
t
)
=
e
x
p
(
e
x
p
(
­
(
t
­
a
)
/
O
)
)
and
to
respond
only
is:

The
ordered
Iogit
is
derived
for
a
standardized
form
with
u
=
O
and
9
=
1.
This
implies
that
variance
of
the
errors
will
be
1.6449.
See
Chapter
20
of
Johnson
and
Kotz
[
1970]
for
more
details.

tThis
description
admits
the
possibility
of
a
model
comparable
to
the
hedonic
framework
used
in
modeling
property
values
(
see
Rosen
[
1974])
or,
more
recently,
adapted
to
a
travel
cost
recreation
demand
framework
by
Brown
and
Mendelssohn
[
1980]

6
­
4
Alternatively,
it
is
possible
to
assume
that
the
indirect
utility
function
is
separable
in
all
commodity
prices
but
the
ones
of
direct
interest.
Moreover,
in
principle,
these
prices
can
be
replaced
by
a
price
index
that
can
be
assumed
to
normalize
the
incomes
and
the
prices
of
goods
and
services
of
interest.
However,
it
should
also
be
acknowledged
that
this
approach
imposes
quite
restrictive
assumptions
on
the
structure
of
individual
preferences.
*
The
primary
conclusion
to
be
drawn
from
these
general
observations
is
that
conventional
neoclassical
models
of
consumer
behavior
can
be
used
as
an
integral
part
of
random
utility
models
when
the
utility
functions
are
interpreted
as
indirect
functions
describing
the
outcomes
of
households'
optimizing
decisions.

A
second
feature
of
the
models
used
in
the
random
utility
framework
stems
from
the
assumption
of
the
independence
of
irrelevant
alternatives.
This
assumption
is
important
to
the
structure
of
any
model
in
the
framework
because
it
implies
that
the
odds
of
one
alternative
being
chosen
over
a
second
alternative
are
not
affected
by
any
other
alternatives.
McFadden
[
1974]
has
conveniently
summarized
the
implications
of
this
assumption
in
discussing
the
limitations
to
the
random
utility
model:

The
primary
limitation
of
the
model
is
that
the
independence
of
irrelevant
alternatives
axiom
is
implausible
for
alternative
sets
containing
choices
that
are
close
substitutes.
.
.
.
application
of
the
model
should
be
limited
to
situations
where
the
alternatives
can
plausibly
be
assumed
to
be
distinct
and
weighed
independently
in
the
eyes
of
each
decisionmaker.
(
McFadden
[
1974],
p.
113)

With
this
background
on
the
random
utility
model
and
its
relationship
to
the
conventional
model
of
consumer
behavior,
it
is
possible
to
consider
the
contingent
ranking
methodology.
The
contingent
ranking
methodology
maintains
that
individuals'
valuation
of
environmental
amenities,
such
as
visibility
or
improved
water
quality,
can
be
described
within
a
random
utility
framework.
Thus,
an
approach
to
estimating
individuals'
values
for
changes
in
these
amenities
could
be
developed
by
estimating
the
deterministic
component
of
the
random
utility
function
­­
i.
e.
,
the
V(*)
in
Equation
(
6.1).
The
process
of
collecting
the
information
necessary
to
derive
these
estimates
involves
presenting
individuals
with
a
set
of
alternatives.
Each
alternative
describes
a
specific
state
of
the
world
in
that
it
characterizes
the
features
of
the
environmental
resource
and
the
cost
to
the
individual
of
having
access
to
the
resource
under
the
specified
conditions.
Individuals
are
then
asked
to
order
the
alternatives
from
most
to
least
preferred.
If
the
determinants
of
V(°)
are
known
and
it
can
be
approximated
using
models
that
are
linear
in
parameters,
the
ranking
of
the
alternatives
provides
sufficient
information
to
estimate
(
relative
to
a
scale
factor)
the
parameters
of
these
models,

*
Applications
of
these
principles
have
been
used
by
Hausman
and
Wise
[
1978]
.
The
restrictive
assumptions
required
are
discussed
In
detail
by
Blackorby
Primont,
and
Russell
[
1978].
Based
on
their
analysis
(
especially
in
Chapter
5),
this
approach­­
used
by
Hausman
and
Wise,
for
example­­
requires
separability
in
commodity
prices
(
called
indirect
separability
by
Blackorby,
Primont,
and
Russell)
and
additive
price
aggregation.
These
assumptions
imply
that
the
utility
function
will
exhibit
homothetic
separability.

6
­
5
The
contingent
ranking
methodology
provides
an
operational
basis
for
benefit
measurement.
However,
several
factors
should
be
considered
in
using
this
methodology
to
estimate
benefits
of
environmental
amenities.
Consistent
benefit
measurement
requires
recognition
of
the
constraints
on
individual
choice.
Thus,
to
define
compensating
variation
or
compensating
surplus
benefit
measures,
V(°)
must
be
considered
an
indirect
utility
function.
Moreover,
when
individuals
are
asked
to
rank
alternatives
that
involve
levels
of
an
environmental
amenity
and
a
fee,
the
role
of
the
fee
must
be
considered
within
an
optimizing
mode!
of
consumer
behavior.
That
is,
the
selection
of
the
payment
vehicle
may
have
an
important
effect
on
the
specification
of
the
random
utility
function.
For
example,
if
the
fee
included
in
each
alternative
is
a
user
charge
associated
with
gaining
access
to
the
resource
whose
features
are
also
being
described,
the
fee
would
be
treated
as
a
price
per
unit
of
use
of
the
resource.
Therefore,
it
would
enter
the
indirect
utility
function
in
a
format
comparable
to
any
other
price.
By
contrast,
if
the
fee
is
described
as
an
annual
payment,
regardless
of
how
much
the
resource
is
used,
it
would
be
expected
to
enter
as
an
adjustment
of
income
rather
than
as
a
price
per
unit
of
use
of
the
resource.
The
indirect
utility
function
can
be
expected
to
be
homogeneous
of
degree
zero
in
income
and
prices.
While
assumptions
that
can
simplify
the
form
of
the
function
and
the
number
of
distinct
prices
need
to
be
considered,
they
impose
significant
restrictions
on
the
types
of
features
of
demand
relationships
between
the
commodities
consumed
by
the
individual.
These
issues
are
discussed
in
more
detail
below.

The
required
assumption
of
independence
of
irrelevant
alternatives
limits
the
generality
of
the
contingent
ranking
methodology
for
benefit
estimation.
The
definition
of
the
alternatives
presented
to
individuals
in
a
contingent
ranking
is
largely
arbitrary
and
is
constructed
to
ensure
a
distinct
ranking
of
the
combinations
presented.
Indeed,
the
literature
to
date
has
not
explicitly
considered
the
issues
associated
with
experimental
design
in
selecting
the
alternatives
used.
While
this
problem
does
not
arise
in
application
of
the
model
to
alternatives
defined
by
what
is
available
in
the
real
world,
it
may
well
be
an
important
consideration
when
the
alternatives
are
specified
to
represent
feasible
alternatives
or
defined
to
provide
the
"
best"
estimates
of
an
individual's
compensating
surplus
for
a
change
in
an
environmental
amenity.

The
framework
used
for
benefit
estimation
(
and
described
later
in
this
chapter)
implies
that
the
level
of
environmental
quality
and
proposed
fee
are
subject
to
continuous
tradeoffs
as
each
varies
over
predefine
ranges.
This
presumption
is
quite
different
from
those
cases
for
which
McFadden
[
1981]
argued
the
random
utility
function
is
best
suited.
Thus,
even
a
brief
consideration
of
the
economic
theory
and
assumptions
underlying
conventional
formulations
of
the
random
utility
model
indicates
there
may
be
problems
with
its
use
in
the
contingent
ranking
methodology
as
a
procedure
for
benefit
estimation
Equally
important,
economic
theory
offers
some
guidance
in
selecting
the
most
appropriate
specification
in
empirical
applications
of
the
model.

6
­
6
~
.
.
.
.
.
.
.
 
 
______
.
.
..__.
_
____
­­
,­­_.,
_­
 
.
 
 
.
 
.
.
.
.
.
6.3
ESTIMATION
OF
RANDOM
UTILITY
MODELS
WITH
ORDERED
ALTERNATIVES
The
random
utility
model
can
be
estimated
using
the
information
provided
in
contingent
rankings
with
a
maximum
likelihood
estimator.
That
is,
once
the
additive
error
associated
with
each
individual's
utility
function
is
assumed
to
follow
a
probability
distribution,
the
decision
rule
given
in
Equation
(
6.2)
describing
how
each
individual
orders
the
available
alternatives
provides
the
information
necessary
to
describe
the
probability
of
a
specific
ordering
of
alternatives.
Of
course,
for
some
assumptions
concerning
the
probability
distribution
for
S(
 
l
 
)
,
the
form
is
simpler
than
it
is
for
others.
Nonetheless,
in
principle,
any
assumed
probability
distribution
provides
the
basis
for
describing
this
probability,
which
is
the
basic
ingredient
in
the
definition
of
the
likelihood
function
(
i.
e.
,
the
joint
probability
of
observing
all
the
orderings
given
in
a
specific
sample
as
a
function
of
the
parameters
of
the
utility
function
The
criteria
of
maximum
likelihood
estimation
can
then
be
used
to
derive
estimates
of
the
parameters
(
relative
to
a
scale
factor)
of
the
deterministic
portion
of
the
utility
function.

The
discussions
to
this
point
as
well
as
the
existing
applications
of
the
contingent
ranking
methodology
have
assumed,
for
analytical
convenience,
that
the
errors
follow
a
Weibull
distribution
in
deriving
an
ordered
Iogit
estimator
for
the
parameters
of
the
function
specified
to
represent
(
or
to
approximate)
v
(
"
)
.
Because
the
logic
underlying
this
derivation
has
been
outlined
in
Beggs,
Cardell,
and
Hausman
[
1981],
some
features
of
the
estimator
are
simply
highlighted
here
as
they
relate
to
the
Iogit
model
applied
to
problems
involving
discrete
choices
(
as
given
in
Equation
(
6.3))
versus
those
based
on
an
ordering
of
several
alternatives.

A
closed
form
expression
for
the
probability
of
an
ordering
of
the
alternatives
can
be
derived
using
the
properties
of
the
Weibull
distribution.
More
specifically,
the
conditional
probability
Prob(
Ui
~
t
Ui
>
Uk,
for
j
#
k)
differs
only
in
its
location
parameter
from
the
unc&
dition'al
distribution,
trated
for
this
two­
alternative
case
in
Equations
(
6.4a)
and
(
6.4b):

Prob(
Uj
~
t
)
=
exp
(­
exp(­(
Uj­
Vj))),
unconditional
distribution.

v.
v.
as
illus
­

(
6.4a)

Prob(
Uj
~
t
U
j
>
U
k
f
o
r
j
#
k
)
=
e
x
p
(­
exp(­(
U­
log
(
e
`+
e
`
)
)
)
)
.
(
6.4b)

Beggs,
Cardell,
and
Hausman
[
1981]
have
outlined
how
this
result
can
be
used
to
derive
the
probability
of­
an
ordering
of
alternatives
as
given
in
Equation
(
6.5):

Prob(
Ul
>
U
2
>
Us
>.
..>
U
h
)
=
~

k=
l
where
H
=
the
number
of
alternatives.
e
`
k
H
Vi
le
i=
k
f
(
6
.
5
)

6­
7
Equation
(
6.5)
describes
for
any
individual
the
probability
of
an
observed
ordering
of
alternatives.
Under
the
assumption
that
each
individual's
decisions
on
ordering
the
alternatives
are
independent
of
all
others,
the
likelihood
function
can
be
defined
for
a
sample
of
T
individuals
as:

By
specifying
the
determinants
pressed
in
terms
of
unknown,
T
H
[
1
evj
k
I­
II­
I
H
V..
(
6
.
6
)

j
=
l
k
=
l
~
eJl
i=
k
of
v.
the
likelihood
function,
L,
can
be
ex­
]
k'
estimatable
parameters.
Thus,
for
example,
if
v
.
is
described
by
Equation
(
6.7),
the
Ii"
kelihood
function
can,
for
a
given
.

]
k
sample,
be
where
Z
i
k
=
expressed
in
terms
of
the
unobservable
parameters,
P:*

v.
Jk
`
zik~
l
(
6
.
7
)

vector
(
lx
K)
describing
the
individual's
characteristics,
attributes
of
alternatives
being
ranked,
and
other
variables
as
detailed
by
economic
model
used
to
describe
behavioral
choice
vector
KxI
of
parameters
to
be
estimated.

Substituting
Equation
(
6.7)
into
Equation
(
6.6)
and
taking
the
logarithm
yields
the
log­
likelihood
function
for
the
ordered
logit
estimator.
~
Maximum
Ii
keli
­
hood
estimation
involves
solving
this
function
for
the
value
of
~,
which
maximizes
the
log­
likelihood
function.
In
most
cases,
this
solution
involves
numerical
optimization
procedures.
Our
analysis
of
the
Iogit
estimator
used
the
Davidon,
Fletcher,
and
Powell
[
1963]
(
DFP)
algorithm
with
numerical
partial
derivations.

The
second
estimator
for
use
with
information
from
contingent
ranking
was
developed
by
Keener
and
Waldman
[
1981
]
and
follows
the
same
behavioral
model.
In
the
Keener
and
Waldman
framework,
the
errors
associated
with
the
*
See
Section
6.3,
above,
for
a
description
of
the
relationship
between
a
general
form
for
the
Weibull
and
the
standardized
form
that
underlies
the
Beggs,
Cardell,
and
Hausman
[
1981]
derivations.

~
This
estimator
is
actually
the
same
method
proposed
by
Cox
[
1972]
for
dealing
with
duration
problems.
That
is,
Cox
proposed
a
conditional
Ii
keli
­
hood
model
based
on
ordering
the
variable
of
interest.
His
framework
maintains
a
proportional
hazard
formulation
of
the
problem.
The
two
likelihood
functions
will
be
identical
in
the
absence
of
ties
(
i.
e.
,
Cox's
analysis
allows
for
ties
in
the
ordering
of
the
dependent
variable,
while
the
ranked
Iogit
does
not).
.

6
­
8
_.
 
.
.._..__
~
a~
dom
utility
function
were
assumed
to
follow
independent
normal
distributions
T
h
e
p
r
o
b
a
b
i
l
i
t
y
o
f
a
n
o
r
d
e
r
i
n
g
o
f
a
l
t
e
r
n
a
t
i
v
e
s
i
s
d
e
s
c
r
i
b
e
d
b
y
the
multi
variate,
normal
cumulative
distribution
function
evaluated
at
`
r(!
2+
l
)
P
­
zr(
fl)
P/
~
=
1
,
2
,
.
.
.
,
H­
1
,
where
r(
n)
is
the
index
of
the
component
of
the
VeCtOr
of
utilities
for
a
given
individual
with
rank
!
2.
In
general,
the
~
olution
to
the
likelihood
function
for
the
normal
distribution
would
pose
a
difficult
nUMeriCal
integration
problem.
However,
Keener
and
Waldman
observe
.­
that
the
error
covar~
ance
matrix
is
tridiagonal
and
propose
a
computationally
tractable
method
of
numerically
evaluating
the
probabilities
composing
the
likelihood
function.
Thus,
the
likelihood
function
for
the
ranked
normal
estimator
is
derived
by
numerically
integrating
these
functions
to
obtain
the
probabilities
of
the
orderings
provided
by
each
sample
respondent.
Numerical
maximization
of
this
function
yields
the
Keener
­
Waldman
estimates.
The
DFP
algorithm
was
also
used
to
maximize
the
likelihood
function
associated
with
this
estimator.
Because
ranked
Iogit
is
globably
concave,
most
experience
with
the
method
indicates
it
converges
rapidly.
Thus,
estimation
with
the
ordered
logit
framework
is
comparatively
inexpensive.
By
contrast,
as
the
above
description
implies,
the
maximum
likelihood
estimator
based
on
the
assumption
of
normality
can
be
an
expensive
approach.
Consequently,
the
ranked
logit
method
has
been
used
here
to
examine
a
wide
array
of
alternative
specifications
for
the
deterministic
component
of
the
random
utility
function
and
the
ranked
normal
for
the
subset
of
those
models
that
were
judged
to
be
the
"
best.
"

6.4
PAST
APPLICATIONS
OF
CONTINGENT
RANKING
The
use
of
contingent
ranking
procedures
for
benefit
estimation
with
environmental
amenities
has
been
a
recent
development.
The
applications
have
been
exclusively
conducted
by
Douglas
Rae
of
Charles
River
Associates
and
have
focused
on
valuing
visibility
changes.
Our
review
considers
two
unpublished
reports
(
Rae
[
1981a,
1981
b])
describing
applications
of
the
methodology
*
Because
the
studies
were
largely
motivated
by
concern
over
the
benefits
associated
with
defining
alternative
visibility
standards
for
Class
I
areas
(
as
mandated
under
the
1977
Amendments
to
the
Clean
Air
Act),
the
surveys
have
been
conducted
at
fairly
unique
recreational
areas­­
the
Mesa
Verde
National
Park
and
the
Great
Smoky
National
Park.

The
experimental
design
used
in
the
two
surveys
was
quite
similar.
In
each
case,
a
sample
of
users
of
a
park
was
asked
to
rank
a
set
of
alternatives
The
set
was
composed
of
two
types
of
alternatives.
One
type
specified
combinations
of
conditions
for
the
park
where
the
survey
was
being
conducted
These
conditions
included
different
visibility
conditions
(
using
photographs
to
display
an
integral
vista
within
the
park),
a
recreational
quality
measure
(
generally
measured
by
waiting
time
at
a
key
landmark
or
availability
of
activities
at
a
park
service
center)
,
and
a
per
vehicle
entry
fee.
The
second
type
of
alternative
included
other
sites.
The
reports
are
not
clear
as
to
*
Since
the
draft
version
of
this
report
was
prepared,
a
third
application
(
Rae
[
1982])
to
visibility
changes
in
Cincinnati
has
been
undertaken,
but
is
not
considered
in
this
review.
Future
references
will
use
the
author's
name
[
Rae]
and
will
refer
to
these
1981
reports.

6
­
9
whether
comparable
attributes
were
reported
on
the
cards
used
to
describe
s
these
other
sites
or
whether
the
evaluation
of
the
characteristics
of
these
sites
was
left
to
the
respondents.
Table
6­
1
describes
the
features
and
selected
results
for
each
of
these
studies.

Each
respondent
was
asked
to
provide
two
rankings.
The
information
detailed
in
Table
6­
1
is
based
on
the
rankings
for
deterministic
conditions.
That
is,
in
the
cases
shown
in
Table
6­
1,
the
alternatives
were
explained
as
having
constant
visibility
at
the
level
prescribed.
In
addition
to
these
rankings
individuals
in
each
study
were
asked
about
alternatives
that
included
deterministic
and
probabilistic
descriptions
of
visibility
conditions
(
i.
e.
,
three
probabilistic
cases
and
four
with
constant
visibility
prescribed).
The
probabilistic
cases
specified
the
percentage
of
summer
daylight
hours
when
one
of
four
conditions
could
be
expected
to
prevail.
Unfortunately,
no
attempt
was
made
to
take
account
of
the
different
probability
structures
used
in
describing
the
visibility
conditions
in
the
estimation
of
the
random
utility
functions
from
these
rankings.

As
Table
6­
1
indicates,
the
empirical
results
from
these
studies
are
mixed.
The
entry
fee
was
found
to
be
a
significant
determinant
of
the
ranking
of
alternatives
in
both
studies.
However,
the
qualitative
variables
for
visibility
conditions
were
not
significant
determinants
of
utility.
The
Great
Smoky
results
were
somewhat
more
definitive.
They
indicated
that
serious
impairments
in
visibility
had
a
negative
and
significant
impact
on
the
level
of
utility.
However,
at
lower
levels
of
impairment
the
results
for
some
specifications
of
the
model
contradict
a
priori
expectations.

These
studies
are
important
because
they
demonstrate
an
alternative
approach
for
soliciting
individuals'
preferences
and
organizing
them
to
test
hypotheses
Nonetheless,
they
are
subject
to
some
shortcomings.

The
most
important
problem
arises
with
the
specification
and
interpretation
of
the
random
utility
function
estimated
in
these
analyses.
As
a
rule,
the
model
specifications
used
in
Rae's
analyses
of
the
respondents'
rankings
include
income,
the
suggested
price
for
use
of
the
area
(
i.
e.,
the
fee
included
as
an
attribute
of
each
alternative
that
is
ranked),
and
one
or
more
measures
of
the
postulated
visibility
conditions.
It
is
thus
clear
from
context,
though
never
explicit
in
the
studies,
that
the
functions
are
to
be
interpreted
as
indirect
utility
functions.
As
a
rule,
an
indirect
utility
function
would
include
the
prices
of
all
the
goods
and
services
consumed
by
the
individual,
not
simply
the
fee
proposed
for
use
of
the
relevant
recreation
site.
Since
these
prices
have
been
omitted
from
the
models,
it
must
be
concluded
that
an
implicit
assumption
consistent
with
one
of
the
appropriate
forms
of
aggregation
has
been
made.
There
are
two
possibilities
­­
that
all
remaining
goods
can
be
treated
as
a
Hicksian
composite
commodity
(
see
Deaton
and
Muellbauer
[
1980]
pp.
120­
122
for
discussion)
or
that
the
utility
function
exhibits
homothetic
separability
in
two
groups
of
commodities.
The
first
group
of
commodities
consists
of
the
services
of
the
site
under
evaluation
and
the
second
includes
all
other
goods
and
services.

6­
10
Table
6­
1.
Summary
of
Rae/
CRA
Contingent
Ranking
Studies
Description
Number
Area
of
of
specific
Sample
environmental
Benefit
Character
Recreational
alter­
alter­
b
estimates
Study
size
amenity
of
fee
quality
natives
natives
Design
choice
Empirical
findingsa
(
1981
dollars)

Cp
Great
213
Visibility
conditions:
Entry
fee
per
Availability
of
14
Smoky
intense
haze
vehicle,
$
0
to
full
program
moderate
haze
$
30
(
existing
of
visitor
slight
haze
fee,
$
0)
center
clear
 
Mesa
Verde
205
Visibility
conditions:
Entry
fee
per
Congestion
as
13
8
22
possible
combinations
intense
plume
vehicle,
$
2
to
measured
by
of
alternatives;
1
of
intense
haze
$
20
(
existing
waiting
time
10
sets
of
8
cards
moderate
haze
fee,
$
2)
at
landmark
randomly
given
to
clear
on
site
survey
respondents;
combinations
of
alternatives
always
include
current
conditions;
no
clearly
dominant
alternative
included
in
combinations
8
29
possible
combinaof
alternatives;
1
of
10
sets
of
8
cards
randomly
given
to
survey
respondents;
alternatives
always
Include
current
conditions;
no
clearly
dominant
alternative
included
in
combinations
Entry
fee,
negative
and
significant;
qualitative
variables
for
pcmr
visibility,
negative
and
insignificant
absence
of
congestion,
positive
and
significa~
t
Entry
fee,
negative
and
significant;
qualitative
variables
for
visibility
provide
some
evidence
for
vitiation
of
better
visibility;
intense
haze,
negative
and
significant;
absence
of
program
not
imtiort=
nt
Intense
haze
to
clear,
$
0.73
to
$
0.79
intense
plume
to
clear,
$
1.03
to
$
1.13
Intense
haze
to
clear,
$
7.39
to
$
11.22
intense
haze
io
slight
haze,
$
11.03
to
$
14.86
l
 
These
results
are
based
on
aggregate
models
and
use
conventional
criteria
for
significance
at
the
5
percent
level
with
asymptotic
t­
statistics.

bBased
on
the
aggregate
model.
Under
the
first
aggregation
assumption,
the
prices
of
all
goods
and
serv.
ices
(
other
than
the
site
under
study)
are
assumed
to
change
in
constant
proportion
and
this
proportion,
say
k,
would
be
the
relevant
argument
in
the
indirect
utility
function.
I
n
this
case,
because
of
the
nature
of
the
assumed
pattern
of
price
movements,
an
individual's
preference
for
one
good
in
the
set
cannot
be
distinguished
from
his
preference
for
any
other.
Ideally,
to
define
and
estimate
an
indirect
utility
function
consistent
with
theory
requires
a
sample
consistent
both
with
the
assumption
of
proportionality
in
the
price
movements
of
all
goods
and
with
some
variation
in
the
proportionality
cons
t
a
n
t
,
k
.
Since
both
of
these
conditions
are
not
often
realized
in
practice,
the
Hicksian
composite
commodity
theorem
is
difficult
to
use
in
empirical
applications
*
For
the
Rae
analyses,
there
is
no
way.
either
to
know
whether
the
prices
of
all
other
goods
and
services
change
in
a
proportional
relationship
across
all
individual
respondents
or
to
measure
the
magnitude
of
these
proportionality
constants.
These
unknowns
are
important
because
proceeding
under
the
assumption
that
a
Hicksian
composite
can
be
defined
and
then
arbitrarily
assuming
a
constant
value
for
it
across
all
individuals
in
a
cross­
sectional
data
base
is
equivalent
to
assuming
that
there
is
no
change
in
prices
across
individuals
If
the
respondents
all
come
from
a
single
geographic
area
(
i.
e.
,
in
a
region
immediately
around
the
site),
this
assumption
may
be
reasonable.
However,
based
on
evidence
of
substantial
regional
variation
in
prices,
this
implicit
assumption
is
untenable
for
sites
that
draw
visitors
from
around
the
nation.
Moreover,
to
the
extent
the
price
variation
is
not
simply
by
a
constant
multiple
for
all
goods
and
services,
the
assumptions
of
the
composite
commodity
approach
to
aggregation
would
be
violated.
t
*
lt
can
be
used
in
controlled
experiments
where
the
prices
confronting
an
economic
agent
(
i.
e.,
household
or
firm)
are
selected
by
the
analyst.
For
the
most
part
it
has
been
an
analytical
device
used
in
theoretical
analysis.
Indeed,
Deaton
and
Muellbauer
[
1980]
raise
comparable
reservations,
noting
that:

The
usefulness
of
this
theorem
[
i.
e.
,
the
Hicksian
composite
commodity
theorem]
in
constructing
commodity
groupings
for
empirical
analysis
is
likely
to
be
somewhat
limited.
.
.
.
in
an
open
economy
with
a
floating
exchange
rate,
considerable
fluctuation
in
relative
prices
can
be
expected
and
even
without
this,
it
is
not
clear
that
we
could
justify
the
types
of
aggregates
that
are
usually
available.
(
pp.
121­
122).

They
do,
however,
note
that
greater
justification
is
available
for
use
of
the
theorem
with
single
period
aggregation.

tThe
Bureau
of
Labor
Statistics
(
BLS)
data
used
to
derive
regional
cost
of
living
indexes
provide
evidence
of
both
variation
in
the
levels
of
prices
by
region
and
differential
patterns
of
change
among
these
prices
for
different
goods
and
services.

6­
12
A
approach
to
structuring
an
indirect
utility
function
so
that
it
The
5econd
the
models.
in
Rae's
analyses
would
involve
assuming
that
the
di
­
~
Oximates
@
utility
func~
ion
$
xhlblts
we~
k
separability.
That
is,
a
given
general
util
­
reCt
itY
function
U(
XI/
X2t
.
.
.,­
X
n),
with
Xi
the
ith
vector
of
goods
and
servw
r
i
t
t
e
n
a
s
U(
U1(
X1),
uz(~
z)
l
.
.
.1
Un(~
n))),
with
each
of
the
.
 
­
<
can
be
,
Ce=
I
(
uj(
 
l
 
)
)
a
hofnothetic
f
u
n
c
t
i
o
n
.
This
specification
implies
that
@
f(
JnctiOns
utility
function
can
be
expressed
in
terms
of
the
price
(
or
fee)
~
be
indirect
siteis
services,
income,
and
a
price
index
for
all
other
goods
and
f~~
the
This
price
index
can
be
normalized
at
unity
for
a
given
set
of
~
erviCes
o
for
the
PriCes
of
all
other
goods
and
services.
However,
if
it
is
values
to
be
unity
for
ail
r
e
s
p
o
n
d
e
n
t
s
,
it
is
implicitly
assumed
that
all
assumed
face
the
same
prices
(
or
different
price
sets
that
always
lead
to
respondents
value
for
the
index).
As
in
the
case
of
the
composite
commodity
a
UnitarY
the
plausibility
of
this
assumption
­­
i
.
e.
,
holding
the
price
index
aggregate
value
for
all
individuals­­
depends
upon
whether
or
not
respond­
at
a
COnstant
from
a
small
region
surrounding
the
site.
Otherwise,
some
varia­
ents
come
tion
can
be
expected,
both
in
price
and
in
the
value
of
the
price
index.

Aside
from
this
issue,
the
use
of
the
homothetic
separability
assumption
a150
restricts
the
nature
of
the
income
effects­
for
goods
within
each
grouping
­­
i,
e.,
subfunction
as
given
earlier,
as
Ui(
Xi)­­
and
the
nature
of
the
substitution
effects
for
commodities
involved
in
different
groupings.
To
illustrate
the
nature
of
these
constraints,
consider
the
case
of
Rae's
applications
where
the
utility
function
is
assumed
to
be
composed
of
two
groups
of
commodities
­­
the
services
of
the
site
under
study
and
the
set
of
all
other
goods
and
services
It
is
convenient
to
use
the
framework
of
conditional
demand
functions
to
illustrate
the
demand
effects
of
the
separability
assumptions.
*
For
example,

the
income
elasticity
of
demand
for
any
commodity
in
the
set
of
goods
and
services
(
other
than
the
site)
can
be
defined
as
a
product
of
the
income
elasticity
of
demand
in
the
conditional
demand
function?
and
the
elasticity
of
the
expenditures
on
this
set
of
goods
with
respect
to
income.
More
formally,
let
q.
designate
the
quantity
demanded
for
the
ith
commodity
in
this
set;
e,
the
e~
penditures
on
all
commodities
in
the
set;
and
y,
the
individual
income.
Thus,
if
q
is
the
use
of
the
relevant
site's
services
and
ps
is
the
price
per
unit
of
use,
s
e
=
Y­
Ps"
qs.
(
6
.
8
)

*
For
a
discussion
of
conditional
demand
functions,
1971].
s
e
e
Pollak
[
1969,
Summaries
of
his
work
are
available
in
Deaton
and
Muellbauer
[
1980]
.

~
This
elasticity
is
the
percentage
change
in
the
quantity
demanded
of
the
900d
with
respect
to
a
percentage
change
in
the
expenditures
on
all
goods
in
the
set.
These
expenditures
play
the
same
role
in
conditional
demand
functions
as
income
would
in
a
conventional
demand
function.
I
n
g
e
n
e
r
a
l
,
thee
determination
of
these
expenditure
levels
will
be
a
function
of
the
level
of
inCOme
and
the
prices
of
all
goods
and
services.
See
Blackorby,
Primont,
and
Russell
[
1978]
and
Pollak
[
1971
]
for
further
discussion.

6­
13
The
conditional­
demand
function
for
q.
will
be
goods
in
its
group,
pit
and
the
expenditure!
on
this
related
to
the
prices­
of
all
group
(
i.
e.,
qi
=
`
i(
pi,
e))

will
be
responsive
to"
income
and
the
prices
of
all
goods.
This
association
can
be
used
to
derive
the
following
relationship
between
demand
responses:

aqi
a!
~
e
3y­=
3E"
3j"
(
6.9)

In
elasticity­
terms,
Equation
(
6.9)
can
be
written
as:

(
6.10)

Homotheticity
of
this
subfunction
that
reflects
decisions
about
all
other
goods
implies
that
the
income
elasticity
in
the
conditional
demand
functions
for
these
goods
will
be
unity.
Thus,
the
first
term
on
the
right
side
of
Equation
(
6.10)
will
be
one.
Thus,
Rae's
model
implicitly
maintains
that
all
goods
consumed
by
the
individual
(
aside
from
site
services)
have
equal
income
elasticities
and
are
equal
to
the
expenditure
elasticity
with
respect
to
income.

This
analysis
can
be
extended
one
step
further.
Budget
exhaustion
implies
that
the
share
weighted
sum
of
the
income
elasticities
will
be
unity,
as
in
Equation
(
6.11)

n
Ks
*&+
Sy
ZKi~.=
l,
i=
l
Iy
where
Ks
=
share
of
income
spent
on
the
K
i
=
share
of
income
spent
on
the
site's
services
ith
commodity
%
y
=
income
elasticity
of
demand
for
a
site's
services
z.
=
Iy
income
elasticity
of
demand
for
the
ith
commodity.

Using
Equation
(
6.10)
to
substitute
for
&
iY
in
Equation
(
6.11)
gives
n
K&
2Ki=
l
,
s
Sy
+
`
e
y
i=
q
where
6­
14
(
6.11)

(
6.12)
While
homothetic
separability
of
the
utility
function
does
not
in
general
reit
does
have
implications
for
the
cases
in
which
it
would
be
likely
strict
`
sy
'

Rearranging
the
terms
in
Equation
(
6.12)
gives
to
be
Piausibie.

n
l­&
2Ki
e
y
i_
l
&
=
Sy
KS
"
(
6.13)

since
the
grouping
implicitly
required
for
Rae's
model
involves
all
other
goods
in
tie
set
designated
as
qi,
i=
l
,2,
.
.
.,
n,
it
is
reasonable
to
expect
that
&
would
be
close
to
unity.
That
is,
expenditures
on
the
majority
of
the
au
it~
ms
in
the
individual's
budget
are
likely
to
change
in
percentage
terms
as
income
does.
This
implies
that
the
income
elasticity
of
demand
for
site
services
will
aiso
have
to
be
ciose
to
unity
to
satisfy
the
adding­
up
condition
on
income
elasticities
(
i.
e.,
Equation
(
6.11
)).
Equation
(
6.13)
illustrates
this
conclusion.

Of
course,
this
conclusion
is
a
judgment.
Indeed,
the
appraisal
of
the
plausibility
of
using
the
composite
commodity
to
explain
Rae's
models
was
also
based
on
a
judgment.
What
is
at
issue
is
an
evaluation
of
the
implicit
assumptions
of
a
model's
specification
for
the
properties
of
its
results
(
or
conclusions
The
forgoing
appraisal
suggests
that
the
assumptions
necessary
to
interpret
Rae's
model
as
an
indirect
utility
function
are
fairly
stringent.
Both
sites
attract
visitors
from
substantial
distances.
Thus,
omitting
the
relevant
price
aggregate
for
other
goods
may
be
an
important
consideration
for
the
properties
of
the
estimates
of
compensating
surplus
derived
from
Rae's
indirect
utility
functions.

Regardless
of
how
one
judges
the
plausibility
of
the
assumptions
required
to
ignore
other
goods
and
services,
there
is
a
further
issue
arising
from
Rae's
definition
of
the
compensating
variation.
To
illustrate
the
proble"
m,
consider
an
example.
Assume
that
Equation
(
6.
14)
defines
the
deterministic
component
(
V)
of
the
random
utility
model,
which
is
assumed
to
be
a
function
of
the
individual's
income
(
Y),
the
entry
fee
(
F),
and
the
specified
level
of
visibility
(
v)
:

v=
alY
+
a2F
+
a3v
.
~
6.14)

Rae's
proposed
benefit
measure
is
the
increment
to
fee
that
must
accompany
a
change
in
visibility
to
hold
utility
constant.
When
Rae
assumes
that
dY
=
O,
this
increment
is
given
for
the
example
by
Equation
(
6.15):*

*
Assuming
dY
=
O,
this
is
derived
by
totally
differentiating
Equation
(
6.14)
as:
dV
=
aldY
+
a2d
F
+
a
3
dv
.

Holding
utility
constant
in
expected
value,
dV
=
O,
or
a2dF
+
a
3
dv
=
O
.

Solving
for
dF
gives:
dF=­
~
dv
Q2
"

6­
15
dF=­
aJ@
a
2
"
(
6.15)

Equation
(
6.
15)
is
not
com~
ensatina
variation.
This
Hicksian
measure
of
con­
 
­
.
.
su­
mer
surplus
is
defined
(
see
pag;
2­
4)
to
be
the
income
change
required
to
hold
utility
constant
in
the
presence
of
a
change
in
the
quantity
of
a
@
od
or
service,
such
as
visibility.

Thus,
the
interpretation
of
these
benefit
measures
depends
upon
the
type
of
fee.
If
it
is
a
fee
per
unit
of
use,
Equation
6.15,
strictly
speaking,
does
not
measure
compensating
variation.
Of
course,
the
extent
of
error
depends
upon
the
level
of
repeated
use.
If,
for
exampie,
"
users
are
expected
to
visit
the
site
only
once,
Rae's
measure
should
not
be
appreciably
different
from
one
based
on
the
income
changes.
However,
if
there
are
repeat
visitors,
it
maY
be
a
source
of
error
in
the
benefit
estimates.
In
pragmatic
terms,
as
shown
beiow,
the
use
of
price
versus
income
for
measuring
the
benefits
associated
with
a
specified
change
in
water
quaiity
markedly
affected
the
resu
Its.
Moreover
in
the
present
study,
the
fee
was
described
as
an
annual
payment
rather
than
a
price
per
unit
of
use.
*

There
are
severai
additional
probiems
with
these
studies.
The
Rae
applications
fail
to
include
respondents'
characteristics
in
the
estimated
utility
functions.
Presumably,
this
approach
was
adopted
because
two
models
were
estimated.
The
first
was
specified
under
the
assumption
of
constant
parameters
across
al
I
respondents
(
the
"
aggregate"
form).
The
second
permitted
these
parameters
to
be
different
for
each
individual.
Thus,
this
second
format
provides
the
flexibility
of
permitting
all
individuals
to
be
different
in
their
determinants
of
utility.
However,
to
estimate
a
modei
with
this
flexibility,
a
reasonably
large
number
of
ranked
alternatives
is
required.
It
is
not
clear
that
this
general
framework
is
heipful
to
interpreting
the
resuits.
Detailed
analysis
of
the
parameter
estimates
across
different
groups
of
individuals
would
be
necessary
to
understand
the
importance
of
an
individual's
attributes
in
determining
his
preferences
for
water
quality.

Despite
these
qualifications,
Rae's
applications
have
been
vaiuabie.
They
have
identified
a
new
approach
for
evaluating
individuals'
preferences
for
nonmarketed
goods
and
services,
and
they
have
contributed
to
an
understanding
of
the
issues
associated
with
using
the
random
utiiity
model
for
consistent
benefit
measurement.

6.5
MONONGAHELA
CONTINGENT
RANKING
EXPERIMENT:
DESIGN
AND
EST
I
MATES
Since
the
Monongahela
survey
was
designed
to
compare
approaches
for
measuring
the
benefits
of
water
quality
improvements,
one
section
of
the
ques
­

*
Since
Rae's
approach
has
been
foilowed,
and
since
the
role
of
the
prices
of
other
goods
and
services
has
been
ignored,
the
probiems
raised
earlier
as
judgmental
issues
may
also
have
contributed
to
these
findings.

6­
16
tionnaire
included
questions
designed
to
elicit
contingent
rankings.
There
are
several
important
distinctions
between
the
Monongahela
survey's
contln
­
~
ent
ranking
component
and
the
procedures
used
by
Rae.

For
the
Monongahela
survey,
individuals
were
given
a
smaller
number
of
alternatives
to
rank:
four
combinations
of
water
quality
and
annual
payments
higher
taxes
and
prices.
in
the
fOrm
Of
This
number
is
approximately
onethird
that
in
Rae's
experiments
and
affects
the
Monongahela
survey's
abi
Iity
to
estimate
what
Rae
describes
as
"
individual"
modeis.
*
Equaily
important,
~
hiie
all
the
Monongahela
survey
respondents
received
the
same
four
sets
of
alternatives
~
individuals
in
the
Mesa
Verde
and
Great
Smoky
experiments
were
randomly
assigned
one
of
ten
different
sets
of
alternatives
to
be
ranked.
sufficient.
experience
has
~
ot
Yet
been
acquired
with
the
estimators
of
these
models
to
Judge
the
Impllcatlons
of
this
difference
in
experimental
design.

A
further
distinction
arises
in
the
composition
of
each
set
of
alternatives
to
be
ranked.
The
procedure
used
in
the
Monongahela
survey
inciudes
four
sets
of
conditions
for
the
Monongaheia
River.
Tabie
6­
2
details
the
combinations
used,
and
Figure
6­
1
provides
an
example
of
the
cards
presented
to
each
respondent
for
ranking.
In
contrast,
the
Rae
surveys
inciuded
other
sites
in
the
set
of
alternatives
to
be
ranked.
Specificali
y,
the
Mesa
Verde
study
included
5
of
the
13
alternatives
as
other
sites,
and
6
of
14
alternatives
in
the
Great
Smoky
study
were
other
sites.
The
rationale
for
this
practice
was
described
as
an
attempt
to:

reflect
the
fact
that
alternative
sites
are
availabie
and
to
cause
respondents
to
focus
broadly
on
all
the
characteristics
of
a
site
that
contribute
to
overali
enjoyment
of
National
Parks
and
outdoor
recreation
areas.
(
Rae
[
1981
b],
p.
3­
1)

Of
course,
to
the
extent
that
one
accepts
the
assumption
of
independence
of
irrelevant
alternatives
that
underlies
the
random
utiiity
models
used
in
these
applications,
these
other
sites
should
not
be
important
to
the
rankings
provided
by
survey
respondents.
?

*
The
ordered
Iogit
estimator
permits
the
estimation
of
different
alternativespecific
effects
for
each
individual
in
the
sample
if
there
are
sufficient
alternatives
ran
ked.
S
e
e
Beggs,
Cardeil,
and
Hausman
[
1981]
for
a
discussion
of
the
identification
problem
In
such
cases.

Rae
refers
to
a
constant
parameter
model
for
ail
individuals
as
the
"
aggregate"
model
and
to
the
model
that
allows
variation
in
the
parameters
describing
the
effects
of
the
characteristics
of
alternatives
across
individuals
as
the
"
individual"
modei.

tThe
procedure
used
in
the
Mesa
Verde
study
involved
asking
respondents
first
to
rank
the
Mesa
Verde
alternatives
and
then
to
place
the
non­
Mesa
Verde
alternatives
within
the
ran
king.
Presumably,
the
same
procedure
was
used
in
the
Great
Smoky
study.

6­
17
 .

Table
6­
2.
Combinations
of
Water
Quality
and
Payment
for
Monongahela
Contingent
Ranking
Survey
Alternative
Water
quality
level
Annual
payment
2
3
1
E
RFF
water
quality
index
No
recreation
possible
D
RFF
water
quality
index
Boating
possible
c
RFF
water
cwalitv
index
$
5
=
0.8
=
2.5
=
5.0
$
50
$
100
4
$
175
Boating
and"
fishing
possible
B
RFF
water
quality
index
=
7.
O
Boating,
fishing,
and
swimming
possible
Figure
&
l.
Rank
order
card.

6­
18
FinallY,
the
PaYment
vehicle
included
in
the
rankings
conducted
by
Rae
was
a
price
per
unit
of
use­­
an
entry
fee
to
the
park.
By
contrast,
the
payment
vehicle
in
the
Monongahela
survey
was
independent
of
the
use
made
of
the
river.
It
is
therefore
an
adjustment
to
income.
This
distinction
affects,
~.
s
we
noted
earlier,
the
interpretation
of
the
specifications
for
the
random
utility
model.

The
rank
order
cards
used
to
describe
each
alternative
included
the
RFF
water
quality
ladder
as
described
earlier
in
Chapter
4
and
repeated
in
Figure
6­
1.
All
survey
respondents
were
asked
to
rank
the
four
alternatives
summarized
in
Table
6­
2.
In
making
these
judgments,
interviewers
were
i
n
­

s
t
r
u
c
t
e
d
to
refer
to
the
value
card
(
see
Figure
4­
6
in
Chapter
4)
and
to
ask
individuals
to
consider
actual
and
anticipated
use
of
the
river.
The
specific
question
used
was:

First,
I
would
like
you
to
rank
the
combinations
of
water
quality
levels
and
amounts
you
might
be
willing
to
pay
to
obtain
those
levels
in
order
from
the
card,
or
combination,
that
you
most
prefer
to
the
one
you
least
prefer.
I
would
like
you
to
do
this
based
only
on
your
use
and
possible
use
in
the
future
of
the
Monongahela
River.
That
is,
keeping
in
mind
only
Parts
I
and
I
I
of
the
value
card.

Two
hundred
thirteen
of
the
301
survey
respondents
provided
usable
rankings
and
family
income
information.
Thus,
they
provide
the
basis
for
the
empirical
analysis.
We
have
followed
Rae's
implicit
assumptions
and
interpreted
our
model
as
an
approximation
to
an
underlying
indirect
utility
function
However,
given
the
incomplete
information
on
an
individual's
other
consumption
choices,
we
have
not
attempted
to
include
the
prices
of
other
goods
or
to
impose
restrictions
on
the
nature
of
the
function
estimated.
A
variety
of
specifications
for
the
model
were
considered
under
this
general
format
and
the
"
best"
selected
based
on
the
ability
of
the
model
to
"
fit"
the
data
and
agreement
of
the
signs
of
the
estimated
parameters
to
a
priori
expectations.
The
final
section
of
this
chapter
discusses
the
implicat~
ons
of
extending
the
model
to
consider
the
role
of
other
prices
in
the
indirect
utility
function.

As
noted
earlier
in
this
chapter,
two
estimators
have
been
developed
for
random
utility
functions.
One
of
them,
an
ordered
Iogit
estimator,
was
used
in
Rae's
analysis
of
the
Mesa
Verde
and
Great
Smoky
contingent
ranking
results
,
Because
it
exhibited
rapid
convergence
and
performed
reasonably
well
in
unpublished
Monte
Carlo
experiments
performed
by
V.
K.
Smith
and
D.
Waldman
to
evaluate
the
estimators,
the
Iogit
has
been
used
to
screen
alternative
specifications
for
the
random
utility
model.
*
The
second
estimator
*
To
evaluate
the
relative
performance
of
the
ordered
Iogit
and
ordered
normal
models,
Smith
and
Waldman
[
1982]
conducted
a
limited
number
of
sampling
studies.
In
general,
each
estimator
performed
best
with
the
experiments
using
the
estimator's
assumed
error
(
i
.
e.
,
Weibull
for
ordered
Iogit,
normal
for
ordered
normal).
However,
the
ordered
normal
was
close
to
comparable
to
the
ordered
Iogit
with
the
Weibull
distribution.

6­
19
based
on
a
normal
specification
for
the
errors
in
the
utility
function
has
much
greater
computational
costs
and
was
therefore
applied
to
only
the
"
final"
model
specifications
for
comparative
purposes.
*

Table
6­
3
reports
a
selected
set
of
results
for
the
random
utility
model
with
the
ordered
Iogit
model.
Variables
describing
the
alternatives
ranked
and
the
features
of
the
individual
respondent
were
included
in
the
model.
The
models
are
distinguished
according
to
the
variable
used
to
interact
with
features
of
respondents
(
payments
or
water
quality);
the
specified
form
of
the
relationship
between
family
income
and
payment
in
the
model;
and
the
attributes
of
respondents
included
in
the
model.
Water
quality
was
measured
using
the
RFF
index
scale
as
it
appeared
on
the
rank
order
cards
presented
to
survey
respondents.
The
income
measure
is
family
income
in
thousands
of
dollars.
Age
(
in
years),
education
(
in
years),
race
(
1
=
white),
and
sex
(
1
=
male)
qualitative
variables
were
also
considered.
Three
additional
qualitative
variables
were
also
included
in
some
of
these
models:
boat
ownership
(
Boat
own
=
1
for
owners);
participation
in
any
outdoor
recreation
in
the
past
year
(
participate
=
1
if
yes);
and
the
individual's
attitude
toward
paying
for
the
costs
of
controlling
water
pollution
(
attitude
=
1
if
individual
considers
himself
very
or
somewhat
willing).

This
study's
results
provided
stronger
support
for
the
methodology
than
Rae's
findings.
Both
the
payment
and
water
quality
measure
are
statistically
significant
and
correctly
signed
in
most
of
the
model
specifications.
The
experimental
design
induced
a
high
correlation
between
payment
and
water
quality
(
simple
correlation
=
0.99),
and
this
may
explain
the
results
for
specification
(
2)
in
the
table.
Each
equation
in
the
table
has
three
columns
to
identify
whether
it
is
an
individual­
specific
variable
entered
individually
in
the
model
(
the
first
column)
or
a
respondent­
specific
variable
entered
in
interaction
form
with
either
the
payment
(
the
second
column)
or
water
quality
(
the
third
column).
Respondent­
specific
variables
must
be
entered
in
interaction
form
because
the
rankings
are
modeled
as
a
function
of
the
differences
between
the
values
of
the
deterministic
portion
of
the
random
utility
function
for
each
of
the
alternatives
being
ranked.
Consider
a
simple
example.
Let
,
J
designate
the
utility
individual
i
derives
from
alternative
j.
Individual
i
v..

will
rank
alternative
j
superior
to
alternative
k
if
V
ij
>
V
i
k.
Thus,
the
probability
that
alternative
j
is
ranked
ahead
of
k
will
be
equal
to
the
probability
t
h
a
t
V
i
j
>
V
i
k.
If
it
is
assumed
that
the
deterministic
component
of
V
is
a
linear
function
of
one
individual
characteristic
(
Z
li)
and
one
variable
describing
the
alternative
(
Zzj),
V..
can
be
rewritten
as:
lJ
Using
the
same
V
i
j­
v
i
k:
V
ij
=
ao
+
al
Z
li
+
a
2
Z
2
.
+
E..
J
IJ
(
6.16)

relationship
to
describe
V
ik
gives
the
following
expression
for
*
Comparability
between
the
results
of
Iogit
and
probit
models
for
bivari
­
ate
dichotomous
problems,
as
found
in
Hausman
and
Wise
[
1978],
do
not
necessarily
apply.
The
two
error
assumptions
will
yield
approaches
that
are
equally
comparable
with
ranked
data.

6­
20
~

b
­
 
.
.

Table
6­
3.
Selected
Results
for
the
Random
Utility
Model
with
Ranked
Logit
Estimator
a
MoM
and
 
l
 
ltm'native­
specific
interaction
(
1)
(
2)
(
3)
(
4)

Interaction
I
nterectlon
Interaction
Interaction
with
individual­
with
individual­
with
individual­
with
individualspeclflc
~
specific
~
specific
~
specificb
Alternative
variables
Alternative
variables
Alternative
variables
variables
Independent
variables
Alternative
specific
P
WQ
Spaciflc
P
WQ
specific
P
WQ
specific
P
WQ
Alternative
specific
Payment
(
P)
­
0.044
­
0.046
­
0.067
(­
0.236)
(­
8.922)
(­
3.957)

Water
quality
(
WQ)
­
0.151
(­
1.437)
0.030
1.364
1.919
(
3.025)
(
8.931)
(
4.121)

PxWQ
­
0.006
(­
9.342)

Individual
speciflcc
Income
(
x)
­
0.10
x
10­
3
(­
1.760)
(
J.
42
x
10+'
(­
0.002)
0.15
x
10­
4
(
0.800)
0
.
2
5
X
1
0­
3
­
0
.
4
3
X
10­
2
(
0.581)
(­
0.370)

Income
(+)

Partlclpate
(
x)
0.150
(
3.384)

Boat
own
(
x)
­
0.055
(­
0.967)
0.005
(
0.949)
­
0.137
0.402
(­
0.942)
(
1.022)

Age
(
x)
­
0.002
(­
1.280)

0.077
(
1.911)

0.016
(
2.762)

­
0.015
(­
0.247)
­
0.004
(­
3.342)
0.0004
­
0.015
(
1.435)
(­
1.846)

Sex
(
x)
0.075
(
1.732)

Education
(
x)

Race
(
x)
0.017
(
2.530)

Attitude
(
x)
(%%

LOO
(
L)
­
6S6.25
­
550.69
­
628.03
­
628.28
l
 
The
numbers
In
parentheses
below
the
estimated
pararnoters
aro
the
~
SyrnptotlC
t­
ratios
for
tha
null
hypothesis
of
no
association;
n
=
213.
(
continued)

b
The
columns
(
i.
e.,
P
or
WQ)
Indicate
which
interaction
is
used
in
each
modal
specification,

c
The
multiplication
signs
(
x)
indicate
that
the
individual­
specific
variable
is
entered
in
multiplicative
inter~
tion
with
either
the
payment
or
watOr
quality.
Tho
division
sign
(+)
indlcatas
that
incoma
is
l
 
ntered
in
as
a
division.

 
_.=.
 
 
_
.
.
­­
­,"
...
.
 
~
Table
6­
3.
(
continued)

0)

IE
Model
and
l
 
lternative­
spaclflc
interaction
(
5)
(
6)
(
7)
(
8)

Intoractlon
Intaractlon
Interaction
Interaction
with
individual
­
wlth
individual
­
wlth
individual
­
wlth
individual
­
apeclfic
~
spaclflc
~
specific
~
specificb
Alternatlva
variables
Altarnatlve
verlables
Alternative
variables
variables
Indapandsnt
varlablss
Alternative
l
 
pociflc
P
WQ
apeclflc
P
WQ
specific
P
WQ
specific
P
WQ
Alternative
specific
Payment
(
P)
­
0.
0s2
­
0.053
­
0.046
(­
9.769)
­
0.043
(­
6.215)
(­
6.101)
(­
7.764)

Watar
quality
(
WQ)
(
H%
0.959
(::%
0.706
(
6.520)
(
5.230)

pxWQ
Indlvldual
stmclflcc
Inwme
(
x)

Income
(+)

Pertlclpata
(
x)

Boat
own
(
x)

Age
(
x)

sax
(
x)

Education
(
x)
­
0.20
x
10­
5
(
0.035)

­
0.002
(­
0.677)

(
H%
­
0.280
(­
7.000)

­
0.003
(­
1.700)

0.0004
(
2.000)
­
0.273
(­
6.926)

­
0.0001
(­
0.066)

0.0006
(
3.374)
­
0.280
(­
7.000)

­
0.094
(­
1.709)

0.010
(
1.667)

Race
(
x)

Attltudo
(
x)
0.012
0.013
(
7.316)
0.351
(
6.944)
(
7.468)

Lq
(
L)
­
MM.
19
­
571.69
­
598.44
­
567.99
blha
~
In
parenthaaes
below
Us.
 
l
 
stlrnatad
parameters
ara
tha
 
l
 
symptotic
t­
ratios
for
the
null
hypothesis
of
no
association;
n
=
213.

%
0
~(
1
l
 
.
,
l
 
or
wQ)
IndlQte
which
Intarsctlon
Is
used
In
l
 
ach
mochl
epeciflcatlon.

Clhe
mitlplicetiert
 
l
 
loos
(.)
lndlcata
that
tho
Indlvldu.
1­
spaclflc
variable
is
entered
in
multiplicative
interaction
with
either
the
payment
or
water
quality.
The
dlvlslon
sign
(+)
Indlcatos
that
Inconm
Is
.
ntgrad
in
l
 
S
a
division.
Vij
­

`
ik
=
(
a
O
+
al
Zli
+
az
Z2j
+
&
ij)
­
(
a.
+
al
Zli
+
a
2
Z
2k
+
Gik)
(
6,17)

simplified,
this
expression
is:

V..
­
V
ik
=
a2(
Z2.
.­
Z
2k
)
+
.
si.
­
Sik
.
IJ
J
J
(
6.18)

Thus,
the­
var~
ables
describing
each
individual
are
not
involved
in
describing
how
that
Individual
ranks
alternatives
since
they
will
remain
constant
for
all
alternatives.
*

One
of
the
most
puzzling
aspects
of
the
results
is
the
effect
of
the
income
variable.
Because
the
payment
vehicle
was
constant
regardless
of
the
level
of
use!
the
multiplicative
interactions
between
income
and
the
payment
or
between
income
and
water
quality
would
have
been
expected
to
provide
better
results
than
income
divided
by
payment.
However,
the
results
indicate
that
the
income
divided
by
payment
f
o
r
m
i
s
a
s
i
g
n
i
f
i
c
a
n
t
d
e
t
e
r
m
i
n
a
n
t
o
f
t
h
e
u
t
i
l
i
t
y
function
implied
by
the
rankings,
while
the
other
forms
are
not.
In
all
cases,
the
signs
for
the
estimated
parameters
are
difficult
to
interpret.
~
p
r
i
o
r
i
expectations
wouid
have
suggested
that
income
relative
to
payment
be
a
positive
determinant
of
utility
and
not
negative.

Of
the
remaining
determinants
considered,
only
education
and
the
attitude
toward
paying
for
the
costs
of
controlling
water
pollution
were
consistently
significant
determinants
of
utility.
Both
variables'
parameters
are
consistent
with
g
priori
expectations.
Based
on
the
value
of
the
log­
likelihood
function
at
the
maximum
(
LOG
IL]
)
and
the
significance
and
consistency
of
the
estimated
parameters,
Specification
(
8)
was
selected
as
the
final
model.
It
was
reestimated
with
the
Keener­
Waldman
[
1981
]
ordered
normal
maximum
likelihood
estimator.
Table
6­
4
reports
these
results
along
with
estimates
for
Model
(
7)
for
comparison
purposes
and
repeats
the
ordered
Iogit
estimates
for
convenience
in
comparing
the
two
estimators
with
each
of
these
specifications.

The
two
estimators
yield
quite
similar
results.
The
signs
and
significance
of
estimated
parameters
are
comparable
for
the
final
model
and
for
Specificat
i
o
n
(
7
)
.
In
general,
the
Keener
­
Waldman
[
1981]
estimated
parameters
are
smaller
in
absolute
magnitude
"
than
the
ordered
Iogit.
There
are
no
specific
implications
of
this
difference,
because
both
estimators
involve
scaled
coeff
i
­
cients
and
the
estimated
parameters
do
not
correspond
to
the
marginal
effects
of
individual
variables
on
the
level
of
utility.
These
difficulties
in
evaluating
the
effects
of
the
estimator
on
the
conclusions
drawn
from
these
methods
suggest
that
the
Rae
measure
of
the
benefits
associated
with
a
water
quality
improvement
should
be
calculated
with
each
of
the
estimator
results
for
t
h
e
final
model
(
i.
e.
,
Specification
[
8]
).
These
results
will
be
considered
in
the
next
section
of
this
chapter.

*
See
Beggs,
Cardell,
and
Hausman
[
1981]
for
further
discussion
of
the
limitations
in
specifying
models
based
on
ordered
data.

6­
23
Table
6­
4.
Comparison
of
Ordered
Logit
anda
Keener­
Waldman
Ordered
Normal
ML
Estimator
(
7)
b
(
8)=
Independent
variable
Ordered
logit
Keener­
Waldman
Ordered
Iogit
Keener­
Waldman
Alternative
specific
Payment
(
P)
­
0.048
(­
8.101)
­
0,039
(­
7.073)
­
0.043
(­
7.764)
­
0.033
(­
7.196)

0.959
(
6.520)
0.760
(
5.630)
0.706
(
5.230)
0.510
(
3.400)
Water
quality
Individual
specific
Income/
p
­
0.273
(­
6.926)
­
0.070
(­
5.667)
­
0.280
(­
7.000)
­
0.170
(­
4.250)

­
0.039
(­
0.796)
Boat
own
­
0.094
(­
1
.709)
­­
­­

0.0006
(
3.374)
0.0006
(
3.000)
0.010
(
1
.667)
0.010
(
2.000)
Education
Sex
­
0.0001
(­
0.066)
0.0009
(
0.643)
­­

0.351
(
7.468)
0.330
(
8.462)
Attitude
­­
­­

Log
(
L)
­
598.44
­
619.46
­
567.99
­
582.34
a
The
numbers
in
parentheses
below
the
estimated
coefficients
are
asymptotic
t­
ratios
for
the
null
hypothesis
of
no
association.

bThis
specification
variables.

CThis
specification
specific
variables.
involves
payment
interaction
with
the
individual­
specific
involves
water
quality
interaction
with
the
individual­

6­
24
6.6
~
ENEF,
T
ESTIMATES
WITH
CONTINGENT
RANKING
MODELS
both
estimators
for
the
random
utility
model
provide
@
r19
~~~
ked
data
I
@
led
~
alue~
of
the
parame~
ers.".
AS
a
consewen.
ce,
the.
estimates
do
not
perof
the
utility
change
associated
with
a
change
In
water
~
lt
direct
evaluation
It
is
nonetheless
possible,
given
that
the
function
is
interpreted
as
Ity.
to
define
the
compensating
surplus
associated
with
@
I
utility
function
I
~
,
Odirect
water
quality"
Compensating
surplus
would
correspond
to
the
e!
i
`
n
&
in
income
that
just
Off
SetS
the
increment
to
utility
associated
with
the
Thus,
it
could
be
derived
by
taking
the
total
differen
­
.
ater
quality
change.
estimated
random
utility
function
with
respect
to
income
and
water
,,
a~
of
the
for
the
income
change
that
would
be
equivalent
(
in
its
and
by
SOlvif@
*
tiditY
on
utility)
to
any
water
quality
change.
This
approach
is
directly
,
ffects
to
the
definition
of
compensating
surplus
in
terms
of
the
expend
i­
~
nalO@
us
*
Thus,
in
principle,
the
model
can
be
used
to
derive
a
theore­
~
on~
istent
benefit
measure
for
changes
in
environmental
amenities.
[
Ure
function
o
~
lCallY
as
noted
earlier,
this
procedure
implicitly
assumes
that
the
indirect
~
oWeverf
is
theoretically
well
behaved.
~
~~
tlity
function
Rae's
Procedure
`
e
f
i
n
e
s
benefits
as
the
change
in
entry
fee
that
would
~
ff5et
a
chan9e
in
the
environmental
amenity
(
see
Equation
[
6.
1S).
The
bene­
,
it
measure
for
the
Monongahela
survey
was
also
defined
in
terms
of
a
total
~,
fferential,
measuring
the
change
in
payment
that
will
offset
a
water
quality
change.
As
we
noted
earlier
,
since
the
payment
vehicle
is
not
a
fee
per
unit
of
use
but
an
adjustment
to
income,
regardless
of
the
indiv~
ual's
use
of
the
river,
the
measure
`
f
c0mpensatin9
surPlus
should
be
invariant
to
the
use
of
income
or
of
the
payment
In
the
total
differential
equation.
If
the
indirect
utility
function
is
theoretically
consistent,
the
two
measures
should
be
equal
md
opposite
in
sign.

Of
course,
it
should
be
acknowledged
that
the
Monongahela
application
has
maintained
Rae's
basic
model
and
therefore
implicitly
assumes
that
ail
other
goods
and
services
are
either
part
of
a
Hicksian
composite
commodity
or
included
in
a
separable
homothetic
subfunction.
To
the
extent
neither
of
these
assumptions
provides
a
plausible
basis
for
treating
other
goods'
and
services'
prices,
estimates
of
compensating
surplus
will
likely
be
affected.
One
area
seems
to
be
an
especially
clear
example
of
the
limitations
of
this
aswmption
The
Monongahela
respondents
may
well
have
used
other
water­
based
sites
in
the
region.
These
sites
provide
services
that
substitute
for
what
is
*
See
Hause
[
1975];
Freeman
[
1979a];
and
Just,
Hueth,
a
n
d
Schmitz
[
1982]
for
further
details.

tThe
properties
of
an
indirect
utility
function
(
IDF)
include:

.
IDF
is
continuous
in
prices
and
income,

l
IDF
is
nonincreasing
in
prices
and
nondecreasing
in
income,

l
I
DF
is
quasi­
convex
in
prices,
and
.
I
DF
is
homogeneous
of
degree
zero
in
prices
and
income.
see
Varian
[
1978],
pp.
89­
92.

6­
25
.
proposed
for
the
Monongahela
sites
under
the
various
hypothetical
water
quality
scenarios.
It
must
be
expected
that
they
will
have
a
different
substitution
influence
than
the
remaining
goods
and
services
consumed
by
these
individuals
This
would
suggest
that,
at
a
minimum,
measures
of
the
"
prices"
(
i.
e.,
travel
and
time
costs)
of
trips
to
these
sites
should
be
included
in
the
specification
of
the
indirect
utility
functions
for
recreationists
in
the
sample.
*
In
addition,
it
implies
the
need
for
careful
consideration
of
the
relationship
between
whether
the
individual
was
a
user
of
the
sites
along
the
Monongahela
and
the
corresponding
specification
of
the
indirect
utility
function.
Since
the
models
used
in
this
study
do
not
reflect
these
considerations,
they
should
be
treated
as
fairly
crude
approximations
of
the
indirect
utility
functions
required
for
benefit
estimation.
T
The
exact
nature
of
the
estimating
equation
for
benefits
will
depend
upon
whether
the
individual­
specific
variables
enter
the
model
as
interactions
with
water
quality
or
with
the
proposed
payment.
To
illustrate
the
difference,
consider
two
simple
specifications
for
the
random
utility
function.
In
Equation
(
6.19),
the
model
includes
payment
(
P),
water
quality
(
WQ),
and
an
individual
specific
variable
(
Z)
using
a
payment
interaction,
whereas
Equation
(
6.20)
uses
the
water
quality
interaction.
Equations
(
6.21)
and
(
6.22)
report
the
corresponding
equations
for
measuring
the
payment
increase
equivalent
to
water
quality
improvements
for
each:

V
a
=
UIP
+
a2WQ
+
ci3P*
Z
.
(
6.19)

V
b
=
p~
P
+

`
p
=
­
e
`
p
a
y
m
e
n
t
#
12WQ
+
~
3WQ*
Z
.
(
6.20)

interaction
format).
(
6.21)

dP=­
(
P2
+
~~
z)
dwQ
(
water
quality
interaction
format).
PI
(
6.22)

It
is
clear
from
the
specifications
that,
in
either
Equations
(
6.21
)
or
(
6.22),
the
benefit
estimates
will
vary
with
the
individual­­
depending
on
the
individual­
specific
variables
entering
the
final
model
used
to
summarize
the
respondents'
rankings.
Table
6­
5
reports
the
average
and
range
of
benefit
estimates
for
the
final
specification
(
i.
e.
,
with
the
water
quality
interactions)
of
the
random
utility
model
for
u
s
i
n
g
both
the
ordered
Iogit
and
ordered
normal
models.
Because
the
final
specification
included
a
term
with
income
measured
relative
to
the
payment,
the
estimated
benefits
for
specified
water
*
These
issues
are
currently
being
considered
in
followup
research.

?
It
should
also
be
acknowledged
that
the
benefit
measures
calculated
with
the
income
change
were
several
orders
of
magnitude
greater
than
the
price
change
and
had
the
wrong
sign.
These
results
would
be
expected
because
the
estimated
parameter
for
the
income
variable
had
an
incorrect
sign
in
all
models.

.

6­
26
Table
6­
5.
Benefit
Estimates
from
Contingent
Ranking
Models
a
Estimator
b
Average
Range
I
Payment
=
5
Water
quality
change
=
Boatable
to
fishable
Ordered
Iogit
­
1
.
4
5
­
72.46
to
208.67
Ordered
normal
­
17.72
­
136.87
to
156.83
II
Payment
=
50
Water
quality
change
=
Boatable
to
fishable
Ordered
Iogit
62.76
39.74
to
83.31
Ordered
normal
64.30
38.54
to
85.51
I
l
l
Payment
=
100
Water
quality
change
=
Boatable
to
fishable
Ordered
Iogit
60.04
36.74
to
74.40
Ordered,
normal
62.12
36.27
to
78.40
iv
Payment
=
175
Water
quality
change
=
Boatable
to
fishable
Ordered
Iogit
59.47
36.12
to
72.66
Ordered
normal
61.65
35.80
to
76.96
v
Payment
=
5
Water
quality
change
=
Boatable
to
swimmable
Ordered
Iogit
­
2
.
6
2
­
130.42
to
375.61
Ordered
normal
­
30.91
­
246.37
to
282.30
VI
Ordered
Iogit
Ordered
norms
VII
Ordered
Iogit
Ordered
norms
Vlll
Ordered
Iogit
Payment
=
50
Water
quality
change
=
Boatable
to
swimmable
112.97
71.53
to
149.96
115.75
69.38
to
153.91
Payment
=
100
Water
quality
change
=
Boatable
to
swimmable
108.06
66.12
to
133.92
111.81
65.29
to
141.12
Payment
=
175
Water
quality
change
=
Beatable
to
swimmable
107.04
65.02
to
130.78
Ordered
normal
110.97
64.44
to
138.53
a
These
estimates
are
based
on
the
213
observations
used
to
estimate
the
random
utility
functions.

b
For
final
model,
Specification
(
8).

6­
27
­

quality
improvements
will
change
with
the
payment
level
at
which
~~
is
eval
­

uated.
The
results
in
Table
6­
5
are
presented
for
each
of
the
fou~
payment
levels
indicated
on
the
rank
order
cards,
as
well
as
for
each
of
two
water
quality
changes
­­
beatable
to
fishable
water
quality
and
boatable
to
swimmable
(
using
the
RFF
index
on
the
rank
order
cards).
The
results
are
clearly
implausible
for
the
lowest
payment
level
(
i.
e.
,
P
=
5).
Because
the
water
quality
change
represents
an
improvement
,
negative
values
imply
that
improved
water
quality
decreases
individual
well­
being.
However,
for
payment
levels
ranging
from
$
50
to
$
175,
the
benefit
estimates
are
stable
for
each
water
quality
change
(
i.
e.,
boatabie
to
fishable
and
beatable
to
swimmable)
and
are
approximately
the
same
order
of
magnitude
as
the
values
derived
from
direct
questioning
of
survey
respondents.
(
More
details
on
these
types
of
comparisons
are
provided
in
the
next
chapter.
)
These
estimates
should
be
interpreted
as
being
comparable
to
an
option
price
for
each
water
quality
change,
because
the
question
identified
both
use
and
anticipated
use
as
the
basis
for
the
ranking
solicited
from
survey
respondents.

The
benefit
estimates
derived
from
the
order
normal
model
seem
slightly
higher
than
the
ordered
Iogit
and
exhibit
a
consistently
wider
range.
Finally,
the
estimates
remain
quite
stable
as
the
payment
level
increases
from
50
to
175.
In
Appendix
C,
comparable
benefit
estimates
are
reported
for
a
model
using
payment
interactions
for
the
individual
specific
variables
(
see
Equation
[
7]
in
Table
6­
3).
For
this
case,
the
results
are
also
implausible
at
the
lowest
payment
level.
There
is
a
somewhat
larger
difference
between
the
ordered
Iogit
and
normal
estimates,
with
the
averages
for
Iogit
ranging
from
$
49.17
to
$
51.40
for
a
change
in
water
quality
from
boatable
to
fishable
(
and
payments
from
$
50
to
$
175)
versus
$
68.75
to
$
72.45
for
the
ordered
normal.
Nonetheless
these
changes
are
rather
modest
overall.
The
estimated
benefits
seem
quite
stable
across
the
alternative
specifications
of
the
random
utility
model.

6.7
IMPLICATIONS
AND
FURTHER
RESEARCH
This
chapter
has
described
and
applied
the
contingent
ranking
methodology
for
evaluating
the
benefits
from
changes
in
environmental
amenities
such
as
water
quality.
In
the
process
of
developing
the
background
for
this
approach
the
first
applications
of
the
approach
by
Rae
were
evaluated.
This
appraisal
indicated
that
the
empirical
results
yielded
a
relatively
weak
association
between
visibility
and
the
individual's
ranking
of
the
alternatives
describing
conditions
at
either
the
Mesa
Verde
or
Great
Smoky
Parks.
The
empirical
results
for
the
Monongahela
study
provide
much
stronger
support
for
the
method.
However,
analysis
of
the
theoretical
foundations
of
the
method
Rae
used
for
benefit
estimation
indicated
it
required
quite
stringent
assumptions
to
be
treated
as
an
approximation
of
a
theoretically
appropriate
benefit
measure.
It
should
be
acknowledged
that
the
evaluation
of
Raeis
approach
was
based
on
an
attempt
to
infer
the
implicit
assumptions
for
his
models.
The
underlying
behavioral
model
and
assumptions
were
not
explicitly
described
in
either
report.
Thus,
this
interpretation
should
not
be
attributed
to
his
reports.

.

6­
28
The
analysis
performed
here
has
begun
the
development
of
the
behavioral
underpinnings
for
the
random
utility
models
applied
to
contingent
rankings
of
alternatives
involving
environmental
amenities,
but
the
process
is
not
complete.
Models
estimated
with
samples
composed
of
users
and
nonusers
of
the
Monon
­
gahela
River
sites
have
been
used.
A
priori
expectations
would
suggest
that
nonusers
may
require
specifications
for
their
indirect
utility
functions
that
are
different
from
those
of
users.
The
latter
should
include
the
prices
(
i
.
e.
,
travel
costs)
for
all
the
relevant
substitute
sites
and
the
payment
as
an
adjustment
to
income.
By
contrast,
nonusers'
indirect
utility
functions
would
not
include
these
travel
cost
arguments.

Extensive
analysis
of
this
alternative
framework
for
modeling
respondents'
rankings
of
the
water­
quality/
payment
alternatives
was
beyond
the
scope
of
the
current
project.
The
primary
intention
of
this
analysis
has
been
to
apply
and
evaluate
the
Rae/
Charles
River
Associates
methodology
for
benefit
estimation
The
analysis
considered
the
appropriate
interpretation
of
their
proposed
benefit
estimator,
defined
an
approach
to
benefit
estimation
that
more
closely
approximated
a
theoretically
consistent
measure,
and
evaluated
several
models
with
two
estimators
of
the
random
utility
framework.

In
an
attempt
to
gauge
whether
these
model
revisions
would
be
important,
the
models
used
were
reestimated
for
a
subset
of
the
respondents­­
those
individuals
who
used
only
one
of
the
sites
on
the
Monongahela
River
(
i
.
e.
,
eliminating
nonusers
and
those
who
used
more
than
one
site).
For
this
sample
(
a
total
of
49
observations),
the
implications
of
treating
all
sites
as
perfect
substitutes
were
considered,
and,
therefore,
only
the
travel
cost
of
the
particular
site
used
was
entered.
The
results
with
the
ordered
Iogit
estimator
for
models
estimated
with
this
sample
under
these
assumptions
were
rather
poor
and
suggest
that
the
full
sample
of
users
and
a
more
complete
specification
of
the
model
will
be
required
to
judge
the
potential
importance
of
the
theoretical
arguments
calling
for
different
random
utility
models
for
users
and
nonusers.

6­
29
b
.
CHAPTER
7
A
GENERALIZED
TRAVEL
COST
MODEL
FOR
MEASURING
THE
RECREATION
BENEFITS
OF
WATER
QUALITY
IMPROVEMENTS
7.
I
INTRODUCTION
while
previous
chapters
have
considered
"
direct"
methods
of
eliciting
individuals'
valuations
of
water
quality
changes,
all
of
which
require
that
individuals
be
directly
asked
about
their
willingness
to
pay
for
water
quality,
this
chapter
describes
an
"
indirect"
method
for
benefit
estimation.
This
method
Uses
individuals'
actions
and
a
behavioral
model
that
describes
indi
­
~
idual
S'
decisions
in
order
to
infer
water
quality
values.
Specifically,
using
~
generalization
of
the
travel
cost
model
to
describe
recreation
site
demand,
this
approach
involves
describing
the
influence
of
recreation
site
characteristics
such
as
water
quality,
on
the
demand
for
a
site's
services.
To
accommodate
variations
in
demand
for
each
site's
services,
the
generalized
travel
cost
model
uses
variations
in
site
attributes
across
a
large
number
of
waterbased
recreation
facilities.

In
the
process
of
developing
the
model,
the
analysis
has
attempted
to
Consider
a
number
of
the
problems
associated
with
the
travel
cost
framework,
including
.

l
l
l
l
the
following:

The
estimation
ing
to
a
site.

The
treatment
relationship
to
of
the
opportunity
cost
of
the
time
spent
travelof
time
spent
at
the
site
during
each
trip
in
additional
trips
to
the
site.

The
specification
of
the
model,
including
the
prospects
for
biased
results
from
conventional
statistical
approaches.

The
implications
of
multiple­
purpose
trips
for
the
validity
of
`
the
model.

The
estimation
of
the
specific
effects
of
site
attributes
on
the
nature
of
each
site's
dem­
and
function.

This
chapter
discusses
each
of
these
issues
in
detail.
Specifically,
Section
7.2
reviews
the
economic
basis
for
the
travel
cost
model
using
Becker's
[
1965
]
household
production
framework,
and
Section
7.3
generalizes
the
conventional
treatment
of
the
travel
cost
model
as
a
derived
demand,
assuming
site
services
are
inputs
to
the
production
of
recreation
activities.
In
particular,
Section
7.3
considers
the
problem
of
modeling
site
attributes
in
developing
an
appropriate
7­
1
.
quantity
index
for
site
services,
and
it
proposes
a
variant
of
Saxonhouse's
[
1977]
generalized
least­
squares
estimator
to
implement
the
model.
Section
7.4
describes
the
recreation
choice
and
site
attribute
data
used
to
estimate
the
travel
cost
model,
and
Sections
7.5
and
7.6
present
and
evaluate
results
from
individual
site
demand
models.
In
these
two
sections,
as
throughout
the
chapter
generally,
a
major
objective
is
to
gauge
the
implications
of
modeling
decisions
for
each
site
demand
model
used
to
develop
the
generalized
travel
cost
model.
Section
7.7
presents
the
generalized
travel
cost
model,
and
Section
7.8
describes
its
use
to
estimate
benefits
with
survey
data
from
users
of
the
recreation
sites
along
the
Monongahela
River
in
Pennsylvania.
Finally,
Section
7.9
presents
a
brief
summary.

7.2
TRAVEL
COST
MODEL
The
travel
cost
model
is
widely
used
to
describe
demand
for
recreation
facility
services
(
see
Dwyer,
Kelly,
and
Bowes
[
1977]
for
a
review).
Indeed,
the
most
recent
Water
Resources
Council
[
1979]
guidelines
for
benefit­
cost
analysis
call
for
travel
cost
methods
to
estimate
the
economic
value
of
recreation
sites.
Although
the
travel
cost
model
is
usually
credited
to
a
suggestion
made
by
Harold
Hotel
ling
to
the
Director
of
the
National
Park
Service
(
that
distance
traveled
can
indicate
the
implicit
"
price"
recreationists
pay
for
using
a
particular
facility),
Clawson
[
1959]
and
Clawson
and
Knetsch
[
1966]
were
the
first
to
develop
empirical
models
based
on
it.
The
travel
cost
model
has
been
refined
since
this
early
literature,
and
it
is
now
recognized
as
an
important
indirect
methodology
for
valuing
environmental
amenities,
especially
water
quality
(
see
Freeman
[
1979a],
Chapter
8,
and
Feenberg
and
Mills
[
1980]).

Of
course,
recognition
of
the
travel
cost
model
has
not
come
without
the
parallel
development
of
a
behavioral
model
for
the
demand
patterns
it
describes.
For
example,
Becker's
[
1965]
household
production
model
can
analyze
individuals'
recreation
choices.
*
While
the
household
production
model
does
not
imply
new
testable
hypotheses
(
see
Pollak
and
Wachter
[
1975]),
it
does
offer
a
useful
conceptual
framework
to
describe
household
behavior,
especially
with
respect
to
outdoor
recreation.
~

The
absence
of
uniform
types
of
household
recreation'
data
and
the
lack
of
organized
markets
for
most
recreation
site
services
have
compounded
the
problems
of
describing
consumer
demand.
Therefore,
a
framework
that
can
be
constructed
using
the
available
recreation
data
has
distinct
advantages
over
frameworks
that
do
not.
Because
these
advantages
have
elsewhere
been
discussed
in
detail
(
see
Smith
[
1975a];
Deyak
and
Smith
[
1978];
Cicchetti,

*
ln
what
follows
Individual
and
household
are
used
synonymously.
Based
on
Becker's
[
197­
such
conventions
do
not
require
models
specifying
a
dictatorial
decision
process
for
the
household.
Rather,
households
can
be
seen
to
act
as
if
guided
by
a
single
utility
maximizer
when
altruistic
behavior
is
recognized
as
an
integral
component
of
the
social
interactions
of
family
members
(=
ee
Becker
[
1981]
`
for
more­
details).

?
It
can
also
provide
a
basis
for
consistent
welfare
measurement.
Bockstael
and
McConnell
[
forthcoming].

7
­
2
.
­
see
.
fi@
r/
and
Smith
[
1976];

redeveloped
here.
and
Bockstael
and
McConnell
[
1981
]),
they
wil
between
the
household
~
roduction
framework
not
The
basic
distinction
,,.
.

other
approaches
and
stems
from
its
portrayal
of
the
household
as
both
producer
That
is,
the
household
is
assumed
to
consume
only
services
>
~
mnsumer.
afw
:""'­

that
It
produces.
For
convenience,
these
services
will
be
designated
as
final
As
with
any
other
production
process,
these
services
req~
~
erviCe
.
f@!!
E"
In
this
case,
however,
the
inputs
involve
the
household's
time,
as
W
well
as
market­
purchased
goods
and
services.
Thus,
the
framework
considers
the
purchased
goods
as
an
indirect
means
to
maximize
utility.

The
household
production
framework
has
two
steps
or
stages,
which,

though
purely
logical
abstractions
,
can
explain
how
households
make
decisions.
The
first
step
Involves
selecting
market
goods
and
services
and
allocating
avaiia~
le
household
time
to
minimize
the
costs
of
each
possible
set
of
final
service
flOwS
.
I
n
the
second
step,
based
on
the
outcomes
of
the
first
step,
the
household
defines
for
itself
the
"
shadow
prices,
"
or
marginal
costs,
of
each
of
the
final
service
flows.
Thus,
along
with
the
relevant
"
full"
income
budget,
marginal
costs
are
implied
by
the
selection
process
for
final
service
flows
.

For
this
study,
constrained
utility
maximization
in
the
household
production
framework
highlights
several
important
aspects
of
the
travel
cost
model,
the
first
of
which
is
the
distinction
between
the
recreation
activities
undertaken
by
a
household­­
such
as
boating,
fishing,
or
swimming­­
and
the
usage
level
of
a
particular
recreation
site.
To
readily
identify
the
implicit
price
of
services
of
a
recreation
facility~
the
former
are
best
treated
as
measures
of
household
recreation
final
service
flows,
and
the
latter
are
best
treated
as
an
input
to
the
production
of
such
service
flows.

Furthermore,
the
household
production
framework
can
readily
identify
the
various
ways
site
services
are
used.
That
is,
the
framework
can
distinguish
whether
an
individual
uses
more
of
a
site's
services
by
visiting
it
a
greater
number
of
times
during
a
recreation
season
or
by
spending
more
time
at
the
site
during
fewer
visits.
This
choice
implies
a
simultaneity
problem
in
modeling
household
decisions
on
visits
and
onsite
time
per
trip.
Past
efforts
have
implicitly
avoided
this
problem
by
assuming
that
all
visits
(
across
all
users)
are
of
fixed
length
(
see
Cicchetti,
Fisher,
and
Smith
[
1976]
)
or
by
estimating
separate
models
for
each
trip
l
e
n
g
t
h
""(
Brown
and
!
vlendelsohn
[
1980]).

Finally,
the
household
production
framework
permits
a
general
discussion
of
a
household's
use
of
multiple
recreation
sites
that
produce
identical
recreation
activities,
thus
allowing
the
incorporation
of
site
attributes
as
determinants
of
the
differences
in
the
demands
for
the
services
of
multiple
sites.

In
its
simplest
form,
the
household
production
model
can
describe
recreation
decisions
by
simply
distinguishing
two
types
of
final
service
flows
produced
and
consumed
by
households.
The
first
is
the
recreation
service
f
l
o
w
,
Zr,
and
the
second
is
a
nonrecreation
service
flow,
Z
Because
nr"

7
­
3
sets
of
service
flows
can
be
expanded
without
fundamental
changes
in
the
implications
of
the
model,
the
present
analysis
has
been
confined
to
this
simple
case.
Following
earlier
developments
of
the
model
(
see
Cicchetti,
Fisher,
and
Smith
[
1976]
as
an
example),
the
production
function
for
recreation
services
can
be
specified
in
terms
of
five
inputs:
the
purchased
goods
associated
with
recreation
(
e.
g.
,
equipment
for
fishing,
boating,
camping,
etc.
),
X
;
the
number
of
visits
to
each
of
two
distinct
recreation
sites,
VI
and
V
2;
`
and
the
time
per
visit
to
each
site,
t
Y
and
t
It
is
important
to
V2
"
note
that
this
specification
greatly
simplifies
the
analysis
by
maintaining
that
onsite
time
per
visit
is
the
same
for
all
v"
isits
to
a
given
facility.

tion
The
Equation
(
7.1
)
provides
a
general
functional
services
production
fun"
ction:

Zr
=
fr(
xr,
VI,
v~,
t
vi
f
representation
tv2
)
.

time
horizon
for
production
activities
is
often
unspecified.
of
the
recrea
­

(
7
.
1
)

However,
the
household
must
be
assumed
to
make
decisions
over"
some
predefine
time
horizon
that
involves
a
full
recreation
season
(
or
some
fraction)
during
which
multiple
visits
to
different
sites
are
possible.

The
production
function
in
Equation
(
7.1)
implicitly
maintains
that
each
(
Vi,
tv
)
pair
ideally
measures
the
services
provided
by
each
site.
Thus,

this
fu~
ction
effectively
skirts
a
significant
index
number
problem*
because
differences
in
the
productivity
of
one
site's
services
for
the
recreation
service
flow
are
embedded
in
the
function
itself.
The
next
section
adds
further
assumptions
to
this
function
to
investigate
the
rationale
for
skirting
the
index
number
problem.

Because
the
focus
here
is
on
decisions
related
to
recreation
activities,
the
nonrecreation
service
flow
can
be
expressed
in
rather
simple
terms
as
related
to
nonrecreation­
re
lated
purchased
goods,
X_
,
and
household
time
spent
on
the
nonrecreation
service
flow,
t
n,
as
in
Equation''
(
7.2):

z
=
fnr(
xn,
t
n
)
.
n
r
(
7
.
2
)

*
1
ndex
number
problems
are
commonplace
in
the
application
of
macroeconomic
theory
to
real­
world
problems.
For
example,
measures
of
the
quantity
of
housing
pose
index
number
problems
because
houses
are
differentiated
by
number
of
rooms,
floor
space,
character
of
external
construction
(
wood
frame,
brick,
etc.
),
as
well
as
a
variety
of
other
features.
Because
it
would
not
reflect
these
differences
,
simply
counting
the
number
of
houses
is
insufficient
to
accurately
reflect
consumer
demand.
Similarly,
in
the
case
of
coffee,
muitiple
end
products­­
ground,
instant,
"
freeze­
dried,
"
decaffeinated
(
as
weil
as
combinations
of
these
attributes)­
­
makes
adding
pounds
of
coffee
consumed
an
insufficient
way
to
refiect
either
how
these
coffee
end
products
are
used
or
the
corresponding
features
of
consumer
demand.

7
­
4
I
.
 
l"
terms
of
its
relationship
to
practical
applications
of
the
travel
cost
one
of
the
most
important
aspects
of
the
household
production
frame­
~
odel
t
arises
with
the
definition
of
the
household's
budget
constraint.
Following
~
ork
IS
[
1965]
original
suggestions,
the
household
is
assumed
to
face
a
"
full
6ecke\,
~
onstraint,
Y,
including
wages,
wt
,
nonwage
income,
R,
and
foreincome
income~
L.
However,
it
is
not
assum%
d
that
the
household
necessarily
gone
treats
the
ma~
k~
t
`
a9e
as
the
opportunity
cost
of
its
time
in
all
household
This
formulation
can
be
seen
as
a
generalization
to
that
production
actlvltles.
in
Cicchetti,
Fisher,
and
Smith
[
1976].
Equation
(
7.3)
defines
this
proposed
budget
Constraint:

Y
=
Wt
w
+
R+
L
=
prxr
+
pnxn
+
(
T"
dl
+
r
t
l
+
w
l
t
JV1
(
7
.
3
)
+
(
T=
d2
+
rt
2
+
w
2
t
V2)
V2
+
Wt
n
,

where
P
P
n
=
r
'
the
prices
of
market­
purchased
recreation­
and
nonrecreation
­
related
goods
T
=
the
travel
cost
per
mile
d
i
=

r
=

ti
=

w.
=
I
the
roundtrip
mileage
to
the
ith
site
the
individual's
opportunity
cost
of
traveling
time
time
for
each
roundtrip
to
the
ith
site
the
individual's
opportunity
cost
for
onsite
time
at
the
ith
site.

Equation
(
7.3)
identifies
three
important
components
of
the
unit
cost
of
each
visit:
the
travel
costs
associated
with
the
vehicle
used
to
reach
the
site,
the
time
costs
of
the
trip,
and
the
opportunity
costs
of
time
spent
on
the
site.
Only
the
last
of
these
costs
is
a
choice
variable,
because
the
distance
and
time
to
reach
a
recreation
facility
are
defined
by
the
location
of
that
faci
I
ity
in
relation
to
the
individual's
origin
point.
Because
the
model
assumes
that
these
locational
choices
are
already
determined,
their
costs
are
outside
the
individual's
control.
*

*
Of
course,
this
statement
assumes
that
the
individual's
opportunity
cost
o
f
t
r
a
v
e
l
i
n
g
,
r
,
is
treated
as
a
fixed
parameter
to
the
recreation
decision
process.

7
­
5
i
1
The
past
literature
has
devoted
considerable
attention
to
the
appropriate
treatment
of
the
travel
and
time
costs
of
a
trip
in
the
formulation
of
travel
cost
demand
models.
Cesario
and
Knetsch
[
1970,
1976]
have
suggested
that
the
opportunity
cost
of
travel
time,
r,
is
less
than
the
wage
rate,
w,
and,
in
some
cases,
that
travel
and
time
costs
may
not
be
additive.
The
latter
component
of
the
Cesario­
Knetsch
argument
has
been
difficult
to
substantiate
without
dropping
the
assumption
that
the
opportunity
cost
of
travel
time
is
a
parameter
in
the
individual's
decision
process.

For
practical
purposes,
the
travel
cost
literature
has
tended
to
focus
on
the
relationship
between
the
cost
of
travel
time,
r,
and
the
wage
rate,
w.
Cesario,
for
example,
has
suggested
that
since
the
cost
of
travel
time
involved
in
urban
transportation
decisions
likely
falls
between
one­
fourth
and
one­
half
the
wage
rate
(
see
Cesario
[
1976],
p.
37),
one­
third
might
be
used
as
a
reasonable
approximation
for
travel
cost
models.
I
n
contrast,
McConnell
and
Strand
[
1981
]
have
estimated
the
fraction
to
be
six­
tenths
for
sports
fishermen
in
the
Chesapeake
Bay
region.
Their
model
assumes
that
the
opportunity
cost
of
travel
time
is
a
parameter
estimated
from
the
data
and
that
travel
costs
and
time
costs
of
travel
have
equivalent
effects
on
the
demand
for
a
site's
services.
McConnell
and
Strand
caution
that
this
parameter
may
vary
among
regions
and
sites.

The
only
notable
exception
to
the
treatment
of
r
as
a
multiple
of
the
wage
rate
arises
in
Wilman's
[
1980]
recent
attempt
to
compare
the
Cesario
and
McConnell
approaches
for
estimating
the
costs
of
recreation
trips.
Wilman's
analysis
sought
to
distinguish
"
scarcity"
and
"
commodity"
values
for
time
in
modeling
the
relationship
between
trips
taken
and
onsite
time
per
trip
to
produce
recreation
service
flows.
*
T
h
e
Wilman
model
specifies
utility
as
a
function
of
goods
and
services
requiring
time,
goods
and
services
not
requiring
time,
and
two
measures
of
a
recreation
site's
use­­
the
number
of
visits
of
a
given
length
to
a
site
and
the
number
of
roundtrips
to
that
site.
Roundtrips
are
intended
to
reflect
any
satisfaction
derived
from
traveling
to
the
recreation
site.
By
assuming
that
the
time
and
budget
requirements
are
fixed
multiples
of
the
number
of
visits
and
roundtrips,
Wilman
Iinks
these
choice
variables
to
the
household's
time
and
income
constraints.

The
basis
for
Wilman's
derivation
of
a
different
implicit
valuation
of
travel
and
onsite
times
is
an
assumption
that
the
number
of
trips
and
visits
to
a
site
are
equal.
The
resulting
first
order
conditions
require
equality
between
the
sum
of
the
marginal
utilities
of
trips
and
vi­
sits
and
the
corresponding
goods
and
time
costs
of
each
(
weighted
by
the
appropriate
Lagrangian
multipliers).

Wilman's
definition
of
commodity
and
scarcity
values
of
time
is
simply
a
rearrangement
in
this
allocation
condition
for
visits
and
trips
in
an
attempt
to
*
lt
should
be
noted
that
Wilman
did
not
explicitly
adopt
a
household
production
framework.
analysis
could
be
cast
in
However,
with
relatively
minor
amendments,
her
these
terms.

7­
6
for
the
potential
utility
derived
from
travel
time.
it
is
important
to
accoUnt
that
the
framework
maintains
that
trips
and
visits
are
delivered
rec09nize
jointlY
on
a
one­
to­
one
basis
in
this
version
of
her
model.
They
must
be
treated
as
a
single
commodity,
and
any
cost
allocation
between
them
is
arbitrary
Indeed,
once
the
equality
assumption
between
trips
and
visits
is
~
ropPed,
Wilman's
model
implies
that
both
types
of
time
should
be
valued
at
their
scarcity
value
(
see
Wilman
[
1980],
Equation
[
24]).
Thus,
the
existing
recreation
literature
does
not
provide
an
unambiguous
theoretical
justification
for
distinguishing
the
valuation
assigned
to
the
travel
and
onsite
time
com­
~
onentS
of
a
recreation
experience.

The
household.
production
framework
and
the
procedures
used
to
compiie
the
data
for
an
.
empl~
ical
estimation
of
trave!
time
costs
permit
direct
investigation
Of
the
`
eiatlonshl
P
between
the
travel
time
costs
and
the
onsite
time
costs
of
the
trip.
Therefore,
the
generalized
statement
of
distinct
opportunity
costs
for
each
time
of
travel
can
be
accommodated
within
the
empirical
model.

To
complete
the
model
it
is
necessary
to
maintain
that
the
household's
utility
is
a
function
of
the
levels
of
the
two
final
service
flows
produced
as
U(
zr,
Znr)"
Maximizing
this
utility
function
subject
to
the
budget
and
production
constraints
yields
a
set
of
conditions
that
can
be
manipulated
to
suggest
~
hat
the
marginal
utility
product
of
each
input
(
i.
e.,
the
product
of
the
mar91nal
utilitY
of
a
service
flow
times
the
marginal
product
of
the
input
in
the
production
of
that
service
flow)
relative
to
its
market
price,
or
implicit
unit
cost,
would
be
equalized
over
all
inputs.
More
formally,
this
result
is
given
in
Equation
(
7.4):

azr
azr
M
U
Z
r
ml
`"
zr3i72
=
(
T=
dl
+
r"
tl
+
wlt
v
1)
Wlvl
=
(
T"
dz
+
r=
tz
+
wzt
%)

azr
azr
azn
M
"
z
~
r
V2
Muzrm­
r
`"
Zn
wn
=
=
W2V42
P
r
=
P
n
(
7
.
4
)

azn
M"
zn
w
=
w
There
are
two
important
aspects
of
these
marginal
conditions.
First,
the
assumption
that
r
and
W
i
are
parameters
allows
all
aspects
of
the
costs
of
an
additional
visit
to
each
site
to
be
added
(
i.
e.,
the
full
cost
of
a
visit
to
the
ith
site
is
T*
di
+
r"
ti
+
witv
)
a
n
d
t
r
e
a
t
e
d
a
s
t
h
e
"
p
r
i
c
e
"
o
f
t
h
a
t
v
i
s
i
t
.

Second,
the
joint
determination'
of
trips
and
onsite
time
implied
by
this
formu
­

7
­
7
Iation
is
clearly
apparent
in
the
dependency
of
the
unit
costs
of
each
of
these
inputs
on
the
selected
levels
of
the
other.
*

Solving
the
necessary
conditions
of
a
utility
maximum
for
the
optimal
number
of
visits
to
each
site
as
a
function
of
the
parameters
in
the
optimization
problem
provides
the
analytical
counterpart
to
the
travel
cost
demand
model.
These
derived
demand
equations
can
be
written
in
general
form
as
Equations
(
7.5)
and
7.6):

VI
=
Ll(
wtw,
R
,
L
,
Pr,
P
n
,
T"
dl
+
r
t
l
,
T"
dz
+
r
t
2
,
Wl,
W
2,
w
)
,
(
7.
s)

V2
=
L2(
wtw,
R
,
L
,
P
r,
P
n
,
T"
d2
+
rt
2
,
T"
dl
+
r
o
l
l
,
W
I
,
W
2,
w
)
.
(
7
.
6
)

The
relationships
in
Equations
(
7.5)
and
(
7.6)
are
clearly
more
general
than
the
conventional
travel
cost
demand
model.
Empirical
estimation
of
these
relationships,
however,
requires
several
simplifying
assumptions.
Specifically,
full
income
(
wt
+
R
+
L)
is
assumed
to
be
approximated
by
family
income,
and
choices
of
`#
arket­
purchased
recreation
and
non
recreation
goods,
as
well
as
time
used
in
nonrecreation
final
service
flows,
are
treated
as
separable
decisions
in
the
consumer's
budget
allocation
process.
These
assumptions
reduce
the
input
demand
travel
cost
specifications.
(
7.7)
would
result:
equatio~
s
to
a
format
more
closely
resembling
the
I
n
the
case
of
the
first
site,
for
example,
Equation
T"
dl
+
rtl,
T"
d2
+
rtz,
Wl,
W
2
)
,
(
7
.
7
)

where
7=
family
income
as
a
proxy
measure
for
full
income
(
Y)
defined
in
Equation
(
7.3).

Before
turning
to
further
refinements
in
this
model
to
accommodate
the
introduction
of
specific
features
of
recreation
sites
as
determinants
of
the
variation
in
the
site
demand
functions,
it
may
be
useful
to
relate
the
amended
travel
cost
model
to
some
of
the
existing
travel
cost
studies.
(
A
comprehensive
review
is
available
in
Dwyer,
Kelly,
and
Bowes
[
1977
].)
It
is
acknowledged
at
the
outset
that
the
features
of
the
existing
work
can
often
be
explained
by
inadequacies
in
the
data
available
on
the
usage
of
recreation
sites.
Indeed,
many
travel
cost
studies
have
been
based
on
aggregate
visit
patterns
rather
than
on
information
on
the
behavior
of
individual
households.
These
data
are
typically
the
result
of
automobile
surveys
or
the
aggregation
of
user
permit
information
at
specific
recreational
sites.
However,
information
is
now
available
on
the
number
of
visitors
to
a
specified
site
from
a
set
of
*
This
framework
can
also
be
extended
to
consider
an
alternative
basis
for
deriving
a
relationship
between
the
opportunity
cost
of
travel
time
and
the
wage
rate
by
assuming
that
straints.
See
Smith,
Desvousges,
individuals
face
different
types
of
time
conand
McGivney
[
1983]
for
details.

7
­
8
~
one~
(
often
counties)
around
the
site.
With
such
information,
the
origin
of
site
usa9e
is
9enerall
Y
expressed
as
a
visitor
rate
(
i
.
e.
,
number
measure
relative
to
countY
Population)
and
is
interpreted
as
an
expected
of
usage"
for
the
"
representative"
individual
in
the
county.
County
of
visits
II
rate
statistics
are
used
as
indicators
of
the
economic
and
demographic
~
u~
fnarY
characteristics
of
this
"
representative"
individual.
As
a
consequence,
there
is
often
no
information
with
which
to
estimate
the
individual's
wage
rate.

I
n
the
presence
of
these
limitations,
researchers
have
taken
either
of
~
Wo
approaches.
The
first
assumes
a
constant
wage
rate
for
all
individuals
in
all
origin
z
o
n
e
s
.
The
second,
somewhat
more
desirable,
uses
an
estimate
based
on
the
wage
implied
by
average
family
income
in
the
origin
zone
(
i.
e.,
family
income
divided
by
an
estimate
of
hours
worked
per
year).
Clearly,
"
either
of
these
options
provides
a
discriminating
index
of
an
individual's
wage
rate.
However,
the
crude
nature
of
the
approximations
required
by
the
data
explain
in
part
Cesario's
[
1976]
willingness
to
propose
a
"
rule
of
thumb"
for
estimating
the
opportunity
cost
of
travel
time.

There
are
several
other
problems
that
arise
with
travel
cost
models
based
on
limited
data
sets.
The
first
of
these
stems
from
controlling
for
trips
o
f
d
i
f
f
e
r
e
n
t
l
e
n
g
t
h
s
w
i
t
h
a
n
a
g
g
r
e
g
a
t
e
d
a
t
a
s
e
t
.
In
some
cases,
researchers
have
separated
data
into
weekend
and
weekday
visits
to
ameliorate
the
problem
(
see
Cicchettij
Fisher,
and
Smith
[
19761
).
An
assumption
of
constant
onsite
time
is
otherwise
invoked
without
empirical
justification.
Equally
important,
the
nature
of
the
trips
may
be
quite
different
as
the
distance
from
the
site
increases.
That
is,
the
trips
may
have
multiple
objectives
that
would
imply
the
full
cost
of
the'
trip
is
not
an
implicit
price
for
the
use
of
the
recreation
site
but,
rather,
provides
other
services
as
well.
*

Recent
empirical
analyses
of
the
stability
of
the
travel
cost
model
using
data
aggregated
as
distance
from
a
site
increases
suggest
it
may
be
possible
to
detect
when
violations
of
these
assumptions
are
severe
(
see
Smith
and
Kopp
[
1980]).
Of
course,
this
analysis
requires
the
assumptions
of
constant
onsite
time
across
aggregated
visits
and
single­
purpose
trips,
which
are
more
untenable
as
the
distance
from
the
site
increases.

The
second
type
of
data
available
for
travel
cost
models
involves
sitespecific
user
surveys.
While
these
data
are
in
principle
superior
to
the
aggregate
visit
data,
incomplete
design
of
the
surveys
has
limited
their
ultimate
usefulness.
One
especially
important
omission
involves
the
treatment
of
usage
patterns
for
recreation
facil
~
ties
that
might
be
for
the
one
whose
users
are
questioned.
considered
substitutes
*
Haspel
and
Johnson
[
1982]
have
considered
this
users
of
the
Bryce
Canyon
National
Park
and
found
assumption
of
single­
purpose
trips
for
visitors
was
issue
for
a
survey
of
that
for
this
site
the
inappropriate.
Their
findings
suggest
that
`
it
would
lead
to
substantial
difference<
in
the
estimated
travel
cost
demand
functions.

7
­
9
The
micro­
level
data
from
the
surveys,
however,
have
permitted
the
investigation
of
a
number
of
issues
in
modeling
recreation
demand,
including
the
treatment
of:
travel
costs,
the
time
costs
of
travel,
and
the
costs
of
onsite
time.
Unfortunately,
these
efforts
have
not
been
entirely
successful.
For
example,
McConnell
and
Strand
[
1981
]
assume
that
increases
in
travel
cost
and
in
the
time
costs
of
travel
should
have
the
same
effect
on
site
demand
to
infer
the
relationships
between
the
opportunity
cost
of
travel
time,
r
in
this
study's
notation,
and
the
wage
rate,
w.
Their
data
do
not
include
wage
rates
that
require
estimation
from
family
income.
The
resulting
demand
equations
can
exhibit
difficulties
in
estimating
precise
(
i.
e.,
statistically
significant)
separate
estimates
of
the
"
price"
and
income
effects
on
site
demand.

The
most
recent
attempt
to
include
both
travel
cost
and
the
time
costs
of
travel
with
micro­
level
data
by
Allen,
Stevens,
and
Barrett
[
1981
]
concludes
that
it
is
difficult
to
distinguish
separate
effects
for
these
two
variables
when
time
is
entered
without
attempting
to
estimate
its
opportunity
cost
(
see
especially
pp.
178­
179).
The
authors
suggest
collinearity
would
seem
to
prevent
precise
estimation
of
separate
effects
of
the
two
variables.
Their
conclusions
contrast
with
earlier
suggestions
by
Brown
and
Nawas
[
1973]
and
Gum
and
Martin
[
1975]
that
disaggregation
would
help
to
resolve
these
estimation
problems.

Theory
does
imply
that
travel
time
should
be
valued
by
an
opportunity
cost
.
Thus,
the
Allen,
Stevens,
and
Barrett
findings
may
simply
be
a
reflection
of
a
failure
to
use
all
available
information
from
theory.
Moreover,
the
McConnell­
Strand
empirical
results
support
this
optimism.

One
important
aspect
of
any
attempt
to
include
both
travel
time
and
onsite
time
costs
of
a
trip
will
be
estimation
of
micro­
level
wage
rates
in
a
way
that
accurately
reflects
individual
rates
of
compensation
and
does
not
preclude
the
use
of
family
income
as
a
proxy
variable.
Such
a
method
is
developed
in
Section
7.4
of
this
chapter.

The
last
remaining
facet
of
the
idealized
travel
cost
model
given
in
Equation
(
7.7)
involves
the
treatment
of
the
influence
of
substitute
sites
on
the
demand
for
any
one
site's
Services.
This
model
explicitly
identifies
sites
that
can
contribute
to
the
production
of
the
recreation
service
flow,
and
it
thus
requires
an
approach
that
treats
the
effects
of
other
sites.
A
variety
of
methods
have
evolved
to
incorporate
the
influence
of
substitute
sites
on
demand.
Because
these
approaches
provide
a
natural
introduction
to
the
extended
travel
cost
model,
which
allows
a
site's
characteristics
to
be
determinants
of
intersite
demand
variation,
they
are
considered
as
a
part
of
the
introduction
to
the
proposed
model
in
Section­
7.3.

7.3
THE
TRAVEL
COST
MODEL
FOR
HETEROGENEOUS
RECREATION
SITES
As
noted
in
the
previous
section,
the
travel
cost
methodology
seeks
to
model
the
demand
for
a
recreation
site's
services.
In
general,
the
operational
forms
of
travel
cost
models
focus
on
estimating
site­
specific
demand
functions,

.­

7­
10
 
­
­
 
and
additional
sites
are
considered
only
to
the
extent
they
might
provide
services
for
a
Particular
site
under
study.
Conventional
practice
substitute
has
incorporated
the
role
of
these
substitute
services
using
one
of
three
methods
:

.
Incorporation
of
an
index
of
the
relative
attractiveness
and
availability
of
other
recreation
sites
into
the
relevant
site's
demand
function
(
see
Ravenscraft
and
Dwyer
[
1978]
and
Talhelm
[
1978]).

.
Specification
of
the
recreation
demand
models
to
include
the
prices
(
i.
e.,
travel
costs
and
time
costs
of
travel)
of
other
substitute
recreation
sites
(
see
Burt
and
Brewer
[
1971
]
and
Cicchetti,
Fisher,
and
Smith
[
1976]).

.
Respecification
of
the
utility
function
in
terms
of
the
attributes
of
recreation
sites
so
that
use
patterns
are
assumed
to
be
in
response
to
utility
maximizing
selections
of
these
attributes
(
see
Morey
[
1981
]).

Of
the
three
methods,
the
first
is
probably
the
least
desirable.
I
t
implicitly
assumes
that
­
an
arbitrary
index
can
account
for
substitute
sites
in
the
demand
fOr
any
given
recreation
site.
Of
course,
the
definition
of
such
an
attractiveness
index
.
not
only
requires
knowledge
of
the
exact
nature
of
the
substitute
relationships
but
also
assumes
that
the
index
form
would
be
a
simple
function
of
the
other
site's
attributes.
Thus,
this
approach
requires
the
very
information
it
is
attempting
to
derive.

The
remaining
approaches
are
consistent
with
economic
models
of
recreation
demand.
The
second
approach
can
be
interpreted
as
an
empirical
statement
of
the
model
given
in
Equation
(
7.7),
which
assumes
that
the
effects
of
substitute
sites
on
any
one
site's
demand
can
be
captured
through
the
specification
that
these
other
sites'
"
prices"
affect
the
demand
for
the
site
of
interest.
Because
the
demand
for
each
site
is
measured
individually,
the
second
approach
avoids
the
quantity
and
price
aggregation
issues
that
would
impede
the
consistent
definition
of
the
attractiveness
index
proposed
for
the
first
approach.

The
last
approach
addresses
the
quantity
and
price
aggregation
issues
directly
by
assuming
a
specific
format
for
them
in
the
site
attribute
specification
of
the
recreationist's
utility
function.
All
recreationists
are
assumed
to
have
the
same
preferences.
This
method
can
be
limited
by
the
plausibility
of
the
specification
of
the
utility
function.

However,
none
of
these
methods
offers
the
ability
to
consistently
relate
conventional
travel
cost
site
demands
to
the
site
features
that
produce
recreation
services.
That
is,
while
the
specification
of
the
household
production
function
for
Z
in
terms
of
several
sites
implicitly
reflects
the
prospects
for
substituting
o~
e
site's
services
for
another's,
there.
is
no
direct
means
for
explaining
the
reasons
for
the
degree
of
substitution
observed
between
any
Pair
of
sites.
This
inability
to
explain
the
source
of,
or
reasons
for,
these
7­
11
substitution
possibilities
is
not
a
limitation
for
many
applications.
As
noted
earlier,
when
sample
information
identifies
the
set
of
sites
considered
by
individuals
as
well
as
their
respective
patterns
of
use,
cross­
price
elasticities
of
demand
can
be
used
to
estimate
measures
of
the
substitution
possibilities.
Unfortunate
y,
this
information
is
not
uniformly
available
in
all
recreation
surveys.
Indeed,
this
study's
data
set,
described
in
the
next
section,
is
a
survey
of
users
at
specific
recreation
sites,
without
information
on
the
other
recreation
facilities
respondents
may
have
used
or
considered
using.
In
such
cases,
the
reasons
why
substitution
prospects
exist
between
recreation
sites
must
be
analyzed
and
some
attempt
made
to
reflect
them
in
the
modeling
of
the
overall
demand
for
these
sites.
In
simple
terms,
what
is
required
is
the
addition
of
further
structure
to
the
household
production
functions­­
assumptions
that
serve
to
explain
why
individual
site
services
contribute
differentially
to
the
production
of
recreation
service
flows
and
,
in
turn,
why
they
substitute
at
different
rates.

Before
the
analysis
is
formally
developed,
its
implications
must
be
described.
This
study's
approach
maintains
that
each
site
has
a
set
of
characteristics
(
e.
g.
,
size,
water
quality,
camping
facilities,
scenic
terrain,
etc.
)
and
that
these
attributes
contribute
to
site
productivity
as
inputs
to
recreation
service
flow
production
functions.
If
the
nature
of
these
contributions
is
restricted
to
a
specific
form,
originally
defined
as
the
simple
repackaging
hypothesis
in
problems
associated
with
constructing
quality
adjusted
price
and
quantity
indexes
for
consumer
demand
(
see
Fisher
and
Shell
[
1968]
and
Muellbauer
[
1974]),
the
measurement
of
the
role
of
site
characteristics
as
determina.
nts
of
the
features
of
site
demand
will
provide
an
explanation
of
the
substitution.
As
Lau
[
1982]
has
demonstrated
in
another
context,
the
simple
repackaging
hypothesis
implies
that
site
services
can
be
converted
into
equivalent
units
based
only
on
their
respective
characteristics.
Thus,
after
adjustment
for
their
attributes
(
with
Lau's
conversion
functions),
all
site
services
are
perfect
substitutes
for
each
other.
*
If
this
description
is
plausible,
a
model
of
site
demand
that
omits
consideration
of
the
role
of
potential
substitute
s
i
t
e
s
will
not
be
biased.
Of
course,
it
should
be
acknowledged
that
this
assumption
is
a
stringent
one
and
that
the
models
developed
from
it
may
be
limited
should
the
assumption
prove
to
be
a
poor
approximation
of
processes
giving
rise
to
substitution.

To
begin
the
formal
development
of
this
model,
the
original
specification
of
the
household
production
function
for
recreation
service
flows
(
i.
e.
,
Equation
[
7.
1
]
)
is
replaced
with
one
that
includes
the
characteristics
of
the
recreation
site,
Equation
(
7.
8):

Zr
=
f
r
(
X
r
,
Vi,
tv
,
a
i)
,
i
(
7
.
8
)
,

,

*
Berndt
[
1983]
has
also
recently
used
this
framework
to
describe
the
effects
of
input
quality
in
neoclassical
production
models.

7­
12
where
Xr
=

v
i
=

tv
=

i
a
i
=
recreation­
related
market
goods
number
of
trips
to
the
ith
recreation
site
time
per
trip
to
the
ith
recreation
site
(
assumed
to
be
constant
across
all
trips)

vector
of
attributes
for
the
ith
recreation
site.

In
this
form~
the
relationship
between
Vi,
tv
,
and
a
i
in
the
household
prod
u
c
tion
function
for
recreation
service
flobs
determines
the
appropriate
index
for
transforming
one
site's
services
into
their
equivalents
for
another
site.
More
speclflcally,
given
strict
monotonicity
of
the
household
production
function,
Equation
(
7.8)
can
be
solved
for
Vi.
*
This
resulting
function
might
be
designated
a
site­
service
requirements
function
and
would
be
given
(
in
general
form)
by
Equation
(
7.9):

vi
=
h(
Zr,
X
r
,
tv
,
a
i)
.
i
(
7
.
9
)

Thus,
to
convert
one
Site's
services
into
equivalent
units
of
another
site,
the
ratio
of
the
equivalent
h(.
)
functions
for
each
site
is
needed.
?
For
example,
if
there
are
two
sites
(
designated
With
subscripts
1
and
2),
and
if
the
differences
in
the
production
technologies
for
Z.
using
each
site
can
be
captured
with
a
i,
the
equivalence
between
trips
to
each
is
given
by
Equation
(
7.10):

h(
Zr,
X
r
,
tv
,
al)
VI
=
hz
r,
X
r
,
t
,
az)
"
v~.
V2
(
7
.
1
0
)

This
relationship
can
be
further
simplified
if
the
ratio
V1/
V2
is
assumed
to
be
independent
of
Zr,
X
r,
and
tv
(
i
=
1,2).
*
Under
this
assumption,
the
i
*
A
monotonic
function
implies
that
there
is
a
one­
to­
one
association
between
the
set
of
independent
variables
and
the
dependent
variable.
In
the
context
of
a
production
function
this
assumption
implies
that,
if
an
output
Q
can
be
produced
with
a
certain
input
bundle
x,
the
same
output
can
be
produced
with
more
of
every
input
(
provided
it
is
possible
to
costlessly
dispose
of
what
is
not
needed).

tThis
analysis
of
the
role
of
site
characteristics
adapts
work
recently
developed
by
Lau
[
1982]
for
the
definition
and
measurement
of
a
raw
materials
aggregate
within
neoclassical
models
of
production.

+
The
assumption
of
independence
of
tv
can
be
easily
modified
by
incorporating
it
as
one
of
the
set
of
attributes"
$
ssumed
to
be
available
with
each
visit
to
the
site.
Indeed,
this
format
is
equivalent
to
the
assumption
made
earlier
that
onsite
time
is
the
same
for
all
visits.

7­
13
site­
service
requirements
function
would
be
given
as
Equation
(
7.11
),
and
the
household
production
function
corresponding
to
it
by
Equation
(
7.12):

where
R(
ai)
=
the
vi
=
6
(
Zr,
Xr,
tv
)
l
 
R(
ai)
,
(
7.11)
i
Zr
`
?
r
(
X
r
,
tv
,
R(
ai)
 
l
 
Vi)
,
(
7.12)
i
augmentation
function.

R(
ai),
the
augmentation
function,
provides
a
specific
index
that
permits
each
site's
services
to
be
transformed
into
equivalent
units.
it
maintains
that
this
transformation
will
be
constant
regardless
of
the
level
of
the
site's
services
used
and
will
only
vary
with
changes
in
the
attributes
(
the
ai's)
for
a
site.
Consequently,
the
augmentation
function
describes
how
sites
would
substitute
for
each
other
in
the
production
of
the
recreation
service
flow,
Z
r"
This
form
of
the
household
production
function
­­
used
in
the
following
discussion
­
­
i
m
p
l
i
e
s
t
h
a
t
t
h
e
e
f
f
e
c
t
s
o
f
a
site's
characteristics
on
household
demands
for
that
site's
services
can
be
derived
if
households
can
be
viewed
as
engaged
in
a
two­
step
optimization
process
to
allocate
their
time
and
resources.
*
One
of
these
steps
involves
minimizing
the
costs
of
producing
a
given
output,
suggesting
that
the
patterns
of
trips
to
recreation
sites
will
be
adjusted
so
the
relative
unit
costs
of
a
trip
to
any
pair
of
sites
would
be
proportionate
to
their
respective
marginal
products
in
contributing
to
the
recreation
final
service
flow.
In
other
words,
the
effective
price
of
a
site's
services
will
be
equalized
across
all
recreation
facilities
considered
for
use
in
the
production
of
the
recreation
service
flow.

If
the
prices
of
site
services
are
equalized
across
sites,
the
augmentation
function,
R(
a.
),
provides
the
means
of
relating
each
site's
marginal
product.
T
h
u
s
,
f
o
r
ex~
mple,
using
the
augmentation
function
to
compare
two
sites
with
different
levels
of
water
quality
(
one
with
levels
permitting
recreation
fishing
and
the
other
permitting
only
boating),
this
distinction
is
captured
analytically
by
a
higher
augmentation
coefficient
for
the
site
with
cleaner
water.
Designating
the
sum
of
the
travel
costs
and
time
costs
of
travel
by
Pi
(
i.
e.,
Pi
=
T.
di
+
r*
ti)
then
yields:?

P,
.
ALt
R(
al)
R(
a2)
(
7.13)

or
the
equivalent
of
a
hedonic
price
function
for
sites'
services:

Pi
=
g(
ai)
.
(
7.14)

*
For
f
u
r
t
h
e
r
d
i
s
c
u
s
s
i
o
n
o
f
t
h
e
application
of
the
household
productiotl
model
to
modeling
outdoor
recreation
behavior,
see
Deyak
and
Smith
[
19781
and
Bockstael
and
McConnell
[
1981]
.

7This
relationship
assumes
that
onsite
time
is
constant
and
equal
for
both
sites
and
that
the
opportunity
costs
of
onsite
time
are
equal
for
the
two
sites.

7­
14
l­
his
approac
h
is
simply
an
alternative
derivation
of
the
first­
stage
estimating
for
the
Brown­
Mendelsohn
(
1980)
hedonic
travel
cost
model.
It
does
equation
not,
hOWever,
necessarily
imply
that
the
marginal
prices
of
attributes
will
be
,.­.
constant.
*

A
second
implication
of
the
above
approach
is
that
the
household's
cost
~
unction
for
producing
Zr
will
be
a
function
of
the
site's
attributes.
More­
­,,
ar
.
the
attribute
augmentation
function,
R(
a.
),
will
adiust
the
effective
Uv­,
o
price
of
the
site's
services
in
the
(
7.15):

c
=
C(
zr,

where
P
r
=
price
of
recreation
related
W
i
=
price
of
onsite
time.
household's
~
ost
function,
as
in
Equation
P
r,
W
i,
Pi/
R(
ai))
,
(
7.15)

commodities
This
cost
function
provides
the
basis
for
a
generalized
travel
cost
model.

It
is
assumed
that
a
given
recreation
site's
attributes
do
not
change
during
a
recreation
season.
Thus,
estimates
of
a
travel
cost
recreation
demand
for
a
sin91e
site
cannot
iSOlate
the
role
of
these
attributes.
Nonetheless,
these
characteristics
should,
in
principle,
affect
the
form
of
these
demand
functions
across
sites,
as
seen
when
Equation
(
7.
15)
is
differentiated
with
respect
to
the
site's
price,
Pi.
Following
Shephard's
[
1953]
lemma,
the
partial
derivative
is
the
individual's
demand
that
this
demand
must
be
vi*
.
K
a
p
i
for
the
site's
services.
Equation
(
7.
16)
illustrates
a
function
of
the
site's
characteristics:~

1
 
C4(
Zr,
P
r,
W
i
,
Pi/
R(
ai))
.
=
R(
ai)
(
7.16)

To
make
the
framework
in
Equation
(
7.16)
operational,
a
number
of
complications
must
be
considered.
The
first
of
these
issues
involves
the
recreation
service
flow,
Z
r,
for
which
there
is
no
measure.
As
a
rule,
the
*
The
Brown
­
Mendelsohn
[
1980]
hedonic
travel
cost
model
proposes
a
twostage
framework.
In
the
first
stage,
the
hedonic
price
function
is
estimated
for
each
origin
zone
by
considering
the
set
of
recreation
sites
available
to
users
in
that
zone,
their
respective
travel
and
time
costs
for
trips,
and
their
attributes.
With
these
data
a
separate
hedonic
price
function
can,
in
principle,
be
estimated
for
each
zone.
The
partial
derivatives
of
these
price
functions
(
which
are
assumed
to
be
linear
in
their
application)
define
the
implicit
prices
of
the
sites'
attributes
for
users
in
each
zone.
Using
the
recreation
site
choices,
their
implied
levels
of
attributes,
and
these
implicit
prices
for
attributes
Brown
and
Mendelssohn
then
estimate
demand
functions
for
each
attribute
across
all
origin
zones.

~
C4(')
is
a
short­
hand
expression
for
the
partial
derivative
of
the
cost
function,
C(
 
l
 
)
,
with
respect
to
its
fourth
argument,
Pi/
R(
ai).

7­
15
flows
are
not
part
of
travel
cost
demand
model
s­­
an
exclusion
that
is
justified
if
the
production
technology
is
homothetic
and
if
the
levels
of
production
are
uncorrelated
with
other
determinants
of
the
demand
for
a
site's
services.
*
That
is,
the
first
assumption
implies
Equation
(
7.16)
can
be
rewritten
as:

y=
H(
zr)
 
l
 
g(
pr,
Wi,
pi/
R(
ai),
R(
ai))
.
(
7
.
1
7
)

Rewriting
Equation
(
7.17)
in
logarithmic
form
yields:

In%=
I
n
H(
Zr)
+
In(
g(
Pr,
Wi,
pi/
R(
ai),
R(
ai)))
.
(
7.18)

When
Z
is
uncorrelated
with
the
arguments
of
g(.
),
and
when
ln(
g(.
))
is
linear
i~
parameters,
the
ordinary
least­
squares
estimates
of
these
parameters
wiil
be
unbiased.
~
Of
course,
this
framework
assumes
that
all
individuals
produce
the
same
types
of
activities.
+

The
second
complication
arises
from
the
treatment
of
onsite
time.
The
model
developed
in
Section
7.2
described
the
cost
of
a
site's
service
by
considering
the
travel
and
time
costs
of
traveling
to
the
site
and
the
time
spent
at
the
site
per
visit.
For
simplicity,
the
time
spent
onsite
was
assumed
constant
for
all
visits.
Thus,
the
full
cost,
C
i
,
of
all
trips
to
the
ith
facility
is
given
as:

where
T
=

d
i
=

r
=
Ci
=
(
T"
di
+
rt
i
+
Witv
)
Vi
,
i
travel
cost
per
mile
(
operating
costs
for
an
automobile)

roundtrip
distance
in
miles
opportunity
cost
of
traveling
time
(
7.19)

*
Any
production
function
that
can
be
written
as
a
monotonic,
increasing
function
of
a
homogeneous
function
is
a
homothetic
function.
This
specification
implies
that
the
marginal
technical
rate
of
substitution
between
all
pairs
of
inputs
will
be
constant
along
rays
from
the
origin.
In
terms
of
the
cost
function
corresponding
to
this
production
function,
the
returns
to
scale
(
as
measured
by
the
elasticity
of
cost
with
respect
to
output)
will
be
a
function
of
the
output
level.

?
To
make
this
judgment,
it
has
been
implicitly
assumed
that
the
site
demand
equations
include
an
additive,
classically
well­
behaved
error.

~
The
framework
implicitly
assumes
that
approximately
the
same
mix
of
recreation
activities
is
undertaken
by
users.
The
rationale
follows
from
the
assumption
that
users
have
comparable
household
production
functions
(
or
that
the
factors
leading
to
differences
in
household
production
technologies
can
be
specified).
The
assumption
on
the
mix
of
recreation
activities
is
equivalent
to
treating
Zr
as
an
aggregate
index
of
aii
of
the
recreation
undertaken
at
the
site.
.

7­
16
~
raveI
time
to
and
from
the
facility
~
ppor­
tunity
Cost
of
onsite
time.

of
one
trip
involves
a
full
cost
of
T"
di
+
rt
i
+
w.
t
The
first
two
Iv.
"
&
e
nf
these
costs
are
9iven
to
each
individual
on~
e
the
recreation
­.
flp~~
e~~~
ected.
However,
this
same
conclusion
does
not
follow
for
witv
.
site
spent
at
the
site~
tv
1
is
a
c
h
o
i
c
e
v
a
r
i
a
b
l
e
.
Thus,
if
onsite
cos\
s
~
be
time
are
included
in
a
travel
costi
demand
model,
amending
Equation
(
7.17)
to
the
restriction
implicit
in
the
previously
described
definition
of
the
of
~
trip,
the
estimation
of
the
model
must
reflect
simultaneity
in
the
reflect
price
of
vi
and
tv
"
In
past
studies,
this
issue
has
been
avoided
by
assure
­
c~
Oice
ing
that
O
n
s
i
t
e
timie
was
constant
for
all
trips.
*
Section
7.6
evaluates
the
.
.
imwrtance
of
this
Slmultanelty
for
the
recreation
sites
in
this
study.

The
measurement.
of
the
opportunity
cost
of
travel
time,
r,
and
of
onsite
tifne,
`
i
~
is
also
a
difficult
issue.
As
noted
in
the
previous
section,
there
has
been
considerable
controversy
over
the
appropriate
treatment
of
the
first
of
these
impliCit
prices.
Cesario
and
Knetsch
[
1970,
1976]
and
Cesario
[
1976]
have
argued
that
the
wage
rate
is
not
an
appropriate
index
of
the
first
of
these
COsts.
Rather,
based
on
individual
travel
choice
studies,
they
have
~
NPosed
that
the
opportunity
cost
of
traveling
time
is
a
fraction
of
the
wage
rate.
In
this
study's
sample,
the
wage
rate
is
estimated
based
on
a
wage
model
derived
from
the
1978
Current
population
Survey
that
permits
specific
wage
predictions
to
be
made
for
each
individual.
These
predictions
take
account
of
the
individual's
background,
including
education,
age,
occupation,
sex
~
racel
and
other
socioeconomic
characteristics.
As
a
result,
it
is
possible
to
separate
the
estimation
of
the
wage
rate
from
the
respondent's
reported
family
income.
The
next
section
provides
more
complete
details
on
the
wage
model
and
its
predictions
for
the
sample
of
recreation
ists.

Finally,
the
theoretical
model
does
not
offer
explicit
guidelines
as
to
how
a
site's
attributes
affect
the
derived
demand
functions
for
that
site's
services.
The
analysis
assumes
that
all
of
the
demand
parameters
can
be
affected
by
a
site's
features.
With
the
natural
log
of
visits
specificied
as
a
function
of
the
travel
and
time
costs
of
visiting
the
site,
income,
and
a
variety
of
other
determinants,
?
using
a
semi!
og
specification
gives
the
generalized
travel
cost
specification
in
its
simplest
form
as:

*
This
assumption
was
one
of
the
reasons
offered
by
Smith
and
Kopp
[
1980]
for
a
spatial
limit
to
travel
cost
models
estimated
from
aggregate
visit
rate
information
by
origin
zone.

?
Earlier
attempts
to
discriminate
between
the
popular
specifications
for
the
travel
cost
model
have
not
met
with
great
success.
Using
tests
for
nonnested
models,
Smith
[
1975b]
found
a
slight
preference
for
the
semi­
log
with
aggregate
visit
rate
data.
Ziemer,
Musser,
and
Hill
[
1980]
have
also
found
support
for
the
semilog
specification.
However,
neither
set
of
results
Co(
lld
be
regarded
as
definitive.

7­
17
InV..
=
ao(
ali,
a
2i
,
.
.
.
,
ski)
IJ
+
al(
alif
azit
.
.
.
/
s
k
i
)
`
ij
(
7.20)

+
~
z(
ali)
azi,
.
.
­.
1
ski)
`
j
.

The
double
subscript
for
Vii
and
p
ii
permits
the
identification
of
the
site
(
i)

a
n
d
t
h
e
i
n
d
i
v
i
d
u
a
l
recreatio~
ist
(
j).
`
Thus,
Vij
is
the
number
of
trips
to
the
ith
site
by
the
jth
individual,
P
ii
is
the
travel
and
time
costs
per
trip
for
the
jth
individual,
and
Y
i
is
that
in­
dividual's
income.
Significantly,
each
parameter
of
the
demand
e~
uation
is
specified
as
a
function
of
the
site
attributes.
Thus,
for
individuals
using
the
services
of
a
single
site,
the
demand
function's
parameters
are
assumed
to
be
constant.
Nonetheless,
this
model
has
the
ability
to
describe
how
the
demand
for
a
site's
services
changes
with
the
attributes
of
that
facility.
Thus,
separate
estimates
of
the
demands
for
individual
recreation
sites
together
with
measures
of
their
characteristics
provide,
in
principle,
the
information
needed
to
determine
the
demand
for
new
sites
or
for
existing
sites
that
experience
changes
in
their
available
characteristics
These
changes
might
include
improvements
in
water
quality,
capital
additions
increasing
access
points,
or
improvements
to
the
camping
facilities.
Thus,
this
analysis
demonstrates
that
the
observed
variation
in
the
estimated
parameters
of
travel
cost
site
demand
models
across
sites
may
be
the
result
of
differences
in
these
sites'
characteristics.
It
therefore
provides
the
basis
for
evaluating
the
implications
of
water
quality
for
recreation
behavior.
Indeed,
as
suggested
in
Section
7.7,
the
estimates
of
travel
cost
demand
models
together
with
the
attributes
explaining
the
variation
in
the
estimated
parameters
of
these
models
can
be
used
to
construct
the
demand
relationships
required
for
a
benefits
analysis
of
water
quality
changes.

It
is
also
important
to
recognize
that
the
structure
of
the
model
provides
sufficient
information
to
permit
efficient
estimation
of
the
role
of
site
attributes
for
the
parameters
of
site
demand.
To
illustrate
this
point,
consider
a
general
statement
of
the
site
demand
model:

Y
i
=
Xipi
+
&
i
,
(
7.21)

where
Y
i
=
N
x
1
vector
of
the
measures
of
The
quantity
demanded
for
the
ith
site's
services
by
each
of
N
individuals
x
i
=
N
x
K
matrix
of
demand
determinants
for
the
N
sampled
users
of
the
ith
site
Pi
=
K
x
1
parameter
vector
for
the
ith
site
&
i
=
N
x
1
vector
of
stochastic
errors
for
the
ith
site.

.

7­
18
the
theoretical
specification
given
in
Equation
(
7.20)
is
equiv­
~
perati0nalizin9
assuming
that
variations
in
the
vector
of
parameter
estimates,
Pi,
can
~
leflt
`
0
be
explained
by
the
attributes
of
each
recreation
site,
as
in
Equation
(
7.22):

Pi
=
eA
i
,
(
7.22)

where
e=
K
x
M
matrix
of
p
a
r
a
m
e
t
e
r
s
d
e
s
c
r
i
b
i
n
g
t
h
e
e
f
f
e
c
t
s
o
f
s
i
t
e
attributes
on
the
parameters
of
the
site
demand
equations
A
i
=
M
x
1
vector
of
the
M
characteristics
of
the
ith
site.

The
specification
for
the
determinants
of
site
demand
parameters
will
affect
the
form
of
the
efficient
estimator
of
this
two­
component
model.
Under
the
present
specification,
a
two­
step
estimation
scheme
can
be
considered.
The
first
would
invOive
the
estimation
of
each
site
demand
equation.
Assuming
there
are
S
sites,
the
process
yields
S
vectors
of
estimates
for
each
of
the
parameters
in
the
pi
v
e
c
t
o
r
.
Consider
the
ith
such
estimate.
If
&
i
is
classi­
.
tally
well
behaved,
the
ordinary
least­
squares
estimate,
P,,
of
~:
will
be
un
­
biased.
It
can
be
written
as:

;
i
=
(
xiT
x
i
)
­
'
xiT
Or,
substituting
for
Yi
from
Equation
(
7.21)

ii
=
pi
+
(
xiT
x
i
)
­
'

Because
Pi
is
not
observed,
it
is
necessary
1
I
Yi
.
(
7.23)

yields:

xiT
Ei
.
(
7.24)

to
consider
the
use
of
estimates
A
in
its
place.
The
ordinary
least­
squares
estimate,
pi,
is
one
such
possibility.

If
the
model
given
in
Equation
(
7.21.)
has
classically
well­
behaved
errors
and
A
nonstochastic
independent
variables
as
determinants,
pi
is
the
best
linear
unbiased
estimate
of
pi.
Substituting
for
f3i
in
Equation
(
7.22)
using
Equation
(
7.24)
provides
the
basis
for
a
second­
step
estimator:

pi
=
ii
­
(
xiT
x
i
)
­
'
xiT
&
i
=
e
A
i
.
(
7
.
2
5
)

Rearranging
terms
yields:

;
i
=
eAi
+
(
x
i
T
x
i
)
­
'
xiT
&
i
.
(
7
.
2
6
)

Equation
(
7.26)
clearly
suggests
that,
even
if
E(
si2)
=
U2
for
all
sites
(
i
.
e.
,

i=
1
to
s),
efficient
second­
stage
estimates
require
a
generalized
leastsquares
estimator.
T
h
a
t
i
s
,
the
model
given
in
Equation
(
7.
26)
must
be
.
estimated
taking
into
account
the
relative
precision
of
estimation
of
the
pi
7­
19
vector
across
sites.
This
will
be
given
in
each
case
by
the
corresponding
diagonal
elements
of
Equation
(
7.27):

E(~
i
­
~
i)(~
i
­
Pi)
T=
u2(
xiTxi)­
1
.
(
7.27)

T
h
e
(
XiT
Xi)­
l
will
not
be
identical
across
sites,
even
if
the
error
variances
are
constant
and
equal.

Unlike
many
instances,
the
nonspherical
errors
in
this
framework
provide
a
consistent
estimate
of
the
covariance
matrix
needed
for
generalized
leastsquares
estimation
of
the
models
in
terms
of
~
i.
These
estimates
are
contained
in
the
ordinary
least­
squares
estimates
of
the
respective
parameter
estimates'
covariance
matrices
(
i.
e.,
Equation
[
7.27]).

The
generalized
travel
cost
model
can
be
efficiently
estimated
with
a
two­
step
procedure.
Each
site
demand
model
is
estimated
with
ordinary
least­
squares
(
ignoring
for
the
moment
any
potential
simultaneity
introduced
by
the
onsite
time
costs
variable).
The
estimated
parameters
in
these
models,
together
with
their
estimated
variances,
provide
the
basis
for
the
second­
step,
generalized
least­
squares
estimates
of
the
role
of
site
attributes
as
determinants
of
the
individual
demand
parameters.
If
the
jth
member
of
ii
for
i
=
1,
2,
.
.
.
.
S,
if
the
vector
of
these
estimates
is
b
j
(
an
S
x
1
vector
of
the
ordinary
least­
squares
estimates
for
the
jth
parameter
in
the
original
pi
 
vector),
and
if
the
generalized
follows:

where
~..
2
is
the
corresponding
diagonal
element
for
~
i2
(
X
i
1
Xi)­
l
j
H
least­
squares
estimator
of
ej
(
the
sth
row
of
(
3)
is
given
as
~
T
j
=
(
AT;
­
lA)­
l
AT;­
lbj
,

tillz
o
0
3222
.

.

62
Ss
c
(
7.28)

A
=
SXM
matrix
of
Ai=
for
each
of
S
sites
.

This
estimator
is
somewhat
different
from
that
described
by
Saxonhous@
[
1977]
.
However,
the
overall
logic
is
completely
parallel.
The
two
generallzod
7­
20
.

 
 
.
 
.
.
.
 
.
.
.
 
 
least­
squares
estimators
differ
in
two
respects.
Saxonhouse
[
1977]
assumes
that
the
first
stage
models
will
be
jointly
estimated
with
a
Zellner
[
1
9
6
2
]
seeming!
y
unrelated
regressions
estimator.
This
approach
is
more
efficient
than
ordinary
least­
squares
estimates
of
the
individual
equations
when
there
is
contemporaneous
correlation
between
the
stochastic
errors
across
the
equations
and
when
the
independent
variables
in
all
models
together
are
not
highly
correlated.
*
As
originally
formulated,
the
Zellner
estimator
maintains
that
there
is
an
equal
number
of
observations
for
all
models.
While
Schmidt
[
1977]
has
developed
variations
on
the
estimator
that
relaxes
this
assumption,
there
is
no
reason
in
this
application
to
expect
contemporaneous
correlation
between
the
errors
of
the
site
demand
equations.
Each
will
be
based
on
independent
surveys
of
users
with
little
prospect
that
the
same
individuals
would
use
more
than
one
site.
I
n
the
absence
of
this
contemporaneous
correlation,
there
is
no
advantage
to
the
Ze!
lner
estimator.
It
is
identical
to
the
ordinary
least­
squares
estimates
for
each
equation.

The
second
distinction
arises
in
the
specification
of
the
covariance
structure
for
the
second
step
estimates.
Saxonhouse's
model
assumes
that
Equation
(
7.22)
includes
a
stochastic
error.
By
maintaining
that
these
errors
are
independent
of
the
site
demand
errors,
it
is
possible
to
develop
consistent
estimates
of
the
required
covariance
matrix
using
the
residuals
from
ordinary
least­
squares
estimates
of
the
second­
step
models.
Saxonhouse's
approach
can
be
viewed
as
a
generalized
random
coefficient
model
because
the
parameters
of
the
site
demand
models
are
treated
as
random
variables.
However,
the
observed
variation
in
these
parameters
(
across
sites)
arises
from
both
systematic
(
i.
e.
,
the
differences
in
each
site's
characteristics)
and
random
influences.
This
interpretation
has
been
avoided
here
in
preference
for
a
framework
that
treats
the
demand
parameters
as
constants
that
change
with
site
attributes.
Because
the
true
parameters
are
unobservable,
estimates
of
them
must
be
used
to
determine
the
role
of
these
attributes.
Thus,
random
influences
enter
the
framework
through
the
estimates
of
these
parameters
and
not
as
an
inherent
component
of
the
demand
model.

I
n
summary,
it
has
been
argued
that
it
is
possible
to
develop
a
theoretically
consistent
method
for
determining
the
effects
of
a
recreation
site's
characteristics
on
the
features
of
the
demand
for
that
site's
services.
Moreover,
the
framework
developed
here
does
not
require
information
on
al
I
recreation
sites
considered
by
each
potential
user.
This
is
an
important
distinction
between
the
approach
developed
here
and
the
Brown­
Mendelsohn
[
1980]
hedonic
travel
cost
model.
Equally
important,
it
is
possible,
using
a
straightforward,
two­
step
estimation
procedure,
to
provide
efficient
estimates
of
the
model.

It
should
be
acknowledged
that
the
approach
presented
here
is
not
new.
Freeman
[
1979a]
suggested
such
a
scheme
(
without
explicit
consideration
of
*
Of
course,
it
is
important
to
recognize
that
the
models
discussed
here
may
be
biased
as
a
result
of
the
assumption
that
all
sites'
services
can
be
transformed
into
common
units
using
conversion
functions
in
terms
of
their
respective
attributes.
This
framework
maintains
that,
after
adjustment
for
these
characteristics,
all
sites
are
perfect
substitutes
in
the
production
of
recreation
service
flows.

7­
21
the
estimation
problems)
as
one
of
a
number
of
ad
hoc
approaches
to
treat­
.
 
ing
water
quality
effects
in
modeling
the
demand
for
recreation
sites.
This
framework
has
extented
Freeman's
suggestion
by
demonstrating
that
it
is
not
ad
hoc.
Rather,
it
is
completely
consistent
with
a
household
production
frame­
 
 
work
of
recreation
participation
patterns
and
with
the
theory
of
adjusting
quantity
and
price
indexes
for
quality
changes
in
goods
and
services.

7.4
SOURCES
OF
DATA
The
1977
Nationwide
Outdoor
Recreation
Survey
was
conducted
by
the
Heritage
Conservation
and
Recreation
Service
as
part
of
the
Department
of
Interior's
mandate
to
periodically
develop
National
Recreation
Plans.
In
contrast
to
past
recreation
surveys,
which
only
included
a
general
population
component,
the
1977
survey
included
general
population
and
site­
specific
user
surveys.

The
Federal
Estate
Survey
component
of
the
survey,
the
primary
basis
of
this
study,
consists
of
interviews
with
recreationists
at
each
of
a
set
of
recreation
f
aci
I
ities.
All
federally
owned
areas
with
public
outdoor
recreation
were
considered
to
comprise
the
Federal
Estate,
and
sites
were
chosen
on
a
basis
of
specific
agency
control.
The
majority
of
interviews
were
conducted
in
areas
managed
by
the
National
Park
Service,
the
National
Forest
Service,
the
U
.
S.
Army
Corps
of
Engineers,
and
the
Fish
and
Wildlife
Service.
Each
agency
was
then
stratified
by
Federal
Planning
Regions,
and
areas
were
randomly
chosen
with
weight
given
to
annual
visitation
in
1975.

Interviewing
time
at
each
site
was
based
on
visitation,
which
also
determined
the
number
of
interviews.
The
final
Federal
Estate
Survey
contains
13,729
interviews
over
155
recreation
areas.
Information
collected
included
socioeconomic
characteristics,
current
outdoor
recreation
activities,
and
attitudes
toward
recreation
for
each
respondent.
Data
requirements
for
developing
travel
cost
models
that
describe
demand
for
individual
recreation
sites
are
met
by
the
Federal
Estate
Survey.

Given
that
the
scope
of
this
study
is
water­
based
recreation
and
that
the
analysis
requires
detailed
descriptions
of
the
activities
at
each
site,
only
U.
S.
Army
Corps
of
Engineer
sites
were
chosen
for
modeling.
These
46
sites
also
ensured
consistent
management
of
recreation
activities.
Three
were
eliminated
from
the
analysis
because
of
data
inconsistency
or
ambiguous
interview
site
locations.

A
number
of
the
sites
selected
for
analysis
from
the
Federal
Estate
Survey
had
observations
with
incomplete
information.
Rather
than
being
eliminated
from
the
sample,
these
observations
were
classified
according
to
whether
o
r
not
the
missing
information
affected
either
the
measurements
of
the
use
of
the
relevant
recreation
sites
or
the
travel
and
time
costs
of
that
use
versus
the
socioeconomic
characteristics
of
the
individuals
involved.
Observations
that
did
not
permit
evaluation
of
recreation
choices
(
i.
e.
,
those
missing
the
u
s
e
and
travel
information)
were
eliminated.
The
remaining
incomplete
observations
were
replaced
by
the
mean
values
of
the
relevant
variables
at
that
site
because
the
demand
models
were
estimated
at
the
site
level.
This
procedure
corresponds
to
the
zero­
order
method
for
treating
missing
observations.

7­
22
.
Section
7.6
discusses
the
results
of
using
regression
diagnostics
to
evaluate
the
sensitivity
of
each
site's
estimates
to
sample
composition.
In
addition
to
9au9in9
the
sensitivity
of
the
estimates
to
the
assumptions
of
our
models,
this
index
also
provided
a
means
to
evaluate
the
implications
of
the
procedures
used
for
missing
observations.

Several
variables
in
the
Federal
Estate
Survey
were
reported
by
discrete
internal
s.'
Answers
to
questions
concerning
time
spent
at
site,
number
of
visitS
to
s
i
t
e
,
travel
hours
to
site,
and
annual
income
were
treated
as
continuous
variables.
In
all
cases
the
interval's
midpoint
was
used.
Open­
ended
intervals
were
converted
using
the
previous
interval,
with
the
difference
between
the
previous
interval's
midpoint
and
minimum
value
added
to
the
open­
ended
minimum
value.

One
component
of
the
model
described
in
Section
7.3
is
the
travel
cost
of
a
trip,
which
is
defined
as
the
number
of
miles
traveled
multiplied
by
a
per
mile
COSt.
An
independent
estimate
of
travel
cost
was
developed
by
measuring
each
respondent's
actual
road
distance
traveled
to
a
site
based
on
his
reported
zip
code.
All
distances
were
calculated
with
the
Standard
Highway
Mileage
Guide
[
Rand
McNally,
1978]
,
which
lists
road
miles
between
1,100
cities.
National
interstate
highways
and
primary
roads
were
used
in
all
calculations
Other
routes
were
used
only
for
the
distance
to
the
nearest
primary
road.
In
cases
where
cities
have
multiple
zip
codes,
the
center
of
the
city
was
used
as
the
origin.

The
second
part
of
the
travel
cost
calculation
requires
a
per
mile
cost
of
a
trip.
The
marginal
cost
of
operating
an
automobile
in
1976
is
estimated
to
be
approximatley
$
0.08
per
mile.
This
estimate
is
based
on
costs
of
repairs
and
maintenance,
tires,
gasoline,
and
oil
as
reported
by
the
U
.
S.
Census
Bureau
,
in
the
U.
S.
Statistical
Abstract
[
1978]
.
Mileage
costs
for
operating
an
average
automobile
were
then
calculated
by
using
the
round
trip
miles
to
the
site
multiplied
by
$
0.08.
This
assumes
that
the
respondent
drove
directly
to
the
site
using
the
routes
in
the
Standard
Highway
Mileage
Guide.
Unfortunately,
information
was
not
available
on
the
primary
purpose
of
the
respondents'
trip
or
further
driving
plans.

The
Federal
Estate
Survey
includes
annual
household
income
of
respondents
but
does
not
indicate
any
hourly
wage
rate.
Because
the
use
of
reported
income
in
calculating
opportunity
cost
of
time
precludes
determination
of
income's
role
in
the
site
demand
models,
an
independent
estimate
of
each
individual's
wage
rate
is
important
to
a
complete
specification
of
the
model
.

A
hedonic
wage
model
estimated
from
the
1978
Current
Population
Survey
(
CPS)
was
used
to
derive
these
estimates.
This
model
specifies
the
market
clearing
wage
rates
to
be
a
function
of
individual­,
job­,
and
location­
specific
characteristics.
`
The
specific
model
was
developed
by
Smith
[
forthcoming,
1983].
By
substituting
each
individual's
characteristics"
(
including
locationspecific
and
occupation­
specific
variables),
predicted
wage
rates
were
derived.
Equation
(
7.29)
provides
a
general
statement
of
the
procedure,
with
X
ij
t
h
e
determinants
of
the
wage
rate:

7­
23
A
N.
W
i
=
exp
(
Z
BjXij)
,
j=
l
where
.
W
i
=
the
predicted
wage
rate
for
the
ith
individual
A
B.
=
the
estimated
coefficient
of
the
jth
variable
J
(
7.29)

x
.
.
=
the
ith
individual's
value
of
the
jth
variable
1]

N
=
the
number
of
explanatory
variables.

Explanatory
variables
usually
include
age,
sex,
education,
occupation,
and
various
other
job­
and
location­
specific
characteristics.

The
estimates
made
in
this
study
should
be
regarded
as
proxy
measures
for
actual
wage
rates.
Since
the
wage
model
is
a
semilog,
the
predictions
can
be
expected
to
understate
the
estimated
conditional
expectation
for
the
wage
rate.
While
Goldberger's
[
1968]
proposed
unbiased
estimator
for
this
conditional
expectation
would
be
superior
for
large
degrees
of
freedom
(
the
CPS
sample
contained
9,077
for
males
and
7,067
for
females)
and
for
a
small
error
variance
of
the
estimated
model,
the
bias
in
this
study's
estimates
will
be
small.
A
10­
percent
discrepancy
would
be
a
generous
outer
bound
on
the
magnitude
of
the
percentage
difference
between
the
direct
predictions
of
these
wage
rates
and
the
estimates
based
on
Gold
berger's
method.
Indeed,
in
most
large
sample
applications
(
see
the
examples
in
Gold
berger
[
1968]
and
Giles
[
1982]
),
the
actual
differences
are
under
5
percent.
Thus,
despite
this
limitation,
these
estimates
provide
a
better
set
of
proxy
measures
for
wage
rates
than
the
available
alternatives
since
they
take
explicit
account
of
individual
and
job
characteristics.
In
specifying
and
estimating
the
wage
model,
consideration
was
also
given
to
measures
of
job
risks,
air
pollution,
climate,
crime,
access
to
cultural
and
sporting
activities,
and
local
labor
market
conditions

The
nominal
wage
model
includes
a
cost­
of­
living
variable
as
one
of
the
determinants
of
wages.
Smith
used
the
Bureau
of
Labor
Statistics
budgetcost
of­
living
index
for
this
variable.
In
the
Standard
Metropolitan
Statistical
Areas
(
SMSAS)
where
the
index
was
not
known,
information
available
for
27
SMSAS
was
used
to
model
the
determinants
of
variations
in
the
cost
of
living.
As
shown
in
E
q
u
a
t
i
o
n
(
7
.
3
0
)
,
t
h
e
i
n
d
e
x
,
C
j,
was
related
to
Populatio
n
density,
D
j;
the
size
of
the
SMSA
population
in
1975
in
thousands,
POP.;
and
J
the
percent
of
the
population
under
125
percent
of
the
poverty
standard,
POOR..
The
t­
ratios
for
the
hypothesis
of
no
association
are
shown
in
paren
­
J
theses:

c
.
=
111.81
+
0.005
D.
­
0.001
POP.
­
1.30
POOR.
J
(
37.73)
(
7.38)
J
(
­
2
.
4
0
)
J
(
­
4
.
3
6
)
J
R
2
=
0.787
.­
F(
3,
23)
=
28.34
.

7­
24
(
7.30)

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
­­­­­­­
,,.
­_
u
~­­­
.
.
 
_
.
.
The
Federal
Estate
Survey
does
not
directly
identify
respondents'
SMSA.
Thus,
a
cost
of
living
variable
was
generated
at
the
State
level
to
minimize
computation
cost.
This
index
was
calculated
as
the
average
of
the
SMSAS
within
each
State,
avoiding
the
need
to
match
each
respondent
to
an
SMSA.

The
estimated
1977
nominal
wages
for
the
recreationists
at
each
site
were
developed
using
the
equations
in
Table
7­
1.
The
characteristics
necessary
for
the
model
were
generally
available
in
the
Federal
Estate
Survey,
and
classifications
between
the
model
and
the
survey
were
compatible.
Problems
do
arise,
however,
for
respondents
who
were
not
labor
force
participants
at
the
time
of
the
survey.
For
example,
students
and
housewives
could
not
be
considered
in
the
sample
used
to
estimate
the
hedonic
wage
model.
In
these
cases,
the
wages
were
treated
as
an
opportunity
cost
estimated
to
be
the
mean
value
by
sex
of
the
predicted
wage
rates
in
the
recreation
survey.
Table
7­
2
provides
a
summary
of
predicted
hourly
wage
rates
by
income
and
occupation
of
the
respondents.
The
predicted
wage
rate
is
used
to
calculate
the
opportunity
cost
of
both
onsite
time
and
travel
time.
For
at
least
two
reasons,
there
are
substantial
differences
in
these
estimates
for
the
upper
income
members
of
the
sample.
The
first
stems
from
the
coding
of
the
wage
measure
in
the
Current
Population
Survey.
Specifically,
the
reporting
format
limits
the
reported
usual
weekly
earnings
(
the
basis
for
the
hourly
wage
rate­­
usual
weekly
earnings
divided
by
usual
hours
worked)
to
$
999.
Thus,
there
is
censoring
in
wages
for
individuals
above
approximately
$
52,000
per
year.
The
second
reason
is
that
family
income
can
reflect
the
effects
of
nonwage
income
and
the
impact
of
dual
earner
households.
Unfortunately,
the
extent
of
these
influences
cannot
be
sufficiently
determined
to
improve
wage
rate
estimates
for
individuals
in
these
higher
income
households.

The
U
.
S.
Army
Corps
of
Engineers
maintains
the
Recreation
Resource
Management
System
for
evaluation
and
planning.
Data
from
this
system
are
compatible
with
the
sites
chosen
for
the
Federal
Estate
Survey
and
have
been
available
since
1978.
Information
is
collected
annually
on
each
water
resource
project
with
5,000
or
more
recreation
days
of
use.
For
1978,
this
information
included
financial
statistics,
facilities
available,
natural
attributes,
recreation
participation,
and
number
of
employees.

The
Recreation
Resource
Management
System
is
used
to
define
attributes
of
the
43
Federal
Estate
Survey
sites.
Attributes
of
an
area
considered
include
land
area,
shore
miles,
pool
elevation,
the
number
of
multipurpose
recreation
areas,
and
facilities
provided.
Table
7­
3
provides
descriptive
statistics
for
both
the
characteristics
of
the
sites
and
of
a
selected
set
of
variables
for
the
survey
respondents
at
these
sites.

The
National
Water
Data
Exchange
(
NAWDEX)
is
a
membership
of
wateroriented
organizations
and
is
a
major
source
of
water
quality
information.
The
NAWDEX
system
is
under
the
direction
of
the
U.
S.
Geological
Survey,
and
its
primary
function
is
to
exchange
data
from
various
organizations.
Major
sources
of
information
are
usually
State
Survey,
t
h
e
U
.
S
.
Army
Corps
of
Engineers,
agencies,
the
U.
S.
Geological
and
the
U
.
S,
Environmental
7­
25
.
.
Table
7­
1.
Hedonic
Wage
Models
Male
Female
t­
statistics
t­
statistics
Variable
a
Coefficient
(
of
no
association)
Coefficient
(
Of
no
association)

Intercept
Education
Education
squared
Experience
Experience
squared
Race
Veteran
Unemployment
Professional
Managerial
Sales
Clerical
Craftsman
Operative
Transport
equipment
Nonfarm
labor
Service
worker
Injury
rate
Cancer
TSP
Household
~
eaci
Union
member
OJT
x
Experience
Crime
rate
Percent
sunshine
Dual
job
holder
Know
x
Cancer
Log
(
cost
of
living
index)
0.631
8.71
0.030
4.01
0.001
3.4.5
0.031
25.83
­
0.001
­
22.35
0.113
8.75
0.035
3.50
­
0.012
­
3.51
0.086
2.76
0.142
4.48
­
0.0003
­
0.01
­
0.099
­
3.01
0.015
0.48
­
0.149
­
4.47
­
0.118
­
3.35
­
0.131
­
3.87
­
0.251
­
7.70
0.011
10.40
0.299
2.93
0.0007
2.31
0.229
16.75
0.178
17.52
­
0.002
­
1.64
0.000005
1.89
­
0.002
­
2.31
­
0.042
­
1.75
3.77
4.58
0.559
7.22
R2
='
0.47
degrees
of
freedom
=
9,077
F
ratio
=
292.92
0.179
2.03
0.028
2.61
0.001
2.15
0.018
15.91
­
0.0002
­
11,83
­
0.024
­
1.73
­
­
­
­
­
­
0.002
0.57
0.563
19.17
0.521
16.15
0.199
6.30
0.390
15.38
0.445
8.68
0.235
8.09
0.366
5.34
0.199
3.97
0.166
6.26
0.012
7.67
0.105
0.86
0.0003
0.97
0.069
6.01
0.191
13.81
­
0.001
­
0,54
­
0.000008
2.39
0.0001
0.12
­
0.025
­
0.81
5.727
4.24
0.606
6.56
R
2
=
0.33
degrees
of
freedom
=
7,067
F
ratio
=
135.52
SOURCE:
Smith
[
1983].

a
The
variable
definitions
are
as
follows:

(
1)

(
2)

(
3
)

(
4
)

(
5
)

(
6
)

(
7
)

(
8
)

T
h
e
Education­­
measured
as
the
years
corresponding
to
the
highest
grade
of
schml
attended
(
this
varlabl{
is
entered
in
linear
and
quadratic
terms).

Experience
­­
measured
using
the
conventional
proxy
of
age
minus
years
of
education
minus
SIX
(
thi!
variable
is
entered
in
linear
and
quadratic
terms).

Socioeconomic
qualitative
variables­­
dummy
variables
for
race
(
white
=
1),
sex
(
male
=
1),
veterar
status
(
vetaran
=
1
and
relevant
only
for
males),
member
of
a
union
(
yes
=
1),
head
of
the
housenolc
(
yes
=
1),
and
dual
job
holder
(
yes
=
1).

Occupational
qualitative
variables­­
dummy
variables
to
define
the
respondents
occupation
as:
profcs.
sional,
managerial,
sales,
clerical,
craftsman,
operative,
transport
equipment
operator,
nonfarm
labor,
or
service
worker.

Cancer­­
index
of
exposure
to
carcinogens.

TSP­­
average
suspended
particulate
in
7978.

OJT­­
on­
the­
job
training
program
available.

Know­­
relative
number
of
workers
within
an
industry
covered
by
collective
bargaining
w!
th
haalth
~
safety
provisions.

omitted
occupational
category
was
defined
to
correspond
to
a
composite
of
those
occupations
that
mfl
lead
the
estimated
hourly
wage
to
understate
actual
earnings.
The
o"
mitted
occupations
were
farm
IX
and
private
household
workers.

A
measure
of
price
uncertainty
was
constructed
to
provide
soma
basis
for
adjusting
the
experience
~
to
reflect
the
different
levels
of
provision
of
on­
the­
job
training
(
OJT)
across
firms.
To
evaluate
t~
`~
tance
of
these
effects,
price
uncertainty
was
measured
as
the
unexplained
variation
(
i.
e.,
1­
R2)
fOf
1'­
trand
models
fit
to
monthly
wholesale
price
indexes
for
each
of
14
product
categories
for
each
Y­
~
the
period
1976
through
1978.
After
evaluating
each
year's
index,
1
9
7
7
w
a
s
selacted
for
th,
s
­
V*
The
indexes
were
assigned
to
individuals
according
to
their
industry
of
employment
in
an
attemot
m
4
products
as
closely
as
possible.
The
variable
was
entered
as
an
interaction
term
with
experience.

7­
26
`/

 
Table
7­
2.
Summary
of
Predicted
Hourly
Wage
Rates
(
1977
$)

Total
sample
Male
Female
 
~
Veral
I
mean
Number
of
observations
Mean
b
y
annUal
ho{
Under
5,999
6,000t0
9,999
10,000
to
14,999
15,000
to
24,999
25,000
to
49,999
~
ojoOO
or
more
a
~
sehold
income
Mean
by
Occupation
of
respondent
professional,
technical,
and
kindred
workers
Farmers
Managers,
officials,
and
proprietors
Clerical
and
kindred
workers
Sales
workers
Craftsmen,
foremen,
and
kindred
workers
Operatives
and
kindred
workers
Service
workers
Laborers,
except
farm
and
mine
Retired
widows
Students
Unemployed
Housewives
Other
No
occupation
given
5.44
3,460
5.08
4.92
5.32
5.72
5.98
5.73
7.05
5.15
7.17
4.34
5.18
5.89
4.97
4.11
4.44
5.92
5.30
5.46
4.37
5.71
5.49
6.27
1,971
5.79
5.49
6.01
6.70
7.17
6.53
7.89
5
.
7
1
7.74
5.94
6.24
6.05
5.15
4.71
4.74
6.27
6.27
6.27
6.27
6.27
6.27
4.34
1,489
4.06
4.10
4.38
4.39
4.65
4.65
5.65
2.75
4.94
4.10
3.29
4,31
3.56
3.18
3.11
4.34
4.34
4.34
4.34
4.34
4.34
aTotal
number
of
observations
is
3,282.

bTotal
number
of
observations
is
3,460.

Protection
Agency
(
EPA).
All
water
quality
data
used
in
the
analysis
were
retrieved
from
NAWDEX
in
a
series
of
steps.
Collection
of
useful
water
quality
data
was
completed
by
identifying
potential
monitoring
stations
and
by
then
obtaining
actual
data.
Potential
monitoring
stations
were
identified
by
defining
the
recreation
area
in
terms
of
latitude
and
longitude.
A
general
retrieval
was
then
obtained
that
listed
station
name,
location,
parameter
Collected,
years
of
data
collection,
and
agency
responsible
for
the
data
collection

7­
27
.
.
.
.
.
.
.
.
.
.
.
..
 
_
 
­
.
 
.
.
.­­
 
 
.
..
..
___________
­­­
­_
\
I
able
/­
3.
The
~
naracteristlcs
of
the
Sites
and
the
Survey
Respondents
Selected
from
the
Federal
Estate
Survey
Characteristics
of
survey
respondents
Site
characteristics
Predicted
Number
wage
rate
Household
Income
Visits
M
i
l
e
s
a
o
f
Property
Recreation
Shore
A
rtsa
(
T+
M)
cost
Project
name
code
days
miles
obser­
b
acres
i
o
x
o
i
a
k
o
k
o
vations
Allegheny
River
System,
PA
Arkabutla
Lake,
MS
Lock
&
Dam
No.
2
(
Arkansas
River
Navigation
System),
AR
Beaver
Lake,
AR
Belton
Lake,
TX
Benbrook
Lake,
TX
Berlin
Reservoir,
OH
Blakely
Mt.
Dam,
Lake
Ouachita,
AR
Canton
Lake,
OK
Clearwater
Lake,
MO
Cordell
Hull
Dam
&
Reservoir,
TX
DeGray
Lake,
AR
Dewey
Lake,
KY
Fort
Randall,
Lake
Francis
Case,
SD
Grapevine
Lake,
TX
Greers
Ferry
Lake,
AR
Grenada
Lake,
MS
4
Herds
Creek
Lake,
TX
&
Isabella
Lake,
CA
co
Lake
Okeechobee
and
Waterway,
FL
Lake
Washington
Ship
Canal,
WA
Leech
Lake,
MN
Melvern
Lake,
KS
Millwood
Lake,
AR
Mississippi
River
Pool
No.
3,
MN
Mississippi
River
Pool
No.
6,
MN
Navarro
Mills
Lake,
TX
New
Hogan
Lake,
CA
New
Savannah
Bluff
Lock
&
Dam,
GA
Nofiork
Lake,
AR
Ozark
Lake,
AR
Perry
Lake,
KS
Philpott
Lake,
VA
Pine
River,
MN
Pokegama
Lake,
MN
Pomona
Lake,
KS
Proctor
Lake,
TX
Rathbun
Reservoir,
TX
Sam
Rayburn
Dam
&
Raaarvolr,
TX
Sardla
Lake,
MS
wuo
LsIu.
TX
WhIUW~
L4he,
1
X
*
ab+
wghemv
nlvor
Lake
 
l
 
A
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
31s
316
317
318
319
320
321
322
323
324
32S
327
328
329
330
331
332
333
334
335
336
337
338
339
340
343
344
345
.

2,011,700
343,700
4,882,600
2,507,000
1,978,000
1,179,000
2,104,300
3,416,500
88s3,000
2,167,900
1,659,700
1,116,800
4,756,000
5,139,100
4,407,000
2,553,800
359,500
1,489,200
2,894,584
712,900
950,600
2,034,600
2,042,300
1,323,700
645,500
1,111,500
335,200
207,600
3,066,500
1,102,000
3,388,000
1,454,800
1,615,100
948,300
1,460,400
975,200
2,332,200
2,728,700
2,488,900
3,371,600
1,976,400
1.122,600
134
96
449
136
37
70
690
45
27
381
207
52
540
60
276
148
11
38
402
80
316
101
65
37
55
38
44
3E
173
160
100
119
53
52
27
156
560
110
60
170
38
52,549
32,415
40,463
30,789
11,295
7,9W
82,373
19,797
18,715
32,822
31,800
13,602
133,047
17,828
45,548
86,826
3,027
15,977
451,000
169
162,100
24,543
142,100
20,350
11,292
14,286
6,162
2,030
54,193
39,251
41,769
9,600
22,177
66,
S42
12,301
15,956
36,072
176,869
98,590
21,342
53,230
4,035
5.45
5.23
5.24
5.59
5.52
5.00
5.44
5.24
5.09
5.43
5.43
5.17
5.83
5.43
5.20
5.15
5.13
5.26
5.64
5.38
6.26
5.90
5.69
5.49
6.36
5.79
5.16
5.57
5.28
5.65
5.02
5.52
5.33
5.95
5.70
5.42
5.49
5.74
5.32
5.41
5.46
5
25
5.56
1.65
1.45
1.03
1.70
1,51
1.21
1.24
1.53
1.54
1.38
1.58
1.58
2.10
1.69
1.58
1.45
1.56
1.42
1.48
1.20
2.07
1.40
1.65
1.87
2.23
1.42
1.41
1.28
1.13
1.61
1.22
1.48
1.55
1.80
1.46
1.36
1.63
1.56
1.35
1,31
1.25
1.29
1.59
15,667
13,184
10,409
18,150
17,279
19,135
16,459
17,144
17,392
17,943
15,491
19,235
18,021
20,696
19,309
15,890
9,199
16,263
15,938
13,849
16,686
18,886
18,087
18,630
29,571
19,589
13,739
18,954
12,609
17,667
12,654
16,565
14,268
20,097
16,816
17,265
17,510
20,543
19,515
13,141
16,396
18,688
16,682
8,625
8,974
3,991
9,946
11,913
10,065
10,161
9,524
10,553
8,456
9,215
10,612
9,559
11,705
10,992
8,562
4,833
9,699
11,445
9,541
5,815
10,986
9,015
1,319
10,895
10,693
4,652
11,270
9,414
8,889
7,568
6,925
6,668
9,370
9,476
7,330
11,167
7,473
11,331
7,223
12,454
11,651
11,051
2.6
5.4
6.8
3.5
6.0
2.3
5.2
4.3
4.6
4.0
5.7
4.8
2.4
3.3
6.3
4.7
6.4
4.4
3.3
4.1
3.3
2.5
4.3
5.6
3.0
4.8
4.6
4.0
5.8
3.2
4.9
4.7
5.8
2.1
3.3
5.4
5.4
4.3
4.1
6.5
6.9
5.0
5
4
2.5
2.7
2.0
3.0
2.8
1.2
2.9
2.8
3.2
2.7
2.9
2.7
2.0
3.1
2.6
3.0
2.6
3.0
2.5
3.0
3.0
1.8
3.0
3.0
2.4
3.0
2.8
3.1
2.7
2.5
3.0
2.7
2.6
;:;
2.8
2.9
2.9
2.7
2.3
2.2
2.8
2.9
45.19
20.04
3.04
94.55
33.18
30.23
21.15
45.39
32.30
50.51
29.65
42.04
90.75
100.29
38.45
54.16
24.57
39.46
55.59
24.91
98.63
104.08
31.48
37.62
99.20
52.23
27.68
34.10
18.65
94.89
58.71
28.79
26.09
69.80
100.63
25.38
46.08
41.78
40.23
36.08
33.02
35.40
24.67
28.30
27.94
13.01
88.64
52.35
58.93
26.63
49.31
22.97
42.24
34.70
43.42
122.44
93.59
64.32
70.00
32.90
48.25
45.54
11.03
130.14
84.35
29.39
55.21
79.14
55.19
30.29
14.55
23.78
59.65
98.54
24.02
46.00
50.54
122.30
23.33
40.96
29.18
31.90
42.17
45.10
38.03
9.48
106
45
55
266
67
73
40
121
95
140
60
115
243
260
92
154
65
108
127
76
338
268
84
90
196
141
61
72
37
268
199
79
47
178
376
65
109
96
85
123
99
96
47
57
90
33
296
142
223
130
139
99
192
87
164
519
295
217
306
165
170
100
258
605
313
137
176
288
240
70
29
77
75
433
109
100
188
590
115
103
41
74
234
263
195
58
69
61
41
226
53
46
96
91
74
74
104
49
46
:;
217
75
54
48
30
37
48
45
53
49
70
42
41
39
42
52
28
38
75
68
31
52
31
67
205
61
201
31
l
(
.
ha­­~
v
d$
stmco
10
Iha
SI1*
NOTES:
x
IS
the
arithmetic
mean.

`
Me.
or
obwrv.
tmn.
 
l
 
.
­
has­
d
o
n
tha
rln.
1
mod.
ls
.
stimat.
d
for
Site.
a
1s
th
­
Standard
d­
viatb"

.
 
.
Z­­
.
.
.
..~
One
major
problem
in
the
data
collection
process
is
the
identification
of
a
p
P
r
o
P
r
i
a
t
e
monitoring
sites.
Ideally,
monitoring
stations
should
be
located
in
the
area
where
recreation
occurs.
Monitoring
sites
could
only
be
identified
by
obscure
station
names.
Furthermore,
information
is
not
available
according
to
area
names
used
by
survey
respondents.
Proximity
of
a
water
quality
monitor
to
actual
recreation
could
not
be
determined.

Monitoring
sites
that
could
be
identified
as
relevant
were
then
chosen,
and
the
actual
water
quality
data
were
obtained
through
NAWDEX.
Several
problems
are
inherent
in
this
type
of
data
collection.
A
brief
discussion
of
the
data
collection
process
and
some
problems
encountered
follow.
The
reader
is
referred
to
Appendix
E
for
a
more
detailed
discussion
of
water
quality.

Water
quality
parameters
were
selected
on
a
basis
of
previous
use
and
availability
among
sites.
The
parameters
CO
I
Iected
are
temperature,
PH
,
dissolved
oxygen,
biological
oxygen
demand,
turbidity,
nitrates,
phosphates,
fecal
coliform,
dissolved
solids,
flow,
and
Secchi­
disk
transparency.
Of
the
43
sites,
16
had
no
data
due
to
a
lack
of
known
monitoring
sites.

Actual
water
quality
data
were
collected
for
27
sites
for
the
years
1972
to
1981.
Most
of
these
sites
were
missing
information
for
the
year
the
survey
was
completed.
As
a
result,
calculations
were
carried
out
using
1972
to
1981
data.
Monthly
means
for
each
site
were
calculated
for
June
through
September.
An
overall
mean
was
also
calculated
using
the
four
monthly
means.
In
cases
where
sites
were
completely
missing
a
parameter,
the
mean
for
all
sites
was
used.

individual
parameters
and
indexes
are
used
in
the
analysis,
including
both
monthly
values
and
a
summer
average.
Index
methods
include
the
National
Sanitation
Foundation
and
the
Resources
for
the
Future
measures.
Linear
combinations
of
parameters
were
also
tested,
although
the
degree
of
correlation
between
parameters
was
regarded
with
caution.

The
treatment
of
missing
values
for
these
variables
led
to
a
lack
of
variation
between
sites.
This
is
caused
by
two
factors.
First,
the
averaging
of
several
years
distorts
the
actual
water
quality
for
a
particular
year.
Consideration
is
not
given
to
improvements
or
deterioration
of
water
quality.
Secondly,
replacing
missing
observations
with
the
means
smooths
out
the
variation
between
sites.
Any
predictions
of
water
quality
benefits
with
t
h
e
travel
cost
model
will
become
more
reliable
as
missing
observations
are
replaced
with
actual
data.

The
choice
of
parameters
to
be
measured
at
a
monitoring
site
varies
according
to
a
water
body's
local
characteristics
and
the
agency
collecting
the
sample.
This
inconsistency
in
data
collection
may
cause
problems
when
the
43
Army
Corps
of
Engineers'
areas
are
compared.
For
example,
if
suspended
solids
are
not
considered
a
problem
in
an
area,
they
are
not
likely
to
be
measured.
Consequently,
several
parameters
were
not
available
in
all
areas
or
during
the
appropriate
time.

I
n
summary,
three
generally
compatible
data
sources
were
used.
Data
obtained
from
each
source
are
consistent
y
defined
across
sites.

7­
29
7.5
EMPIRICAL
RESULTS
FOR
SITE­
SPECIFIC
TRAVEL
COST
MODELS
The
theoretical
model
of
the
consumer's
recreation
decisions
identified
three
aspects
of
the
process
that
may
influence
the
use
of
the
travel
cost
model
for
an
analysis
of
the
benefits
(
or
costs)
of
a
change
in
the
attributes
of
a
recreation
site.
Two
of
these
aspects
arise
in
defining
the
relevant
measure
of
site
usage
and
the
associated
cost
to
the
individual
for
a
"
unit"
of
the
site's
services
(
assuming
an
ideal
quantity
index
could
be
derived).
In
the
formal
model
of
household
choice,
the
individual
was
able
to
produce
additional
units
of
the
recreation
service
flow
with
more
trips
of
a
given
length
or
by
increasing
the
time
spent
onsite
during
a
fixed
number
of
trips.
The
household
production
framework
did
not
specify
these
choices
as
perfect
substitutes,
but
it
did
admit
the
possibility
of
substitution.
This
type
of
input
substitution
is
plausible
because
the
time
horizon
for
production
has
been
interpreted
to
be
the
recreation
season.
This
specification
of
the
problem
implies
that
the
number
of
visits
to
a
given
site
and
the
times
spent
onsite
per
visit
will
be
jointly
determined
variables.
Indeed,
the
demand
model
for
visits
(
i
.
e.
,
Equation
[
7.7]
)
was
expressed
as
a
reduced
form
equation.
Of
course,
the
specific
analytical
model
simplified
the
issues
involved
by
assuming
the
time
spent
onsite
was
the
same
for
all
the
visits
in
a
given
season.
Actual
behavior
is
more
complex,
with
the
prospects
for
different
amounts
of
time
spent
onsite
for
every
visit.
There
are
several
aspects
of
this
problem
described
below
in
greater
detail.
The
discussion
portrays
the
treatment
of
each
issue
in
this
analysis
and
how
this
treatment
compares
with
earlier
literature.

The
second
aspect
of
modeling
an
individual's
recreation
choices
arises
in
the
definition
of
the
cost
of
a
visit
to
a
given
recreation
site.
The
analytical
model
indicated
that
this
cost
would
be
composed
of
the
costs
of
transportation
to
the
site
(
i.
e.
,
the
product
of
roundtrip
mileage
and
a
vehicle
operating
cost
per
mile)
and
the
opportunity
costs
associated
with
the
time
spent
traveling
to
the
facility.
As
noted
earlier,
the
appropriate
definition
of
these
opportunity
costs
has
been
addressed
in
several
papers
in
the
past
literature.
The
model
identifies
the
cost
as
r
and
does
not
attempt
to
relate
it
to
the
individual's
wage
rate.
Of
course,
in
practice
r
is
unknown
and
requires
estimation.
Since
the
treatment
of
this
variable
has
important
implications
for
the
estimated
costs
of
a
trip,
the
issues
involved
in
this
study's
modeling
choices
are
detailed
below.

Finally,
the
third
aspect
of
the
representation
of
recreation
decisions
stems
from
this
chapter's
overall
objective,
which
is
to
evaluate
the
influence
of
site
characteristics
on
the
demand
for
the
services
of
a
recreation
facility.
As
developed
in
Section
7.3,
some
analytical
restrictions
on
the
role
of
site
attributes
for
the
production
of
recreation
service
flows,
together
with
a
diversity
of
these
features
across
sites,
provide
sufficient
information
t
o
estimate
the
relationship
between
each
site's
demand
model
and
its
attributes.
To
estimate
this
relationship,
however,
requires
the
adoption
of
a
common
demand
specification
for
all
the
individual
site
demand
equations.
While
the
sample
sites
provide
the
ability
to
engage
in
an
approximately
comparable
set
of
recreation
activities,
this
is
not
a
sufficient
reason,
in
itself,
for
expecting
the
site
demand
models
to
be
comparable.
Thus,
before
turning
to
the
generalized
least­
squares
models
for
explaining
the
variation
in
an
individual
.

7­
30
I
site's
estimated
demand
parameters,
the
implications
of
using
a
common
specification
must
also
be
considered.
To
adequately
treat
these
three
issues,
a
fairly
detailed
set
of
statistical
analyses
of
site
demand
models
was
undertaken.

The
explanation
of
these
results
will
be
developed
in
this
and
the
next
two
sections
of
this
chapter.
This
exposition
begins
with
a
more
detailed
discussion
of
the
conceptual
dimensions
of
each
of
these
issues
in
the
first
three
subsections
of
this
section.
The
ordinary
least­
squares
estimates
for
the
general
model
applied
to
all
43
sites
follow
that
discussion.
The
remainder
of
this
section
discusses
the
implications
of
using
conventional
pretesting
criteria
for
selecting
individual
specifications
for
each
site
demand,
as
well
as
the
influence
of
different
approaches
for
treating
the
opportunity
cost
of
travel
time
to
each
site.
Section
7.6
discusses
the
results
of
the
analysis
of
onsite
time
and
visits
within
a
simultaneous
equation
model
and
the
several
specific
statistical
issues
that
arise
for
travel
cost
models
because
of
the
nature
of
the
available
measures
of
site
usage.
The
final
component
of
the
model
is
developed
in
Section
7.7,
where
the
results
of
the
generalized
leastsquares
model
for
the
determinants
of
the
features
of
recreational
site
demand
equations
are
presented.

7.5.1
The
Treatment
of
Onsite
Time
Ideally,
the
measurement
of
site
demand
models
would
involve
both
the
number
of
trips
to
a
particular
site
and
the
time
spent
onsite
for
each
trip.
Unfortunately,
in
practice
this
information
is
rarely
available.
*
The
source
of
data
for
this
analysis
(
the
Federal
Estate
Survey)
includes
information
on
the
amount
of
time
spent
at
the
site
during
the
trip
in
which
the
respondent
was
interviewed
and
not
the
corresponding
information
for
all
trips
taken
during
the
season.
Thus,
any
attempt
to
deal
with
the
relationship
between
onsite
time
and
visits
will
require
further
assumptions.

There
have
generally
been
two
treatments
of
onsite
time
in
the
recreation
demand
literature.
The
first
of
these
corresponds
to
the
most
common
practice
in
the
Iiterature­­
onsite
time
is
assumed
to
be
constant
across
trips
and
across
individuals.
In
this
case,
the
number
of
visits
is
a
consistent
in=
of
the
use
of
a
site's
services.
With
this
approach,
the
onsite
time
(
or
cost)
term
is
dropped
from
the
travel
cost
model
(
and
thus
the
wage
rate
would
not
enter
Equation
[
7.7]
).~

*
Brown
and
Mendelssohn
[
1980]
is
one
notable
exception.

tThis
practice
is,
strictly
speaking,
not
correct.
Even
though
onsite
time
is
constant
and
not
considered
a
choice
variable,
it
does
influence
the
cost
of
a
trip
(
see
Equation
[
7.4]).
Moreover,
it
cannot
be
treated
as
a
constant
displacement
to
the
demand
model's
intercept
because
the
opportunity
costs
of
time
can
be
expected
to
vary
across
individuals.

We
considered
a
role
for
onsite
time
under
the
assumptions
that
adjustm
e
n
t
f
o
r
simultaneity
was
unnecessary
and
that
the
results
w
e
r
e
u
n
i
f
o
r
m
l
y
unsatisfactory.
Without
an
explicit
recognition
of
the
simultaneity
between
Visits
and
onsite
time
costs,
ordinary
least­
squares
estimates
of
the
role
of
Onsite
time
costs
would
lead
to
the
conclusion
that
these
costs
were
unimportant
influences
on
the
demand
for
each
site's
services.

7­
31
The
second
approach
specifies
the
travel
cost
demand
function
for
each
site
to
include
the
costs
of
onsite
time
for
the
trip
in
which
the
individual
was
interviewed.
This
case
implicitly
assumes
that
the
time
spent
onsite
is
constant
for
all
trips
but
may
well
be
different
across
individuals.
Thus,
the
empirical
model
corresponds
to
the
theoretical
structure
developed
at
the
o
u
t
s
e
t
of
this
chapter.
The
first
approach
corresponds
to
the
basic
model
and
is
reported
in
this
section.
The
second
approach
is
used
to
gauge
the
implications
of
ignoring
onsite
costs.
These
results
are
summarized
in
Section
7
.
6
.

7.5.2
The
Opportunity
Cost
of
Travel
Time
As
noted
earlier,
it
has
often
been
argued
that
the
opportunity
cost
of
travel
time
is
less
than
the
wage
rate.
If
this
cost
is
known,
theory
suggests
that
travel
costs
and
the
cost
of
travel
time
have
equivalent
effects
on
the
demand
for
the
site's
services
(
i.
e.
,
their
parameters
in
a
linear
demand
model
would
be
equal).
In
the
absence
of
information
on
these
opportunity
costs
,
and
if
it
is
possible
to
assume
they
are
a
constant
fraction
of
every
individual's
wage
rate,
separate
effects
can
be
identified
for
travel
cost
and
the
cost
of
travel
time.
The
relationship
between
the
estimated
parameters
provides
one
basis
for
estimating
the
constant­­
essentially
the
McConnell­
Strand
[
1981
]
approach.
Of
course,
to
apply
this
approach,
independent
estimates
of
roundtrip
distance
to
the
site
and
travel
time
must
be
available.
Since
few
travel
cost
studies
have
had
access
to
this
type
of
information,
many
studies
accept
Cesario's
[
1976]
suggestion
that
the
opportunity
cost
of
travel
time
is
a
multiple
of
the
wage
rate
ranging
from
one­
fourth
to
one­
half
and
use
it
in
calculating
the
cost
of
a
trip.
In
these
cases,
travel
costs
and
travel
time
are
both
based
on
roundtrip
distance.
Of
course,
the
latter
also
requires
an
assumed
velocity
of
travel,
a
wage
rate,
and
the
Cesario
constant
to
estimate
the
opportunity
cost
of
travel
time.

Since
the
Federal
Estate
Survey
reports
travel
time
and
the
Zip
codes
of
each
respondent's
residential
location,
it
was
possible
to
develop
independent
estimates
of
both
components
of
the
cost
of
a
trip.
Thus,
tests
for
each
model
evaluate
the
appropriate
treatment
of
travel
costs
and
the
costs
of
travel
time.
These
tests
simply
translate
the
economic
Issues
and
ad
hoc
practices
into
restrictions
on
the
parameters
of
the
site
demand
models.
 
7.5.3
Results
for
the
Basic
Model
Table
7­
4
provides
the
ordinary
least­
squares
estimates
for
the
semilog
specification
of
our
travel
cost
demand
models.
The
general
form
for
the
model
is
given
in
Equation
(
7.31)
below:

I
n
(
V
i
)
=
a.
+
al(
TCi+
MCi)
+
a
3
l
N
C
i
+
&
i
,
(
7
.
3
1
)

where
vi
=
number
of
visits
during
the
recreation
season
for
the
ith
respondent
.

7­
32
Table
7­
4.
Regression
Results
of
General
Model,
by
Site
LN
VISITS
=
CYo
+
al
(
T+
M)
cosTsa
+
~
3
lNCONIEb
Site
Site
Fnumber
Intercept
T
+
M
cost
Income
R
2
DF
ratio
Allegheny
River
System,
PA
Arkabutla
Lake,
MS
Lock
and
Dam
No.
2
(
Arkansas
River
Navigation
System),
AR
Beaver
Lake,
AR
Belton
Lake,
TX
Benbrook
Lake,
TX
Berlin
Reservoir,
OH
Blakely
Mt.
Dam,
Lake
Ouachita,
AR
Canton
Lake,
OK
4
Clearwater
Lake,
MO
I
Cordell
Hull
Dam
and
Reservoir,
TN
z
DeGray
Lake,
AR
Dewey
Lake,
KY
Ft.
Randall,
Lake
Francis
Case,
SD
Grapevine
Lake,
TX
Greers
Ferry
Lake,
AR
Grenada
Lake,
MS
Herds
Creek
Lake,
TX
Isabella
Lake,
CA
Lake
Okeechobee
and
Waterway,
FL
Lake
Washington
Ship
Canal,
WA
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
0.53
(
2.04)

1.58
(
9.99)

2.31
(
9.76)

1.61
(
16.07)

1.69
(
9.38)

1.83
(
10.70)

1.40
(
8.47)

1.70
(
10.08)

1.77
(
8.61)

1.51
(
5.97)

1.86
(
14.13)

1.79
(
7.71)

0.42
(
2.27)

1.32
(
6.00)

1.80
(
16.12)

1.48
(
14.08)

2.04
(
12.61)

1.73
(
8.22)

1.26
(
5.55)

1.68
(
3.68)

0.96
(
2.69)
­
0.0005
(­
0.13)
8
.
2
x
1
0­
6
(
0
.
7
4
)

­
0.0093
(­
3.09)
6
.
2
X
10­
6
(
o.
67)

­
0.0125
(­
2.30)
­
1.8
x
10­
5
(
­
1
.
0
8
)

­
0.0066
(­
12.77)
­
3
.
5
X
10­
6
(­
o.
78
­
0.0052
(­
2.47)
2
.
6
x
1o­
6
(
o.
29)

­
0.0054
(­
4.11)
6.0
x
10­
6
(
0
.
8
0
)

0.0014
(
0.43)
­
4.1
x
10­
7
(
­
(
)
.
0
5
­
0.0079
(­
5.14)
­
7.6
x
10­
6
(
­
0
.
9
8
"

­
0.0206
(­
5.28)
7.1
x
10­
6
(().
86)

­
0.0032
(­
1.42)

­
0.0139
(­
6.00)

­
0.0070
(­
3.00)

­
0.0024
(­
2.95)

­
0.0066
(­
5.93)

­
0.0073
(­
8.80)

­
0.0065
(­
9.02)

­
0.0095
(­
4.36)

­
0.0050
(­
2.11)

­
0.0073
(­
3.15)

­
0.0268
(­
1.72)

­
0.0037
(­
3.79)
­
1.()
x
10­
5
(­
1.21)

­
1
.
2
x
10­
8
(
­
0
.
0
1
)

­
6
.
9
X
10­
5
(­
o.
73)

2.
OX
10­
5
(
2.02)

7.5
x
10­
6
(
0.91)

8.5
x
10­
6
(
1
.
7
0
)

8
.
4
x
1o­
6
(
1.42)

­
1.0
x
10­
5
(­().&
j)

­
2.1
X
10­
5
(
­
1
.
7
6
)

7
.
9
X
10­
6
(
0.81)

1
.
9
x
10­
7
(
0
.
0
1
)

1.7
X
11)­
5
(
o.
84)
0.01
0.15
0.14
0.43
0.12
0.30
0.01
0.24
0.28
0.04
0.34
0.17
0.18
0.43
0.47
0.28
0.22
0.19
0.20
0.10
0.26
66
0.29
58
4.93
38
3.11
224
86.07
50
3.39
43
9.11
93
0.09
88
13.67
71
13.98
71
1.61
101
25.57
46
4.68
43
4.72
47
17.61
89
39.12
214
40.79
73
10.02
51
5.95
45
5.47
27
1.56
41
7.18
DF
=
Degrees
of
freedom.
(
continued)

aT+
M
represents
the
respondents'
round
trip
cost.
It
is
composed
of
travel
time
cost
(
TCOST)
and
a
constant
per
mile
cost
of
operating
an
automobile
(
MCOST).

I
bt­
values
of
no
association
are
shown
in
parentheses.
8
Table
7­
4.
(
c
o
n
t
i
n
u
e
d
)

Site
FSite
number
Intercept
T
+
M
cost
Income
R
2
DF
ratio
Leech
Lake,
MN
Melvern
Lake,
KS
Millwood
Lake,
AR
Mississippi
River
Pml
No.
3,
MN
Mississippi
River
Pool
No.
6,
MN
Navarro
Mills
Lake,
TX
New
Hogan
Lake,
CA
New
Savannah
Bluff
Lock
&
Dam,
GA
Norfork
Lake,
AR
Ozark
Lake,
AR
Perry
Lake,
KS
Philpott
Lake,
VA
Pine
River,
MN
Pokegama
Lake,
MN
Pomona
Lake,
KS
Proctor
Lake,
TX
Rathbun
Reservoir,
TX
Sam
Rayburn
Dam
&
Reservoir,
TX
Sardis
Lake,
MS
Waco
Lake,
TX
Whitney
Lake,
TX
Youghiogheny
River
Lake,
PA
321
322
323
324
325
327
328
329
330
331
332
333
334
335
336
337
338
339
340
343
344
345
0.87
(
3.88)

1
.
3
0
(
4.47)

1.43
(
7.94)

1.33
(
4.20)

1.41
(
7.45)

1.66
(
6.40)

1.04
(
2.58)

1.88
(
8.39)

1.13
(
4.27)

1.66
(
8.52)

1.50
(
4.17)

1.90
(
9.28)

0.81
(
4.65)

1.44
(
7.28)

1.54
(
5.35)

2.06
(
13.61)

0.77
(
1.85)

1.46
(
7.06)

1.81
(
20.73)

1.95
(
15.04)

1.41
(
13.07)

0.29
(
0.60)
­
0.0022
(­
1.83)

­
0.0079
(­
1.66)

­
0.0081
(­
3.99)

­
0.0057
(­
4.62)

­
0.0074
(­
4.39)

­
0.0057
(­
1.39)

­
0.0040
(­
0.41)

­
0.0067
(­
1.44)

­
0.0047
(­
2.55)

­
0.0046
(­
4.44)

­
0.0042
(­
0.74)

­
0.0087
(­
4.40)

­
0.0017
(­
1.27)

­
0.0033
(­
4.46)

­
0.0058
(­
1.11)

­
0.0134
(­
7.50)

­
0.0015
(­
0.27)

­
0.0094
(­
2.83)

­
0.0030
(­
3.17)

­
0.0006
(­
0.32)

­
0.0025
(­
1.80)

0.0263
(
1.61)
3.5
x
10­
b
4.1
x
10­
6
1
.
8
x
1
0­
5
4.7
x
10­
6
1.3X
10­
5
­
1.4
x
10­
5
7.1
x
lt)­
6
­
9.8
X
1
0­
6
9.3
x
10­
5
­
8
.
8
x
10­
6
­
1.()
x
10­
5
­
1.7
x
10­
6
­
6.4
x
10­
6
­
1.4
x
10­
5
8
.
4
x
10­
6
1.2
x
10­
6
1.4X
10­
5
1.
OX
10­
6
4
.
3
x
10­
6
­
7.4
x
10­
6
3.2
x
10
­
6
1.7X
10
­
5
(
0.37)

(
0.32)

(
2.14)

(
0.54)

(
1.53)

(­
1.14)

(
0.60)

(­
0.70)

(
0.79)

(­
0.66)

(­
0.68)

(­
0.13)

(­
0.91)

(­
1.57)

(
0.62)

(
0.19)

(
0.82)

(
0.13)

(
0.78)

(­
1.25)

(
0.72)

(
1.55)
0.07
0.06
0.25
0.34
0.22
0.06
0.01
0.06
0.14
0.31
0.03
0.36
0.04
0.24
0,13
0.54
0.02
0.11
0.05
0.03
0.02
0.14
45
42
50
46
68
39
38
36
39
49
25
35
72
67
28
49
28
64
202
58
201
28
1.68
1.37
8.26
11.67
9.68
1.33
0.23
1.25
3.30
11.18
0.41
10.03
1.39
10.36
1.35
28.39
0.34
4.10
5.22
0.93
1.80
2.35
DF
=
Degrees
of
freedom.

aT+
M
represents
the
respondents'
t­
outld
trip
cost.
It
is
composed
of
travel
time
cost
(
TCOST)
and
a
constant
per
mile
cost
of
operating
an
automobile
(
MCOST).

bt­
values
of
no
association
are
shown
in
parentheses.
TCi
=
time
costs
of
travel
for
the
ith
respondent,
defined
as
product
of
the
estimated
wage
rate
for
the
person
(
see
Section
7.6)
and
the
roundtrip
travel
time
MCi
=
travel
costs
for
the
ith
respondent
INCi
=
family
income
for
the
ith
respondent
&
i
=
stochastic
error
for
ith
respondent.

Several
alternative
functional
forms
were
considered.
However,
the
results
uniformly
favored
the
semilog
form
based
on
the
ability
to
precisely
estimate
the
site
demand
parameters.
Moreover,
this
specification
is
generally
selected
in
evaluations
of
functional
forms
for
the
travel
cost
model
(
see
smith
[
1975al
I
smith
and
KOPP
[
19801,
and
Ziemer,
Musser,
and
Hill
[
1980]).

In
general
the
implicit
price
(
TC+
MC)
of
a
trip
to
the
site
is
statistically
significant
and
correctly
signed.
There
is
a
fairly
large
range
for
values
for
the
estimated
parameters
for
the
implicit
price­­
ranging
from
­
0.0005
to
­
0.0139.
Only
one
site
exhibited
a
positive
coefficient
for
the
implicit
price,
and
in
this
case
the
coefficient
would
not
be
judged
to
be
significantly
different
from
zero.
In
the
balance
of
the
models,
27
sites
had
coefficient
estimates
t
h
a
t
would
lead
to
the
judgment
o
f
a
d
e
m
a
n
d
e
f
f
e
c
t
s
i
g
n
i
f
i
c
a
n
t
l
y
d
i
f
f
e
r
e
n
t
f
r
om
z
e
r
o
at
least
at
the
5
­
p
e
r
c
e
n
t
l
e
v
e
l
.
The
balance
of
the
estimated
price
c
o
e
f
f
i
c
i
e
n
t
s
i
s
n
e
g
a
t
i
v
e
a
n
d
in
m
a
n
y
c
a
s
e
s
i
s
a
l
s
o
s
t
a
t
i
s
t
i
c
a
l
l
y
d
i
f
f
e
r
e
n
t
f
r
om
z
e
r
o
at
a
higher
s
i
g
n
i
f
i
c
a
n
c
e
l
e
v
e
l
­
­
i
.
e.
,
10
percent.

The
effect
of
income
is
poorly
measured
in
all
of
these
models.
In
most
cases
the
parameter
estimates
would
lead
to
the
conclusion
that
income
is
not
a
significant
determinant
of
the
demands
for
these
sites.
Indeed,
in
a
number
of
the
models
the
estimated
parameters
were
negative.
However,
these
estimated
parameters
would
lead
to
the
conclusion
that
income's
effect
was
not
significantly
different
from
zero.

At
first,
the
lack
of
significance
of
income
may
seem
surprising.
However
when
it
is
considered
in
comparison
to
other
recreation
applications
of
the
travel
cost
framework,
it
is
more
plausible.
For
the
most
part
these
sites
provide
high­
density
camping,
swimming,
boating,
etc.
These
are
activities
where
the
participation
decision
and
level
of
use
decisions
were
either
somewhat
insensitive
to
family
income
or
where
income's
marginal
,
effect
increased
and
then
decreased
with
increases
in
the
level
of
income.
Table
7­
5
summarizes
the
role
of
income
in
the
Cicchetti,
Seneca,
and
Davidson
[
1969]
analysis
of
recreation
participation
decisions.
Of
course,
it
should
be
acknowledged
that
these
participation
models
are
reduced
form
equations
reflecting
the
influence
of
both
demand
and
supply
influences
(
see
Smith
[
1975a]
and
Deyak
and
Smith
[
1978]
for
further
discussion
of
these
approaches).
Nonetheless,
they
provide
some
information
based
on
the
likely
implications
of
the
mix
of
activities
a
site
can
support
for
the
nature
of
the
demand
for
that
site's
services.

7­
35
Table
7­
5.
Summary
of
Cicchetti,
Seneca,
and
Davidson
[
1969]
Participation
Models
Equations
a
Activity
Probability
of
participation
Level
of
participation
Water­
based
Swimming
Marginal
effect
of
income
on
Effect
sensitive
to
probability
changes
with
region
of
residence
level
of
income
Water
skiing
Constant
marginal
Other
boating
Constant
marginal
Canoeing
Constant
marginal
Other
Activities
effect
b
Income
not
a
signi
­
ficant
determinant
effect
b
Marginal
effect
of
income
changes
with
level
of
income
b
effect
Income
not
a
significant
determinant
Camping
developed
Income
not
a
significant
Constant
marginal
determinant
effect
of
income
a
These
results
are
based
on
the
estimates
reported
in
Chapter
5
of
Cicchetti,
Seneca,
and
Davidson
[
1969].
b
These
estimated
parameters
were
substantially
smaller
in
numerical
magnitude
than
the
estimated
parameter
for
income
in
the
probability
equation
for.
fishing.

Finally
the
overall
explanatory
power,
as
measured
by
R
2,
is
also
quite
variable
across
sites.
In
some
cases,
such
as
sites
303
(
Beaver
Lake,
Arkansas),
313
(
Ft.
Randall,
Lake
Francis
Case,
South
Dakota),
314
(
Grapevine
Lake,
Texas),
and
337
(
Proctor
Lake,
Texas),
the
R
2
is
comparable
to
most
cross­
sectional
analyses.
For
the
remainder
it
is
somewhat
low,
indicating
that
there
may
be
other
major
factors
influencing
these
site
demands.

7.5.4
Results
for
the
Tailored
Models
It
should
be
acknowledged
that
while
the
basic
model
provides
a
plausible
specification
for
a
site
demand
equation,
there
may
well
be
a
number
of
other
determinants
of
these
demands.
Indeed,
the
low
R
2
would
certainly
support
this
conclusion.
Since
the
overall
objective
is
to
develop
a
general
model
for
projecting
the
effects
of
changes
in
any
water­
based
site's
characteristics
on
the
site
demand,
site
demand
equations
must
adhere
to
a
common
.

7­
36
­'
l
A
Nonetheless,
this
does
not
prevent
an
appraisal
of
the
sensitiv­
~
Pecification.
ity
of
t
h
e
basic
model's
parameter
estimates
to
the
inclusion
of
additional
As
a
consequence,
the
analysis
plan
considered
a
wide
array
of
variables
o
specifications
of
each
demand
function.
These
models
include
alternative
~
dditional
socioeconomic
information
­­
a
g
e
,
s
e
x
,
e
d
u
c
a
t
i
o
n
,
a
n
d
r
a
c
e
­
­
a
s
w
e
l
l
a
s
an
attitudinal
v
a
r
i
a
b
l
e
(
c
o
d
e
d
a
s
z
e
r
o
a
n
d
1
)
,
w
i
t
h
1
d
e
s
i
g
n
a
t
i
n
g
i
n
d
i
v
i
d
u
a
l
s
who
regarded
o
u
t
d
o
o
r
r
e
c
r
e
a
t
i
o
n
a
s
v
e
r
y
i
m
p
o
r
t
a
n
t
i
n
c
o
m
p
a
r
i
s
o
n
t
o
t
h
e
i
r
other
interests
(
R
EC
IMp).

Table
7­
6.
Comparison
of
Basic
Model
With
Tailored
Model:
Coefficient
f
o
r
(
TC+
MC)

Range
of
estimates
Site
name
Site
No.
Basic
model
tailored
models
Lock
and
Dam
No.
2
(
Arkansas
River
Navigation
System
)­,
AR
Beaver
Lake,
AR
Blakely
Mt.
Dam,
Lake
Ouachita,
AR
Cordell
Hull
Dam
and
Reservoir,
TN
.
Dewey
Lake,
AR
Grapevine
Lake,
TX
Greers
Ferry
Lake,
AR
Genada
Lake,
MS
Lake
Washington
Ship
Canal,
WA
Melvern
Lake,
KS
Millwood
Lake,
AR
Mississippi
River
Pool
No.
3,
MN
Mississippi
River
Pool
No.
6,
MN
Ozark
Lake,
AR
Philpott
Lake,
VA
Pine
River,
MN
Proctor
Lake,
TN
Sardis
Lake,
MS
Whitney
Lake,
TX
302
303
307
310
312
314
315
316
320
322
323
324
325
331
333
334
337
340
344
­
0.0125
­
0.0066
­
0.0079
­
0.0139
­
0.0024
­
0.0073
­
0.0065
­
0.0095
­
0.0037
­
0.0079
­
0.0081
­
0.0057
­
0.0074
­
0.0046
­
0.0087
­
0.0017
­
0.0134
­
0.0030
­
0.0025
­
0.010
to
­
0
.
0
1
3
­
0.0060
to
­
0.0070
­
0.0070
to
­
0.0080
­
0.0013
to
­
0.0015
­
0.0020
to
­
0.0030
­
0.0060
to
­
0.0090
­
0.0060
to
­
0.0070
­
0.0080
to
­
0.0100
­
0.0030
to
­
0.0400
­
0.0070
to
­
0.0090
­
0.0070
to
­
0.0090
­
0.0050
to
­
0.0060
­
0.0070
­
0.0030
to
­
0.0050
­
0.0070
to
­
0.0090
­
0.0010
to
­
0.0020
­
0.0013
to
­
0.0014
­
0.0030
to
­
0.0040
­
0.0020
to
­
0.0030
7­
37
Table
F­
5
in
Appendix
F
presents
a
sample
of
these
models
for
a
selected
set
of
the
43
sites.
These
cases
represent
the
site
demands
where
one
or
more
alternative
specifications
would
have
been
regarded
as
equivalent
or
better
to
the
basic
model.
In
evaluating
these
models,
the
focus
was
on
the
estimated
parameters
for
variables
that
were
common
between
the
basic
model
and
each
variation
to
it.
In
general,
the
most
important
parameter­­
the
coefficient
for
the
implicit
price­­
was
remarkably
stable.
Table
7­
6
provides
a
comparison
of
these
estimates
from
the
tailored
specifications
with
the
basic
model
estimates
reported
in
Table
7­
4.

Since
it
is
widely
acknowledged
in
the
econometrics
literature
that
pretesting
and
sequential
estimation
practices
affect
the
kinds
of
inferences
that
can
be
drawn
concerning
the
properties
(
i.
e.
,
unbiasedness,
efficiency,
etc.
)
of
the
"
final"
model's
estimated
parameters,
these
types
of
sensitivity
analyses
gauge
whether
the
decisions
required
to
select
the
final
models
were
important
to
the
parameters
of
central
importance
to
the
overall
objectives.
*
The
general
criteria
used
for
selecting
the
specifications
reported
in
Table
7­
4
were
based
on
three
considerations;
(
1)
agreement
between
the
sign
of
the
estimated
parameters
with
what
was
expected
from
economic
theory;
(
2)
statistical
significance
of
the
estimates
using
conventional
criteria
as
appropriate
indexes
of
the
precision
of
the
estimates;
and
(
3)
robustness
of
the
mea$
ured
effects
for
important
variables
(
such
as
TC+
MC)
to
model
specifications.

7.5.5
Evaluation
of
Measures
of
the
Opportunity
Cost
of
Travel
Time
Tables
7­
7
and
7­
8
report
the
results
of
two
sets
of
tests
for
the
basic
model
and
tailored
models,
respectively.
The
tests
have
been
structured
to
evaluate
alternative
definitions
of
the
opportunity
cost
of
travel
time.
The
two
models
can
be
readily
described.
The
first
maintains
that
the
wage
rate
is
the
most
appropriate
measure.
This
would
imply
that
the
measure
of
the
time
costs
of
travel,
TC,
can
be
added
to
the
travel
costs
as
in
Equation
(
7
.
3
1
)
.
A
l
t
e
r
n
a
t
i
v
e
l
y
,
i
f
,
as
several
authors
have
argued,
the
opportunity
cost
is
a
different,
constant
multiple
of
the
wage,
the
model
should
be
written
as:

l
n
V
i
=
60
+
ilTCi
+
ti2MCi
+
&
31NCi
+
&
.
(
7.32)

Thus,
if
the
~
age
rate
is
the
~
ppropriate
measure
of
the
opportunity
cost
of
travel
time,
al
should
equal
a2.
Rejection
of
this
null
hypothesis
would
therefore
provide
support
for
the
arguments
against
the
use
of
the
wage
rate
as
the
opportunity
cost.
The
sixth
column
of
Table
7­
7
reports
the
relevant
F­
statistic
and
significance
levels
for
this
hypothesis
using
the
basic
model.
Overall
the
hypothesis
is
rejected
for
9
of
the
43
sites
with
the
basic
model
at
the
5­
percent
significance
level.
These
decisions
are
generally
repeated
with
the
tailored
models
for
the
sites
reported
in
both
cases.

*
This
approach
is
et
al.
[
1978]
for
dealing
clearly
in
the
spirit
of
the
suggestion
made
by
Klein
with
estimation
problems.

7­
38
Table
7­
7.
F­
Test
for
Restriction
of
Genera\
Mocie\
 
 
 
.
­,

Hypothesis
1,
Full­
Time
Cost:
LN
Visits
=
aO
+
al
(
T
+
M)
Cost
+
us
Income
Hypothesis
2,
Cesario
Hypothesis:
a
LN
Visits
=
;
O
+
~
1
(
T
1/
3
+
M)
Cost
+
;
3
Income
Unrestricted
model:
LN
Visits
=
;
O
+
~
1
T
Cost
+
;
Z
M
Cost
+
~~
Income
Sum
of
Sum
of
Sum
of
squared
squared
squared
residuals,
F­
statistic
level
of
significance
Site
residuals,
residuals,
unrestricted"
Site
number
Hypothesis
1
Hypothesis
2
model
Ho:
al
=
az
Ho:
a
=
1/
3;
2
Allegheny
River
System,
PA
300
45.27
45.27
44.99
0.53
0.53
Arkabutla
Lake,
MS
301
24.84
24.12
23.93
0.14
0.50
Lock
and
Dam
No.
2
(
Arkansas
River
302
7.58
8.02
6.91
0.07
0.02
Navigation
System
),
AR
Beaver
Lake,
AR
303
104.64
109.61
104.09
0.27
0.01
Belton
Lake,
TX
304
25.90
25.82
23.71
0.04
0.04
f3enbrook
Lake,
TX
305
11.52
11.29
11.20
0.28
0.56
Berlin
Reservoir,
OH
306
62.13
61.93
61.70
0.43
0.56
4
Elakely
Mt.
Dam,
Lake
Ouachita,
AR
307
45.00
44.02
43.96
0.15
0.73
&
Canton
Lake,
OK
308
41.48
43.49
41.25
0.53
0.06
a
Clearwater
Lake,
MO
309
45.84
45.51
45.37
0.40
0.64
Cordeil
Hull
Dam
and
Reservoir,
TN
310
47.11
46.18
46.18
0.16
0.99
DeGray
Lake,
AR
311
22.45
22.62
22.45
0.99
0.56
Dewey
Lake,
AR
312
16.03
16.45
15.91
0.58
0.24
Ft.
Randail,
Lake
Francis
Case,
SD
313
24.34
26.34
24.05
0.46
0.04
Grapevine
Lake,
TX
314
22.64
25.40
21.34
0.02
0.01
a(
T
1/
3
+
M)
cost
represents
the
total
cost
of
a
round
trip
where
travel
time
is
evaluated
at
one­
third
of
the
predicted
wage
rate.
(
continued)
Table
7­
7.
(
continued)

Hypothesis
1,
Full­
Time
Cost:
LN
Visits
=
aO
+
al
(
T
+
M)
Cost
+
as
Income
Hypothesis
2,
Cesario
Hypothesis:
a
LN
Visits
=
;
O
+
;
I
(
T
1/
3
+
M)
Cost
+
~~
Income
Unrestricted
model:
LN
Visits
=
;
O
+
;
I
T
Cost
+
;
2
M
Cost
+
;
3
Income
Sum
of
Sum
of
Sum
of
squared
squared
squared
residuals,
F­
statistic
level
of
significance
Site
residuals,
residuals,
unrestricted
Site
number
Hypothesis
1
Hypothesis
2
model
Ho:
;
I
=
~
2
Ho:
;
=
1/
3;
2
Greers
Ferry
Lske,
AR
315
110.96
120.65
104.06
0.01
0.01
Genada
Lake,
MS
316
27.65
27.39
27.39
0.41
0.99
Herds
Creek
Lake,
TX
317
30.61
30.32
30.18
0.40
0.63
Isabella
Lake,
CA
318
23.59
23.45
23..
45
0.61
0.99
Lake
Okeechobee
and
Waterway,
FL
319
21.84
22.71
21.59
0.59
0.26
Lake
Washington
Ship
Canal,
WA
320
26.22
28.31
24.90
0.15
0.02
Leech
Lake,
MN
321
21.18
20.78
20.64
0.29
0.59
Melvern
Lake,
KS
322
31.37
31.16
31.15
0.59
0.91
Millwood
Lake,
AR
323
28.67
28.36
28.35
0.46
0.90
Mississippi
River
Pool
No.
3,
MN
324
20.68
22.59
20.63
0,74
0.04
Mississippi
River
Pool
No.
6,
MN
325
37.73
39.47
37.49
0.51
0.06
Navarro
Mills
Lake,
TX
327
23.44
23.59
23.30
0.64
0.50
New
Hogan
Lake,
CA
328
30.71
30.76
30.60
0.72
0.66
New
Savannah
Bluff
Lock
&
Dam,
GA
329
16.67
16.65
16.44
0.49
0.51
Norfork
Lake,
AR
330
18.45
19.58
17.53
0.17
0.04
Ozark
Lake,
AR
331
24.31
25.53
21.93
0.03
0.01
Perry
Lake,
KS
332
12.06
12.01
12.00
0.73
0.89
a(
T
1/
3
+
M)
cost
represents
the
total
cost
of
a
round
trip
where
travel
time
is
evaluated
at
one­
third
of
the
predicted
wage
rate.
(
continued)

I
1
1
e
Table
7­/.
(
continued)

Hypothesis
1,
Full­
Time
Cost:
LN
Visits
=
aO
+
al
(
T
+
M)
Cost
+
as
Income
Hypothesis
2,
Cesario
Hypothesis:
a
LN
Visits
=
&
o
+
&
l
(
T
1/
3
+
M)
Cost
+
;
3
I
n
c
o
m
e
Unrestricted
model:
LN
Visits
=
;
O
+
;
I
T
Cost
+
;
2
M
Cost
+
~~
I
n
c
o
m
e
Sum
of
Sum
of
Sum
of
squared
squared
squared
residuals,
F­
statistic
level
of
significance
Site
residuals,
Site
residuals,
unrestricted
number
Hypothesis
1
Hypothesis
2
model
Ho:
;
I
=
&
H
o
:
;
=
1/
362
Philpott
Lake,
VA
Pine
River,
MN
Pokegama
Lake,
MN
Pomona
Lake,
KS
Proctor
Lake,
TN
Rathbun
Reservoir,

Sam
Rayburn
Dam
&

Sarclis
Lake,
MS
y
Waco
Lake,
TX
&
J
Whitney
Lake,
TX
Youghiogheny
River
333
334
335
336
337
10
338
Reservoir,
TX
339
340
343
344
Lake,
PA
345
10.42
22.96
37.31
14.42
13.25
21.70
34.21
64.10
20.07
113.80
20.17
9.97
23.44
38.26
14.18
12.41
20.83
33.16
66.42
20.02
115.40
21.35
9.85
21.25
36.81
13.27
12.24
17.29
33.15
52.76
17.23
96.77
18.17
0.77
0.02
0.35
0.13
0.05
0.01
0.16
0.01
0.01
0.01
0.10
0.52
0.01
0.11
0.18
0.42
0.03
0.89
0.01
0.01
0.01
0.04
a(
T
1/
3
+
M)
cost
represents
the
total
cost
of
a
round
trip
where
travel
time
is
evaluated
at
one­
third
of
the
predicted
wage
rate.
t
Table
7­
8.
F­
Test
for
Restriction
of
Tailored
Models
a
Site
F­
statistic
level
of
significance
Site
number
Model
1
Model
2
Model
3
Model
4
Model
5
Lock
and
Dam
No.
2
(
Arkansas
River
Navigation
System),
A
R
Beaver
Lake,
A
R
Blakely
Mt.
Dam,
Lake
Ouachita,
AR
Cordell
Hull
Dam
and
Reservoir,
TN
Dewey
Lake,
KY
Grapevine
Lake,
TX
Greers
Ferry
Lake,
AR
Grenada
Lake,
MS
Lake
Washington
Ship
Canal,
WA
Melvern
Lake,
KS
Millwood
Lake,
AR
Mississippi
River
Pool
No.
3,
MN
Mississippi
River
Pool
No.
6,
MN
Ozark
Lake,
AR
Philpott
Lake,
WA
Pine
River,
MN
Proctor
Lake,
TX
Sardis
Lake,
MS
Whitney
Lake,
TX
302
303
307
310
312
314
315
316
320
322
323
324
325
331
333
334
337
340
344
0.07
0.29
0.18
0.20
0.49
0.03
0.01
0.35
0.59
0.41
0.49
0.99
0.54
0.03
0.16
0.03
0.06
0.01
0.01
0.07
0.32
0.13
0.20
0.63
0.04
0.01
0.47
0.20
0.61
0.46
0.88
0.28
0.02
0.08
0.03
0.09
0.01
0.01
0.05
0.06
0.17
0.46
0.16
0.02
0.01
0.36
0.10
0.46
0.84
0.88
0.56
0.03
0.17
0.02
0.16
0.01
0.01
0.03
0.20
0.30
0.16
0.58
0.03
0.01
0.35
0.18
0.61
0.46
0.64
0.76
0.03
0.14
0.02
0.07
0.01
0.01
0.05
0.34
0.30
0.22
0.87
0.05
0.02
0.20
0.16
0.99
0.46
0.75
0.55
0.13
0.04
0.02
0.05
0.01
0.01
aF­
tests
are
calculated
using
the
five
restricted
models
in
Table
7­
6
against
unrestricted
models
where
travel
time
and
mileage
cost
are
separate.
The
second
hypothesis
considers
Cesario's
suggestion
that
the
opportunity
is
a
multlple
o
f
t
h
e
w
a
g
e
r
a
t
e
.
The
explanation
for
the
parametric
cost
treatment
of
this
hypothesis
stems
from
the
definitions
of
the
components
of
of
a
trip.
TC,
the
time
costs
of
travel,
the
COSt
is
defined
as
the
predicted
A
wage
rater
s
a
y
w
,
times
the
travel
time,
t,
or
fit.
If
the
opportunity
cost
of
travel
time
is
some
multiple,
k
(
k
<
1
)
of
the
wage
rate
and
can
be
assumed
to
be
COfIStant
across
individuals,
the
true
measure
of
TC
(
designated
TC)
Both
travel
costs
and
the
time
costs
of
travel
should,
when
~
hould
b
e
kfit.
the
latter
is
correctly
measured,
have
the
same
effe~
t
on
the
demand
for
a
~
itejs
services.
Thus,
if
the
maintained
hypothesis
(
TC
=
kfit)
is
correct,
al
can
be
expected
to
be
equal
to
ci
2
.
However,
k
cannot
be
measured
.­
By
Usin9
~
t
a
s
a
proxy
and
assuming
that
k
is
co­
rnstant,
the
estimates
of
al
in
the
mgdel
using
Tc
=
WI
can
be
expected
to
be
al
=
kal.
Since
it
is
expected
that
al
=
`
1
a
n
d
t
h
a
t
a2
will,
under
ideal
conditions,
­
equal
a2,
the
Cesario
suggestion
can
be
treated
as
the
hypothesis
that
;
I
=
ka2
in
terms
of
Equation
(
7.32).
Since
­
Cesario's­
specific
suggestion
was
that
k
=
1/
3,
the
second
h
y
p
o
t
h
e
s
i
s
i
s
al
=
1
/
3
d
2
.
The
seventh
column
of
Table
7­
7
reports
the
results
for
this
test.
Nearly
twice
as
many
sites
(
16)
reject
this
null
hypothesis
with
the
basic
model
.

Thus,
there
is
greater
support
for
the
use
of
the
wage
rate
as
a
measure
of
the
opportunity
cost
of
travel
time
than
the
Cesario
one­
third
adjustment
to
the
wage.
However,
there
is
no
unambiguous
choice,
because
some
sites
fail
to
reject
both
sets
of
restrictions.

7.6
FURTHER
EVA
LLJATICIN
OF
THE
TRAVEL
COST
MODELS
Section
7.5
presented
estimates
of
the
final
models
for
each
of
43
recreation
sites.
As
noted
earlier,
the
methodology
developed
in
this
chapter
requires
that
the
individual
site
demand
equations
adopt
the
same
specification.
I
n
some
cases
this
specification
would
have
been
adopted
as
"
best,
"
and,
for
others,
the
choice
was
not
as
clearcut.
As
a
consequence,
it
was
necessary
to
evaluate
the
sensitivity
of
important
demand
parameter
estimates
to
the
model
specification.
There
are
several
additional
aspects
of
these
travel
cost
models
that
require
further
consideration.
Therefore,
this
section
collects
the
results
of
the
further
evaluations
of
these
models.
This
analysis
was
conducted
in
an
attempt
to
identify
potential
shortcomings
with
the
models
and
to
appraise
their
importance
for
the
estimated
values.
Most
of
these
difficulties
arise
from
either
econometric
problems
with
the
model
or
limitations
that
would
be
expected
based
on
the
economic
model
of
consumer
behavior
developed
at
the
outset
of
the
chapter.

The
first
aspect
of
these
travel
cost
models
requiring
further
consideration
arises
from
the
data
and
the
model
specification
themselves.
The
visit
measure
used
in
this
analysis
is
a
positive
integer
by
definition.
This
raises
a
number
of
potential
econometric
problems.
For
the
purpose
of
this
study
these
problems
have
been
ignored.
*
However,
where
possible,
appraisals
have
been
made
*
The
implications
of
these
features
for
the
site
demand
models
and
benefit
estimates
are
currently
being
evaluated
using
appropriately
structured
maximum
likelihood
estimators
and
recent
method
of
moments
approximations
proposed
by
Greene
[
1983]
.

7­
43
.:
of
the
potential
implications
of
one
of
the
most
important
aspects
of
the
sampie­­
that
it
observes
only
the
behavior
of
individuals
who
have
visited
each
site
at
least
once.
To
evaluate
the
potential
importance
of
the
bias
in
ordinary
leastsquares
estimates
as
a
result
of
the
truncated
form
of
the
measure
of
site
usage,
Olsen's
[
1980]
method
of
moments
approximation
of
the
maximum
likelihood
estimates
for
models
with
truncated
dependent
variables
has
been
used.

The
Olsen
method
relies
on
approximating
the
mean
of
the
conditional
distribution
for
the
dependent
variable
(
i.
e.
,
E
(
y
I
y
>
0)).
His
proposed
scaling
factors
use
a
first­
order
approximation
to
derive
a
relationship
between
the
ordinary
least­
squares
estimates
of
a
model's
parameters
and
the
maximum
likelihood
estimates.
They
can
be
estimated
from
the
moments
of
the
incomp
l
e
t
e
(
i.
e.,
truncated)
distribution.
These
scaling
factors
are
used
to
gauge
the
magnitude
of
the
differences
between
an
approximate
maximum
likelihood
estimator
and
ordinary
least
squares.
Thus,
as
Olsen
suggests,
they
provide
a
crude
index
of
the
potential
severity
of
the
problems
with
truncation.
Greene
[
1981
]
has
also
proposed
an
approach
for
adjusting
ordinary
leastsquares
estimates
in
the
present
Tobit
and
truncated
dependent
variable
`
models.
He
found
that
Olsen's
approximation
tends
to
overstate
the
bias.
Olsen's
approximation
will
be
closest
to
Greene's
approach
for
models
with
small
coefficients
of
determination
(
i.
e.
,
R
2)
.
A
s
R
2
i
n
c
r
e
a
s
e
s
t
h
e
Olsen
adjustment
will
tend
to
overstate
the
extent
of
bias.
Thus,
this
study's
screening
of
estimated
site
demand
models
provides
a
fairly
conservative
basis
for
gauging
the
bias
due
to
the
truncation
of
the
measure
of
site
usage.
Table
7­
9
reports
these
scaling
factors
for
the
33
sites
in
which
the
general
model
performed
well.

The
scaling
factors
in
the
fourth
column
of
Table
7­
9
can
be
interpreted
as
the
multiplicative
adjustment
coefficients
required
for
the
ordinary
leastsquares
parameter
estimates
to
approximate
the
maximum
likelihood
estimates
(
based
on
the
assumption
of
a
truncated
distribution).
Thus,
for
site
No.
301,
the
maximum
likelihood
estimates
would
be
15
percent
greater
than
the
ordinary
least­
squares
parameter
estimates
in
absolute
magnitude.
These
comparisons
suggest
that
several
sites
exhibit
pronounced
truncation
effects.
For
at
least
11
of
these
sites,
the
bias
associated
with
the
ordinary
least­
squares
estimates
may
well
be
quite
substantial.
As
a
consequence,
the
potential
for
differential
bias
in
the
estimates
of
these
site
demand
functions
is
accounted
for
in
the
final
model.
That
is,
the
generalized
least­
squares
estimates
relating
the
features
of
each
site
demand
function
to
the
site's
characteristics
have
been
derived
using
two
samples­­
one
including
all
sites
with
complete
data
(
i.
e.
,
sites
with
plausible
demand
models
and
complete
information
on
water
quality
and
other
site
characteristics)
and
a
second
omitting
those
sites
with
potentially
important
truncation
effects.

A
second
source
of
qualification
to
the
travel
cost
demand
model
arises
from
the
assumption
that
all
users
of
each
individual
site
have
the
same
derived
demand
for
that
site's
services.
In
most
cases,
disparities
in
onsite
time
could
not
be
accounted
for.
Moreover,
it
has
not
been
possible
to
adjust
for
the
different
mixes
of
activities
undertaken
by
different
individuals
at
the
7­
44
Table
7­
9.
Effects
of
Truncation
on
the
Travel
Cost
Models'
Estimates
Incomplete
mean/
Olsen
Site
standard
ML
sca~
ng
5ite
name
number
deviation
factor
~
rkabutla
Lake,
MS
Lock
&
Dam
No.
2
(
Arkansas
River
Navigation
System),
AR
Beaver
Lake,
AR
6elton
Lake,
TX
Benbrook
Lake,
TX
Blakely
Mt.
Dam,
Lake
Ouachita,
AR
Canton
Lake,
OK
Cordell
Hull
Dam
&
Reservoir,
T~
DeGray
Lake,
AR
Dewey
Lake,
KY
Ft.
Randall,
Lake
Francis
Case,
SD
Grapevine
Lake,
TX
Greers
Ferry
Lake,
AR
Grenada
Lake,
MS
Herds
Creek
Lake,
TX
Isabella
Lake,
CA
Lake
O
keechobee
and
Waterway,
FL
Lake
Washington
Ship
Canal,
WA
Leech
Lake,
MN
Melvern
Lake,
MS
Millwood
Lake,
AR
Mississippi
River
Pool
No.
3,
MN
Mississippi
River
Pool
No.
6,
MN
New
Savannah
Bluff
Lock
&
Dam,
GA
Norfork
Lake,
AR
Ozark
Lake,
AR
Philpott
Lake,
VA
Pine
River,
MN
Pokegama
Lake,
MN
Proctor
Lake,
TX
Sam
Rayburn
Dam
&
Reservoir,
TX
Sardis
Lake,
MS
Whitney
Lake,
TX
301
302
303
b
304
305
307
308
310
311
312:
313
314
315
316
317b
3
1
8
b
319b
3
2
0
b
321
322
3
2
3
b
324
325
329
3
3
0
b
331
333
b
334b
335
337
339
340
344
2.115
4.115
0.975
2.080
3.001
1.425
1.299
1.855
1.818
0.866
0.817
2.458
1.493
2.401
1.374
1.169
1.119
0.876
0.994
1.269
1.739
1.020
1.557
2.137
1.139
1.577
2.413
0.949
1.018
1.960
1.474
3.107
1
.82'
I
1.15
1.01
13.55
1.18
1.01
2.18
2.92
1.29
1.34
13.55
13.55
1.05
1.85
1.07
2.44
5.33
7.87
13.55
13.55
3.30
1.39
13.55
1.75
1.15
6.69
1.67
1.07
13.55
13.55
1.25
1.95
1.01
1.34
aThese
scaling
factors
are
assigned
approximately
using
Olsen's
Table
I
by
selecting
the
closest
value
for
the
reported
mean
to
standard
deviation
with
the
incomplete
distribution.
*
b
These
sites
were
omitted
for
truncation
bias
in
second
estimation
of
the
model.

7­
45
same
site.
*
Thus,
it
might
be
conjectured
that
the
same
demand
model
is
not
equally
well
suited
to
all
survey
respondents.
Such
a
hypothesis
would
imply
that
the
parameter
estimates
would
be
sensitive
to
sample
composition.
That
is,
deleting
individual
observations
associated
with
individuals
with
especially
long
onsite
time
or
rather
different
sets
of
activities
may
well
have
a
pronounced
effect
on
the
ordinary
least­
squares
estimates
of
the
model's
parameters
Moreover,
this
impact
may
be
differentially
important
to
subsets
of
the
sites
considered
for
this
analysis
because
there
are
substantial
differences
in
the
number
of
respondents
for
these
sites.

To
investigate
this
issue,
DFBETA
was
calculated
(
Belsley,
Kuh,
and
Welsch's
[
1980]
regression
diagnostic).
This
index
was
designed
to
act
as
an
aid
in
identifying
influential
or
outlying
observations.
It
is
not
a
statistical
test.
It
has
been
used
to
judge
the
"
influence"
of
specific
observations
on
this
study's
estimates
of
site
demand
parameters.
With
this
evaluation,
it
is
then
possible
to
consider
the
features
of
these
survey
respondents
to
evaluate
whether
there
are
economic
reasons
for
expecting
that
the
demand
patterns
of
these
individuals
must
be
explained
in
a
different
framework.
The
specific
index
used
is
defined
as
the
difference
between
the
ordinary
least­
squares
estimate
for
each
parameter
based
on
the
complete
sample
and
the
corresponding
estimate
based
on
the
sample
with
the
omission
of
one
observation.
These
indexes
were
calculated
for
each
parameter
and
each
observation.
A
review
of
these
estimates
indicated
that
no
single
observation
had
an
important
effect
on
the
estimated
parameters.
This
conclusion
was
found
for
all
sites,
including
those
with
a
somewhat
limited
number
of
sampled
recreation
ists.
While
this
finding
does
not
guarantee
that
the
effects
of
onsite
time
and
the
mix
of
recreation
activities
are
inconsequential
influences
on
site
demand,
it
does
suggest
that
they
are
unlikely
to
have
pronounced
effects
on
these
estimates.

The
final
aspect
of
the
travel
cost
models
that
requires
further
consideration
stems
from
the
relationship
between
decisions
on
the
number
of
trips
to
each
recreation
facility
and
the
amount
of
time
spent
onsite
per
trip.
As
noted
previously,
the
onsite
time
measure
relates
to
the
trip
each
respondent
was
undertaking
at
the
time
that
he
was
interviewed,
but
there
is
no
information
as
to
how
representative
this
trip
was.
That
is,
the
survey
does
not
identify
for
all
visits
during
the
season
the
amount
of
time
spent
onsite
per
t
r
i
p
.
Thus,
the
analyses
of
travel
and
onsite
time
costs
(
both
the
time
and
the
distance
components)
implicitly
assume
that
the
onsite
time
for
the
current
trip
is
a
good
indicator
of
the
onsite
time
for
all
past
trips.

If
this
assumption
is
reasonable,
then
it
is
also
plausible
to
consider
the
prospects
for
a
simultaneous
equation
model
to
describe
the
decisions
for
visits
to
a
site
and
the
time
on
the
site
per
trip.
When
using
simultaneous
equation
models
with
the
Federal
Estate
Survey
data,
two
aspects
of
consistency
in
recreation
choices
must
be
considered.

*
The
feasibility
Of
the
second
stage
models
being
investigated.
including
measures
of
the
activities
undertaken
into
for
the
estimated
site
demand
parameters
is
currently
.
First,
individuals
maY
decide
the
amount
of
time
to
be
spent
onsite­­
first
the
activities
they
wish
to
undertake
and
then
based
on
the
number
based
`
n
to
a
site
to
engage
in
these
activities.
Within
this
decision
frame­
~
f
visits
~
orkt
onsite
time
can
be
treated
as
exogenousiy
determined.
Visits
may
be
upon
these
onsite
time
choices.
This
would
not
imply
that
onsite
conditional
not
important
to
decisions
on
visits
to
a
recreation
facility.
~
i~
e
was
Rather,
suggest
that
they
are
not
joint
decisions.
Indeed,
for
some
cases
it
it
would
may
be
necessarY
to
se9ment
the
samPles
of
users
according
to
their
lengths
of
stay
on
the
site.
*

secOndl.
Y/
`
he
`
nsite
time
maY
not
be
constant
for
all
trips,
and
thus
t
h
e
~
easure
available
for
Per­
trip
time
onsite
is
inappropriate.
These
prospective
difficulties
in
evaluating
the
relationship
between
visit
and
onsite
time
deci
­

5ionS
wilt
therefore
influence
any
effort
to
model
their
respective
roles
in
recreation
site
demand
functions.
Nonetheless,
in
an
attempt
to
account
for
these
simultaneous
equation
effects,
onsite
time
has
been
treated
as
an
endo
­

9et10US
variable,
and
a
variety
of
specifications
have
been
considered
for
it
as
Well
as
for
the
site
demand
models
themselves.
in
general,
this
study
has
attempted
.
t?
`
nstrume~
talize
the
measures
of
the
variable
costs
of
onsite
time.
More
SpeClflCally,
onslte
cost
is
specified
as
a
nonlinear
combination
of
exogenous
and
endogenous
variables
as
a
result
of
the
respective
roles
for
the
,
opportunity
cost
of
time
and
onsite
time.

Following
conventional
practice
(
see
Kelejian
[
1971
]),
the
combination
is
treated
as
a
right­
hand­
site
endogenous
variable
and
the
models
were
estimated
with
two­
sta9e
least
squares.
~
The
first­
stage
instruments
were
composed
of
the
included
predetermined
variables
in
each
specification
for
the
travel
cost
model
along
with
age,
sex,
and
a
qualitative
variable
to
reflect
whether
the
recreation
activities
included
camping.
Several
variations
in
these
instruments
were
considered.
However,
this
set
of
variables
provided
acceptable
models
for
the
largest
set
of
site
demands.
Table
7­
10
reports
the
two­
stage
estimates
for
21
of
the
sites.
t
As
with
earlier
results
(
i.
e.
,
using
ordinary
least
squares
and
ignoring
onsite
time),
the
role
of
income
appears
quite
limited
for
nearly
all
sites.
Only
one
site
demand,
Millwood
Lake,
Arkansas
(
No.
323)
yields
a
statistically
significant
estimate
for
the
coefficient
of
family
income.
The
results
for
onsite
time
are
encouraging
but
certainly
not
clearcut.
As
suggested
by
the
theoretical
model,
onsite
time
(
SCOST)
affects
the
"
price"
of
a
trip
to
the
site
(
since
the
model
assumes
all
trips
have
the
same
onsite
time),
and
it
also
contributes
to
the
production
of
recreation
service
flows.

*
Our
analysis
with
regression
diagnostics
indicated
that
these
problems
Were
unlikely
to
be
present
in
our
models
because
the
results
were
not
sensitive
to
deleting
individual
observations.

~
ldeally,
the
Kelejian
method
calls
for
polynomials
in
the
predetermined
variables
as
first­
stage
instruments.
This
was
not
attempted
in
our
case
because
of
the
limited
number
of
observations
for
several
of
the
sample
recreation
sites.

fThe
21
sites
are
the
result
of
two
screenings
of
the
43
sites
in
the
Survey.
The
first
screening
eliminates
10
sites
with
implausible
demand
functions
The
second
eliminates
12
sites
that
experienced
truncation
bias.

7­
47
Based
on
the
first
of
these
impacts,
it
would
be
expected
to
have
a
negative
impact
on
the
demand
for
visits
to
a
site.
It
is
a
component
of
the
price
of
a
visit.
I
n
a
d
d
i
t
i
o
n
,
h
o
w
e
v
e
r
,
increases
in
the
time
spent
onsite
provide
one
means
of
substituting
for
visits.
Thus,
one
might
hypothesize
a
positive
"
substitution"
effect
on
the
demand
for
trips
to
a
recreation
site.
Of
course,
the
demand
model
reflects
a
composite
of
these
two
influences.

The
empirical
results
are
consistent
with
the
presence
of
these
opposing
influences
on
site
demand.
For
some
sites,
the
effect
of
SCOST
is
positive,
while,
for
others,
it
is
negative.
Five
of
the
21
site
demands
exhibit
statistically
significant
estimates
for
on
site
costs,
based
on
the
asymptotic
t­
ratios.
I
n
all
of
these
cases
the
estimated
coefficients
are
negative.

Table
7­
10.
Two­
Stage
Least­
Squares
Estimates
for
Selected
Travel
Cost
Site
Demand
Models
Estimated
travel
cost
model
S
a
Site
Site
Name
No.
Intercept
TC+
MC
SCOST
INC
AGE
~
z
___
Beaver
Lake,
A
R
Benbrcmk
Lake,
TX
Elakely
Mt.
Dam,
Lake
Ouachita,
AR
Canton
Lake,
OK
Cordell
Hull
Dam
&
Reservoir,
TX
DeGray
Lake,
KY
Ft.
Randall,
Lake
Francis
Case,
SD
Grapevine
Lake,
TX
Greers
Ferry
Lake,
AR
Grenada
Lake,
MS
Herds
Creek
Lake,
TX
Lake
Washington
Ship
Cana~,
WA
Leech
Lake,
MN
Millwood
Lake,
AR
Mississippi
River
Pool,
No.
6,
MN
Norfork
Lake,
AR
Philpott
Lake,
MN
Pokegama
Lake,
MN
Proctor
Lake,
TX
Sam
Rayburn
Dam
&
Reservoir,
TX
Whitney
Lake,
TX
3U3
305
307
308
310
311
313
314
315
316
317
320
321
323
325
330
333
335
337
339
344
1.705
(
11.45)
1.999
(
6.05)
1.721
(;:#

(
5.83)
1.603
(
7.58)
1.587
(
4.51)
1.778
(
4.74)
2.154
(
12.73)
1.607
(
10.60)
1.551
(
5.06)
1.938
(
5.81)
0.505
(
0.66)
0.293
(
0.79)
0.829
(
2.33)
1.198
(
3.81)
0.666
(
1.65)

(::%
1.344
(
3.62)
1.783
(
6.47)
1.157
(
3.38)
1.527
(
8.30)
­
0.0056
(­
8.12)
­
0.0052
(­
3.84)
­
0.
W81
(­
4.57)
­
0.0172
(­
2.83)
­
0.0137
(­
4.53)
­
0.
W83
(­
3.24)
­
0.
W42
(­
2.43)
­
0.
W53
(­
5.06)
­
0.
W66
(­
7.70)
­
o.
W73
(­
2.86)
­
0.0050
(­
2.06)
­
0.
W38
(­
2.40)
­
0.0032
(­
2.33)
­
0.0091
(­
3.88)
­
0.0062
(­
3.24)
­
o.
W55
(­
2.85)
­
o.
ooi'
4
(­
3.75)
­
0.0030
(­
3.54)
­
0.0149
(­
5.99)
­
0.0098
(­
2.92)
­
0.
W27
(­
1.44)
­
0.0W3
(­
2.21)
­
0.0001
(­
0.
s5)
­
0.0002
(­
0.40)
­
0.0008
(­
0.60)
­
0.0002
(­
0.48)
­
0.0004
(­
1.38)
­
o.
W21
(­
2.28)
­
0.0001
(­
2.31)
o.
00002
(
0.08)
­
0.
W16
(­
2.64)
­
0.0001
(­
0.19)
0.0465
(
1.07)
0.0004
(
1.26)
o.
000oW
(
0.02)
­
0.0006
(­
1.66)
­
0.0008
(­
0.28)
­
0.0007
(­
1.49)
­
0.0004
(­
1.32)
0.0003
(
0.84)
­
0.
wol
(­
0.29)
­
0.
oow
(­
4.15)
­
0.000003
(­
0.61)
0.000003
(
0.34)
­
o.
ooOOW
(­
0.81)
o.
ooOOW
(
0.47)
o.
ooooo3
(
0.3s)
­
o.
wooo2
(­
0.21)
0.000009
(
0.79)
O.
ooww
(
1.72)
o.
ooOOW
(
1.48)
o.
oOoo1
(
0.71)
­
0.00002
(­
1.77)
o.
00002
(
0.78)
O.
000011
(
0.9s)
o.
00002
(
2.
s4)
0.00002
(
1
.9s)
o.
oOOo1
(
0.91)
o.
m2
(
0.18)
­
o.
ooOOW
(­
0.86)
o.
mz
(
0.33)
0.000002
(
0.19)
o.
woO06
(
1.11)
­
0.0009
(­
0.27)
­
o.
W20
(­
0.29)

(%'
o.
W43
(
0.
S8)
O.
W72
(
1.71)
0.0104
(
1.44)
­
O.
W87
(­
0.94)
­
0.0129
(­
2.91)
­
0.0045
(­
1.09)
0.0100
(
1.99)
­
o.
W30
(­
0.36)
­
0.0018
(­
0.15)
0.0069
(
0.93)
0.0134
(
1.88)
0.0070
(
0.97)
0.0149
(
1.
s5)
­
0.0062
(­
1.07)
0.0020
(
0.34)
0.0049
(
1.01)
0.0102
(
2.
W)
0.0078
(
1.88)
0.42
0.31
0,21
0.26
0.35
0,21
0.38
0.50
0.28
0.26
0.20
0.21
0.17
0.30
0.24
0.20
0.47
0.24
0.56
0,17
0.10
 
 
 
`
The
numbers
in
Parentheses
below
the
estimated
coefficients
l
 
re
asymptotic
t­
ratios
for
the
null
hypothesis
d
*
association.

7­
48
The
remaining
sites
also
were
modeled
within
a
simultaneous
framework.
HoWevert
in
these
cases
the
parameters
estimates
were
inferior
to
those
derived
using
ordinary
least
squares
under
the
assumption
of
constant
onsite
time.
As
a
rule,
the
estimated
effect
of
travel
cost
(
TC+
MC)
was
not
statistically
significant
and,
in
some
cases,
suggested
a
positive
effect
on
site
demand.
Moreover,
the
estimated
effects
of
onsite
costs
were
generally
statistically
insignificant.
Thus,
the
models
reported
in
Table
7­
10
are
the
cases
in
which
the
simultaneous
estimates
were
judged
to
be
equivalent
or
better
than
the
ordinary
least­
squares
results
reported
in
Section
7.5.

These
results
are
important
for
two
r'easons.
They
attempt
to
deal
with
Onsite
time
costs
and
travel
costs
within
a
single
demand
framework.
Most
authors
(
see
­
Brown
and
Mendelssohn
[
1980]
as
a
notable
example)
have
either
attempted
to
partition
their
samples
according
to
the
time
spent
onsite
and
estimate
separate
demand
models
for
each
grouping
or
have
assumed
that
onsite
time
was
not
important
to
the
decisions
for
trips
to
a
recreation
facility.
This
latter
assumption
might
be
the
result
of
features
of
the
recreation
activities
undertaken
and
site
selected
or
simply
because
the
time
onsite
was
approximately
constant
across
trips.

Table
7­
11.
Comparison
of
Ordinary
Least­
Squares
and
Two­
Stage
Least­
Squares
Estimates
of
Travel
Cost
(
TC
i
+
MC
i)
Parameters
Two­
Ordinary
stage
least­
leastsquares
Site
name
squares
Site
No.
estimate
estimate
Beaver
Lake,
AR
Benbrook
Lake,
TX
Blakely
Mt.
Dam,
Lake
Ouachita,
AR
Canton
Lake,
OK
Cordell
Hull
Dam
&
Reservoir,
TX
De
Gray
Lake,
AR
Ft.
Randall,
Lake
Francis
Case.,
SD
Grapevine
Lake,
TX
Greers
Ferry
Lake,
AR
Grenada
Lake,
MS
Herds
Creek
Lake,
TX
Lake
Washington
Ship
Canal,
WA
Leech
Lake,
MN
Millwood
Lake,
AR
Mississippi
River
Pool
No.
6,
MN
Norfork
Lake,
AR
Philpott
Lake,
VA
Pokegama
Lake,
MN
Proctor
Lake,
TX
Sam
Rayburn
Dam
&
Reservoir,
TX
Whitney
Lake,
TX
303
305
307
308
310
311
313
314
315
316
317
320
321
323
325
330
333
335
337
339
344
­
0.0066
­
0.0054
­
0,0079
­
0.0206
­
0.0139
­
0.0070
­
0.0066
­
0.0073
­
0.0065
­
0.0095
­
0.0050
­
0.0037
­
0.0022
­
0.0081
­
0.0074
­
0.0047
­
0.0087
­
0.0033
­
0.0134
­
0.0094
­
0.0025
­
0.0056
­
0.0052
­
0.0081
­
0.0172
­
0.0137
­
0.0083
­
0.0042
­
0.0053
­
0.0066
­
0.0073
­
0.0050
­
0.0038
­
0.0032
­
0.0091
­
0.0062
­
0.0055
­
0.0074
­
0.0030
­
0.0149
­
0.0098
­
0.0027
7­
49
..&
23!
sP
,.
._.,
.­
­
 
 
i
Table
7­
12.
Hausman
Test
for
Differences
Between
Two­
Stage
Least­
Squares
and
Ordinary
Least­
Squares
Estimates
Site
2SLS
&
l
OLS
2SLS
­
G1
G1
O
LS
d~
`
A
R2SLS
­
VAROLS
t­
statistic
303
305
307
308
310
311
313
314
315
316
317
320
4
321
Jl
o
323
325
330
333
335
337
339
344
­
0.0010
0.0002
­
0.0002
0.0034
0.0002
­
0.0013
0.0024
0.0020
­
0.0001
0.0022
­
0.0000248
­
0.0001
­
0.0010
­
0.0010
0.0012
­
0.0008
0.0013
0.0003
­
0.0015
­
0.0004
­
0.0002
0.0000005
0.0000018
0.0000031
0.0000368
0.0000092
0.0000066
0.0000030
0.0000011
0.0000007
0.0000065
0.0000058
0.0000027
0.0000018
0.0000055
0.0000037
0.0000037
0.000003859
0.0000007
0.0000062
0.0000112
0.0000034
0.0000003
0.0000017
0.0000024
0.0000152
0.0000054
0.0000054
0.0000012
0.0000007
0.0000005
0.0000047
0.0000056
0.0000010
0.0000014
0.0000041
0.0000028
0.0000034
0.000003895
0.0000005
0.0000032
0.000011
0.0000019
0.000447
0.000316
0.000837
0.004648
0.001949
0.001095
0.001342
0.000633
0.000447
0.001342
0.000447
0.001304
0.000633
""­
0.001183
0.000949
0.000548
0.0!;
447
0.001732
0.000447
0.001225
2.237
0.633
­
0.239
0.731
0.103
­
1.187
1.788
3.160
­
0.224
1.639
­
0.05
­
0.077
­
1.580
­
0.845
1.264
­
1.460
NA
0.671
­
0.866
­
0.894
­
0.163
Notes:

NA
=
The
t­
statistic
could
not
be
calculated
as
the
variance
since
the
ordinary
leastsquares
estimate
was
greater
than
the
two­
stage
least­
squares
estimate.

61
=
the
estimated
coefficientof
the
travel
plus
mileage
cost
variable.

VAR
=
the
variance
ofdl.

2SLS
=
the
two­
stage
least­
squares
model.

OLS
=
the
ordinary
least­
squares
model.
of
COUr=
el
this
perspective
is
implicitly
adopted
for
the
results
in
the
preViOUS
S
e
c
t
i
o
n
.
T
h
u
s
,
t
h
e
s
e
c
o
n
d
p
o
t
e
n
t
i
a
l
u
s
e
o
f
t
h
e
s
e
f
i
n
d
i
n
g
s
i
s
t
o
how
Important
a
n
`
rrOr
the
failure
to
take
account
Of
simultaneity
m
i
g
h
t
9auge
he
use
of
t
h
e
g
e
n
e
r
a
l
m
o
d
e
l
s
t
o
d
e
r
i
v
e
a
b
e
n
e
f
i
t
e
s
t
i
m
a
t
i
o
n
f
r
a
m
e
w
o
r
k
.
be
for
t
Table
7­
1
1
reports
a
comparison
of
the
ordinary
least­
squares
estimates
of
the
travel
cost
parameter
versus
the
two­
stage
results
for
each
of
the
sites
where
least
squares
were
judged
to
be
at
least
as
good
as
the
ordinary
the
twO­
sta9e
least­
squares
models.
Overall
the
results
are
quite
similar.
There
are
two
types
of
comparisons
that
can
be
made
between
these
estimates.
As
a
practical
matter
~
for
benefit
estimation,
the
numerical
differences
between
the
ordinary
least
squares
and
two­
stage
least­
squares
estimates
are
of
concern.

For
the
most
part,
the
two
sets
of
estimates
for
the
(
TC.
+
MC.
)
parameter
are
quite
comparable.
A
second
comparison
involves
considering
`
whether
the
~
UII
hypothesis
that
the
parameters
for
the
travel
and
time
cost
variable
were
equal
in
the
two
models
would
be
rejected
based
on
these
estimates.
It
is
possible
to
develop
an
asymptotic
test
for
this
hypothesis
using
Hausman's
[
1978]
approach
to
specification
tests.
Hausman
derives
an
expression
for
the
variance
of
the
difference
between
two
estimators
of
the
same
parameter.
These
estlmat~
rs
are
defined
for
two
hypotheses.
It
must
be
assumed
that
one
is
a
consistent
estimator
under
both
the
null
and
alternative
hypotheses
and
that
the
second
estimator
is
asymptotically
efficient
under
the
null
but
inconsistent
under
the
alternative
hypothesis.
Given
asymptotic
normality
and
these
assumptions
the
variance
of
the
difference
between
the
estimators
is
the
difference
in
their
respective
variances.
This
application
considers
the
difference
between
the
two­
stage
least­
squares
and
ordinary
least­
squares
estimates
of
the
coefficient
for
the
travel
cost
variable.
Constructing
the
corresponding
t­
ratio
gives
the
following:

2SLS
­
~
OLS
t
Gl
=
.
(
7.33)

VAR(
iil
2SLS)
­
VAR(
iilOLs)

Table
7­
12
reports
the
details
of
the
calculation
of
these
test
statistics.
The
t­
ratio
will
follow
an
asymptotically
normal
distribution.
Considering
these
statistics
as
an
approximate
basis
for
testing
the
difference
between
these
coefficient
estimates
gives
only
two
cases
(
Sites
303
and
314)
in
which
the
null
hypothesis
of
equality
would
be
rejected
at
the
5­
percent
significance
level.
Thus,
these
findings
largely
confirm
the
informal
judgmental
inspection
and
indicate
that
the
ordinary
least­
squares
models,
which
assume
onsite
costs
to
be
constant,
are
unlikely
to
have
serious
errors
because
of
this
assumption.

7.7
ANALYZING
THE
ROLE
OF
WATER
QUALITY
FOR
RECREATION
DEMAND
The
last
step
in
the
empirical
modeling
involves
estimating
the
role
of
water
quality
and
other
site
attributes
in
the
demands
for
a
site's
services.
The
structure
of
the
model
has
been
detailed
in
Section
7.3.
Thus,
what
remains
to
be
presented
is
a
specific
description
of
the
results
of
the
application
The
overall
objective
is
to
attempt
to
explain
the
observed
variation
in
7­
51
each
of
the
estimated
demand
parameters
across
sites
by
the
characteristics
of
those
sites.
With
such
a
model,
it
is
possible,
in
principle,
to
characterize
the
change
in
a
site's
demand
in
response
to
a
change
in
any
of
the
factors
inf
Iuencing
those
demand
parameters.
Thus,
it
would
be
feasible
to
evaluate
the
implications
of
a
change
in
water
quality
for
the
demand
for
the
site's
services,
even
though
the
change
has
not
been
experienced.
This
ability
arises
from
the
fact
that
this
model
provides
a
general
description
of
the
factors
that
influence
the
features
of
site
demands
within
a
single
framework.

The
model
has
been
derived
from
two
subsets
of
the
43
site
d
e
m
a
n
d
models
described
in
Section
7.5
above.
The
first
of
these
included
33
sites
with
plausible
site
demand
function
s.*
The
second
restricts
the
sample
further
by
eliminating
11
of
these
sites,
based
on
estimates
of
the
Olsen
scaling
factors
reported
in
Table
7­
9.
As
noted
earlier,
these
scaling
factors
provide
some
indication
of
the
prospects
for
bias
due
to
the
truncation
in
the
measures
of
site
usage.
These
11
sites
exhibited
the
largest
values
of
the
estimated
scaling
factor,
ranging
from
5.33
to
13.55.
The
specific
sites
eliminated
from
the
sample
are
footnoted
in
Table
7­
9
on
page
7­
45.

To
develop
estimates
of
the
influence
of
site
characteristics
on
the
parameters
describing
a
site's
demand
function,
the
attributes
involved
must
be
identified.
As
indicated
in
Section
7.4,
the
information
on
the
site
characteristics
was
obtained
from
U.
S.
Army
Corps
of
Engineers.
These
data
were
augmented
with
information
on
water
quality
from
the
U.
S.
Geological
Survey.
As
a
rule,
the
Corps
of
Engineers
data
were
measures
of
the
size
of
the
area
and
types
of
equipment
available.
The
water
quality
information
consisted
of
monthly
readings
from
June
through
September
of
the
year
of
the
survey
for
seven
measures
of
water
quality,
including
dissolved
oxygen,
fecal
coliform
density,
pH,
biochemical
oxygen
demand,
phosphates,
turbidity,
and
total
suspended
solids.
Two
water
quality
indexes
were
also
developed
from
these
data
for
each
month­­
the
RFF
water
quality
index
(
see
Vaughan
in
Mitchell
and
Carson
[
1981
]
)
and
the
NSF
index.
Since
the
specific
features
of
these
indexes
were
described
in
Section
7.4,
their
definitions
will
not
be
repeated
here.
Table
7­
13
summarizes
the
primary
site
characteristics
considered
from
the
Corps
of
Engineers
data.

Unfortunately,
t
h
e
r
e
a
r
e
f
e
w
~
p
r
i
o
r
i
insights
one
can
derive
from
economic
theory
regarding
which
subset
of
these
variables
is
most
likely
to
influence
the
estimated
parameters
of
site
demand
models.
While
the
primary
focus
was
on
the
water
quality
measures,
the
analysis
considered
a
number
of
alternative
specifications,
including
subsets
of
the
site
characteristics
reported
in
Table
7­
13.
The
variables
with
the
most
consistent
association
with
the
demand
parameters
over
the
specifications
considered
included
a
measure
of
the
size
of
the
site
(
i.
e.
,
SHORMILE),
i
t
s
a
c
c
e
s
s
p
o
i
n
t
s
(
i
.
e
.
,
M
U
L
T
I
+
ACCESS),
and
the
size
of
the
water
body
relative
to
the
overall
site
size
(
i.
e.
,
AREAP/
AREAT).
This
selection
does
not
seem
particularly
surprising.
Each
variable
can
be
interpreted
as
a
crude
measure
of
the
capacity
of
the
*
Appendix
F
presents
the
benefit
estimates
if
all
33
sites
are
used
in
the
model.

4
7­
52
1
Table
7­
13.
Description
of
U.
S.
Army
Corps
of
Engineers
Data
on
Site
Characteristics
~
ariable
name
Description
@
RMILE
Total
shoreline
miles
at
the
site
during
peak
visitation
period
pREAT
Total
site
area,
land
and
water
in
acres
Af?
EAp
Pool
surface
acreage
on
fee
and
easement
lands
during
peak
visitation
period
MULTI
Number
of
developed
,
multipurpose
recreation
areas
onsite
ACCESS
Number
of
developed
onsite
access
areas
CORPICK
Number
of
Corps­
managed
onsite
picnic
locations
OTH
PICK
Number
of
other
agency­
managed
onsite
picnic
locations
CORCMPD
Number
of
Corps­
managed
developed
camp
sites
OTH
CMPD
Number
of
other
agency­
managed
camp
sites
CORLN
Number
of
Corps­
managed
onsite
ing
lanes
OTH
LN
Number
of
other
launching
lanes
agency­
managed
developed
boat
launchonsite
boat
DOCK
PR
Number
of
onsite
private
boat
docks
DOCKCO
Number
of
onsite
community
docks
FLOAT
Number
of
onsite
floating
facilities
(
e.
g.
,
water
ski
jump,
swimming
floats,
fishing
floats,
etc.
)

7­
53
\
I
site
to
provide
services
that
would
support
different
types
of
recreation
service
flows.

It
was
more
difficult
to
isolate
measures
of
water
quality
that
appeared
to
influence
the
estimated
site
demand
parameters.
While
the
final
generalized
least­
squares
estimates
for
the
determinants
of
site
demand
parameters
seem
exceptionally
good,
there
are
a
number
of
reasons
for
caution
in
interpretin
g
these
findings,
as
shown
by
a
review
of
the
approaches
used
to
develop
them.

T
h
e
modelina
of
the
role
of
water
aualitv
considered
a
wide
arrav
of
potential
specif
ica~
ions
of
its
effects,

.
The
monthly
and
average
season)
readings
for
the
ures
of
the
variation
in
considered.
­,­.
including
each
of
the
following:

(
across
the
4
months
of
the
summer
two
water
quality
indexes
and
meas
­
the
index
over
the
4
months
were
.
The
monthly
and
average
readings
for
specific
components
of
the
index
(
i.
e.
,
dissolved
oxygen,
total
suspended
solids,
etc.
)
were
considered
individually
and
in
sets
using
existing
information,
where
possible,
to
avoid
the
joint
presence
variables
that
might
be
measuring
common
phenomena.

.
Temporal
effects
of
individual
pollutants
were
considered
in
an
attempt
to
isolate
"
best"
or
most
relevant
indexes
of
water
quality.

With
a
few
notable
exceptions
these
results
led
to
either
insignificant
or
unstable
estimates
of
the
effects
of
water
quality
on
the
site
demand
parameters.

Only
in
the
case
of
dissolved
oxygen
did
this
pretesting
of
model
specifications
lead
to
a
stable
and
statistically
significant
association
between
the
variation
in
the
estimated
site
demand
parameters
and
the
mean
and
variance
in
the
level
of
dissolved
oxygen
over
the
summer
period.
This
association
is
more
clearcut
with
the
smallest
samples.
Clearly,
these
findings
are
consistent
with
the
earlier
Vaughan­
Russell
[
1981
]
and
Nielsen
[
1980]
analyses
supporting
the
use
of
dissolved
oxygen
as
an
ideal
measure
of
water
quality
for
evaluating
recreation
fishing.
Nonetheless,
it
should
be
acknowledged
that
the
missing
data
problem
is
especially
important
for
this
study's
water
quality
variables
(
see
Section
7.4
above).
The
procedure
has
been
to
use
the
sample
mean
for
those
sites
with
missing
water
quality
information.
Thus,
a
smaller
number
of
actual
readings
on
water
quality
are
what
should
be
regarded
as
the
basis
of
the
measured
association
between
water
quality
and
the
estimated
site
demand
parameters.
This
does
not
imply
that
the
use
of
means
was
inappropriate.
Rather,
it
indicates
that
there
was
little
observed
variation
in
any
of
the
water
quality
variables
to
associate
with
the
estimated
demand
parameters.
l
*
The
indexes
of
water
quality
(
i
.
e.
,
the
RFF
and
NSF)
tend
to
reduco
the
variation
present
in
their
components.
Thus,
there
was
very
little
variation
in
these
indexes
across
sites.

7­
54
half
of
the
22
sites
in
the
restricted
sample
had
complete
water
~
pproxitnatel
Y
information.
Thus,
the
preference
for
the
dissolved
oxygen
measure
~
UalitY
.
might
~
ell
be
altered
with
more
complete
water
quality
data.

Table
7­
14
N3pOrts
the
generalized
least­
squares
estimates
for
the
final
model
with
both
samples.
*
The
parameters,
Uo,
al,
and
Ua,
correspond
to
the
general
model
specifications
as
given
in
Equation
(
7.31).
These
results
clearly
favor
the
model
based
on
the
restricted
sample.
Increases
in
the
average
level
of
dissolved
oxygen
would
be
improvements
in
water
quality.
The
results
using
this
restricted
sample
indicate
that
such
increases
w
o
u
l
d
the
demand
at
all
implicit
prices
(
i.
e.
,
travel
costs)
and
would
also
increase
increase
the
degree
Of
inelasticity
in
the
demand
curve.
This
second
effect
the
site's
ability
to
support
a
wider
range
of
recreation
activi
­
~
imply
reflects
ties
with
the
improved
water
quality.

Given
the
Poor
Performance
of
income
as
a
determinant
of
the
demand
for
any
one
of
the
site's
services,
it
is
not
surprising
that
the
second
step
model
for
the
income
parameter
is
incapable
of
explaining
the
variation
in
the
site
demand
parameters.

The
most
striking
difference
between
the
results
estimated
with
the
two
samples
arises
with
the
estimated
coefficients
for
the
travel
cost
variable.
The
estimated
effects
of
the
site
attributes,
including
the
water
quality
measures,
are
all
significantly
different
from
zero
and
generally
consistent
in
sign
with
2
priori
expectations.
The
differences
between
the
two
samples
would
seem
to
provide
indirect
evidence
of
the
importance
of
truncation
effects
on
the
travel
cost
site
demand
models.

These
generalized
least­
squares
results
do
not
include
R*
measures
of
goodness
of
fit
because
the
conventional
R*
statistic
is
no
longer
confined
to
the
O
to
1
interval
when
calculated
based
on
the
generalized
least­
squares
residuals.
Thus,
it
does
not
have
the
same
interpretation
as
the
R*
statistics
reported
with
the
ordinary
least­
squares
results
(
see
Cicchetti
and
Smith
[
1976]
Appendix
B
for
more
details).

*
See
Section
7.3
for
a
detailed
discussion
of
the
construction
of
the
generalized
least­
squares
estimator.
It
should
be
noted
that
Vaughan
and
Russell
[
1981
]
have
used
a
similar
methodology
in
their
valuation
of
recreation
fishing
days.
However,
their
approach
combined
the
two
equations
by
substituting
the
second
step
model
for
the
determinants
of
site
demand
parameters
(
Equation
7.22)
into
Equation
(
7.21
)
to
derive:

Y
i
=
x
i
eA
i
+
&
i
.

This
model
includes
interaction
terms
in
the
determinants
of
site
demands
and
site
attributes.
It
provides
an
equivalent
description
of
the
two­
step
approach
used
in
this
study.
However,
there
is
one
advantage
to
the
two­
step
approach
in
specification
analysis
of
the
models.
It
allows
the
specification
of
the
determinants
of
site
demand
to
be
treated
separately
from
the
determinants
of
variations
in
site
demand
parameters.
Each
specification
for
the
combined
model
includes
assumptions
about
both.

7­
55
Table
7­
14.
Generalized
Least­
Squares
Estimates
of
Determinants
of
Site
Demand
Parameters
a
tin
ii,
&
q
Independent
variable
33
site
22
site
33
site
22
site
33
site
22
site
1
ntercept
1.2959
1.5106
0.0005
­
0.0246
0
.
5
3
x
10­
5
0
.
5
4
x
10­
5
(
3.768)
(
4.081)
(
0.203)
(­
9.480)
(
0.330)
(
0.308)

SHORMILE
­
0.0003
0.0003
0
.
4
7
x
10­
6
­
0
.
1
3
x
10­
4
­
0
.
1
4
x
10­
7
0
.
9
7
x
10­
9
(­
1
.304)
(
1
.250)
(
0.256)
(­
6.763)
(­
1.408)
(
0.089)

(
MULTI
+
ACCESS)
0.0017
­
0.0059
­
0
.
4
1
x
10­
4
0
.
7
7
x
10­
4
0
.
2
2
x
10­
6
0
.
4
7
x
10­
6
(
0.464)
(­
1
.502)
(­
1.586)
(
2.810)
(
1.299)
(
2.562)

AREAP/
AREAT
­
0.1686
­
0.3950
­
0.0025
0.0033
O.
lox
1
0
­
4
­
0
.
1
9
x
10­
5
(­
1.116)
(­
1
.752)
(­
2.190)
(
2.273)
(
1
.423)
(­
0.181)

4I
Mean
dissolved
0.0049
0.0045
­
4
.
2
x
Io­
4
0.0002
­
0.12X
1
0
­
6
­
0
.
1
2
x
10­
6
~
oxygen
(
1.220)
(
1.065)
(­
1.514)
(
5.992)
(­
0.642)
(­
0.604)

Variance
in
0.0003
0.0005
­
0
.
1
7
X
1
0
­
5
0
.
9
8
x
1
0­
5
­
0
.
7
3
x
10­
8
0
.
9
4
x
1
0
­
'
0
dissolved
oxygen
(
1.131)
(
1
.862)
(­
0.751)
(
4.077)
(­
0.617)
(
0.007)

a
The
numbers
in
parentheses
below
the
estimated
coefficients
are
the
asymptotic
t­
ratios
for
the
null
hypothesis
of
no
association.
,.
8
~
MEASURE
OF
THE
BENEFITS
OF
A
WATER
QUALITY
CHANGE
The
Objective
of
the
­
analySiS
of
recreation
behavior
has
been
to
develop
~
~
Odel
capable
Of
measuring
the
benefits
associated
with
improving
the
water
for
any
site
that
provides
water­
based
recreation
activities.
Given
~
ualitY
information
on
the
site
characteristics
found
to
be
important
determinants
of
it
is
possible
to
use
the
model
for
each
demand
parameter
to
site
demand
J
estimate
a
"
representative
individual's"
demand
function
for
the
desired
water­
~
ased
faCi
IitY.
Consequently,
this
section
reports
the
results
of
such
an
application
using
information
on
the
13
sites
along
the
Monongahela
River
that
were
used
by
the
contingent
valuation
survey
respondents.

Table
7­
15
provides
a
description
of
these
sites
and
their
attributes.
The
model
estimates
the
representative
individual's
demand
for
each
site's
serviceS.
Because
the
survey
asked
each
respondent
about
his
use
of
the
riverl
including
an
identification
of
the
site
(
or
sites)
used,
it
is
possible
to
develop
an
estimate
of
these
demand
functions
for
each
site.
Moreover,
because
the
model
includes
water
quality
information,
the
change
in
these
demands
can
be
est~
mated
to
accompany
each
Of
the
water
quality
changes
used
In
the
survey
instrument.
In
Chapter
8,
this
information
provides
the
basis
for
a
comparison
Of
direct
and
indirect
methods
for
measuring
the
Table
7­
15.
Recreation
Sites
on
the
Monongahela
River
Identification
MULTI
AREAP/
Site
name
number
SHORMILE
+
ACCESS
AR
EAT
Pittsburgh
area
15
2
1
0.99
The
confluence
of
the
16
2
2
0.99
Youghiogheny
and
Monongahela
Rivers
Elrama
Town
of
Monongahela
Donora
and
Webster
Near
Charleroi
California
and
Brownsville
Maxwell
Lock
and
Dam
Point
Marion
Morgantown
Fairmont
9th
Street
Bridge
Cooper's
Rock
7
2
2
0.99
8
3
4
0.99
9
2
1
0.99
20
21
23
25
26
29
37
44
3
12
2
2
4
3
1
2
4
6
7
1
2
1
1
1
0.96
0.96
0.93
0.99
0.77
0.67
0.99
0.99
SOURCE:
U.
S.
Army
Corps
of
Engineers
Resource
Management
System.

7­
57
A
benefits
from
a
water
quality
improvement.
The
direct
methods
correspond
to
the
results
from
the
survey,
while
the
indirect
methods
use
the
information
on
the
survey
respondents'
recreation
behavior
together
with
the
generalized
travel
cost
model
developed
in
this
chapter.

Seventy­
five
of
the
survey
respondents
were
users
of
1
or
more
of
the
13
sites
along
this
section
of
the
Monongahela
River.
Because
several
individuals
used
more
than
one
site,
there
are
a
total
of
94
observations
identified
as
an
individual/
site
combination.
These
data
provide
the
basis
for
constructing
94
separate
demand
models
to
evaluate
the
implications
of
water
quality
changes
as
measured
by
dissolved
oxygen.
For
example,
the
model
implied
that
the
estimated
price
elasticities
of
demand
(
at
the
average
travel
costs
for
users
in
the
survey
used
for
the
contingent
valuation
experiment)
for
the
13
recreation
sites
along
the
Monongahela
River
­­
the
area
for
the
contingent
valuation
survey­­
ranged
from
­
0.069
to
­
0.075
at
current
water
quality
levels.
Improving
the
water
quality
to
permit
game
fishing
would
imply
a
change
in
DO
from
45
to
64
(
percent
saturation).
These
changes
reduce
the
absolute
magnitude
of
the
estimated
demand
elasticities
to
­
0.052
to
­
0.059.

The
benefits
from
a
water
quality
improvement
are
measured
by
the
increment
to
the
ordinary
consumer
surplus
experienced
by
each
individual.
*
This
increment
can
be
defined
for
each
individual
user
as
follows:

I
p.*
Jk
I
p*
]
k
B.
=
jk
Fjk(
p,
wQ*)
dp
­
Fjk(
pt
WQ)
dp
P.
Jk
P.
Jk
(
7.34)

where:

P.
Jk
=
travel
cost
(
mileage
plus
travel
time)
experienced
by
the
jth
user
to
kth
site
P!=
Jk
maximum
price
the
jth
user
would
be
willing
to
pay
for
the
kth
site's
services
(
i
.
e.
where
the
quantity
demanded
is
zero)

wQ*
.
improved
water
quality
level
WQ
=
initial
water
quality
level
(
i.
e.
,
WQ*
>
WQ)

Fjk(.
)
=
demand
function
for
the
kth
site's
services
by
the
jth
user.

*
The
measurement
of
the
benefits
from
water
quality
improvement
has
ignored
the
potential
for
congestion
effects.
It
has
been
assumed
that
conqest
on
is
negligible
this
assumption,
considered
in
the
[
1983]).
both
befor=
and
after
the
change
in
water
quality.
Without
the
implications
of
management
practices
would
need
to
be
definition
of
the
benefit
measure
(
see
McConnell
and
Sutinen
.

7­
58
Implementation
of
this
benefit
estimator
required
several
amendments.
The
specification
of
the
site
demand
functions
in
semilog
terms
implies
that
they
~
ill
not
have
a
price
intercept.
Rather,
they
asymptotically
approach
the
horizontal
(
price)
axis.
To
estimate
a
finite
consumer
surplus,
a
maximum
for
the
price,
P~
k,
was
selected
to
correspond
to
the
maximum
travel
cost
paid
by
any
of
the
survey
users
of
any
Monongahela
site.
The
specific
value
was
$
22.65
for
a
roundtrip,
including
both
the
mileage
and
time
costs
of
travel
.*
This
modification
implies
the
benefit
estimates
for
the
water
quality
improvement
WIII
be
ABCD
as
given
in
Figure
7­
1,
w
i
t
h
P
j
c
o
r
r
e
s
p
o
n
d
i
n
g
to
the
jth
User's
tra@
costs
and
P*
the
maximum
value
for
the
travel
cost..

Visits/
yr
)

q
p*
Price
$
22.65
{$/
visit)

Figure
7­
1,
Measurement
of
consumer
surplus
increment
due
to
water
quality
improvement
(
WQ
to
WQ*).

Table
7­
16
details
the
dissolved
oxygen
levels
associated
with
each
of
three
use
designations
(
see
Vaughan
in
Mitchell
and
Carson
[
1981])
employed
in
the
calculations
rather
than
the
actual
water
quality
levels
for
the
sites
along
the
river.
The
reason
for
this
approach
follows
from
the
key
project
objective­­
to
compare
benefit
estimates
based
on
the
travel
cost
models
with
those
based
on
the
survey
responses.
All
survey
respondents
were
told
that
the
water
quality
was
consistent
with
boatable
conditions.
Thus,
the
corresponding
value
for
dissolved
oxygen
was
used
as
the
base
value
for
the
estimates
Because
the
model
requires
a
mean
level
of
dissolved
oxygen
for
the
*
This
maximum
travel
cost
is
generally
smaller
than
the
maximum
travel
costs
experienced
by
the
Federal
Estate
Survey
respondents
used
to
estimate
the
generalized
travel
cost
model.
Indeed,
it
is
less
than
the
majority
of
the
sample
means
of
the
FES
travel
costs
(
see
Table
7­
3).

7­
59
,
Table
7­
16.
Dissolved
Oxygen
a
Levels
for
Recreation
Activities
Assumed
level
of
dissolved
Use
designation
oxygen
required
Boatable
water
conditions
45
Fishable
water
conditions
64
Swimmable
water
conditions
83
`
These
estimates
for
dissolved
oxygen
are
based
on
Vaughan
in
Mitchell
and
Carson
[
1981].

summer
recreation
season,
the
means
were
assumed
to
correpond
to
each
of
the
levels
given
in
Table
7­
16.
The
variance
in
monthly
levels
of
dissolved
oxygen
was
set
at
the
sample
mean
for
the
sites
used
to
estimate
the
model­­
8.
187­­
and
was
assumed
to
be
unaffected
by
water
quality
changes.

Table
7­
17
presents
the
mean
values
for
the
incremental
benefits
associated
with
three
types
of
changes
in
water
quality
conditions:

.
An
assumed
deterioration
in
water
quality
making
it
unavailable
for
boating
or
other
recreation
activities.

.
An
improvement
in
water
quality
from
boatable
conditions
to
fishable
conditions.

.
An
improvement
in
water
quality
from
boatable
conditions
to
swimmable
conditions.

All
three
of
these
changes
were
assumed
to
take
place
at
all
13
of
the
Monongahela
sites.
The
first
was
treated
as
the
equivalent
of
losing
the
use
of
the
recreation
site
completely.
The
benefit
loss
was
measured
as
the
consumer
surplus
associated
with
the
site
under
boatable
condition
s­­
PjADP*
in
Figure
7­
1.

The
remaining
two
scenarios
correspond
to
different
levels
of
the
new
demand
functions
for
the
water
quality
associated
with
fishable
and
swimmable
conditions.
Table
7­
17
presents
the
mean
consumer
surplus
increment
for
each
of
the
three
changes
for
our
94
user­
site
combinations.
It
also
reports
the
range
of
values
for
the
increment
to
consumer
surplus.
The
mean
benefits
correspond
to
the
increase
in
an
"
average"
individual's
willingness
to
pay
over
the
recreation
season.
The
average
user
in
the
survey
used
one
or
more
Monongahela
sites
7.22
times.
Thus,
the
loss
of
the
site
completely
translates
to
a
loss
of
$
7.39
per
unit
in
1977
dollars,
or
$
11.46
in
1981
dollars,
the
date
of
the
contingent
valuation
survey.
*

*
This
adjustment
used
the
consumer
price
inae)(
(
CPI
)
for
all
Cornmoditic%.

Using
a
1967
base,
the
1977
CPI
for
all
items
was
181.5.
In
1981
it
closed
the
year
at
281.5.
See
Economic
Report
g
~
President
1982,
Council
of
Economic
Advisors
[
1982].

7­
60
Table
7­
17.
Mean
and
Range
of
Bene~
t
Estimates
for
Water
Quality
Scenarios
b
Minimum
Maximum
Water
quality
change
Mean
value
value
Scenario
(
1)
$
53,35
$
0.00
$
70.80
Loss
of
use
of
site
(
7
.
3
9
)
under
boatable
conditions
Scenario
(
2)
!$
4
.52
$
0,00
$
8.60
Improvement
of
(
0.63)
water
quality
from
boatable
to
fishable
conditions
Scenario
(
3)
$
9.49
$
0.00
$
18.30
Improvement
of
(
1.31)
water
quality
from
boatable
to
swimmable
conditions
a
These
calculations
are
in
1977
dollars,
the
year
of
the
Federal
Estate
Survey.
b
The
numbers
in
parentheses
below
the
overall
increment
report
the
corresponding
consumer
surplus
increment
on
a
per
visit
basis.

Because
these
benefits
estimates
are
available
for
each
of
the
94
user/
site
combinations,
the
estimates
in
several
classifications
were
also
tabulated­­
by
size
of
family
income
reported
by
the
respondents
and
by
the
magnitude
of
their
travel
costs.
The
results
for
the
consumer
surplus
loss
due
to
loss
of
the
use
of
the
river
for
boating
are
given
in
Tables
7­
18
and
7­
19.
The
results
for
each
of
the
two
increments
to
water
quality
compared
with
income
are
given
in
Tables
7­
20
and
7­
21.
It
should
be
noted
that
the
income
levels
are
in
1981
dollars
while
the
consumer
surplus
increment
is
in
1977
dollars.
Scaling
the
latter
by
1.55
will
convert
them
to
equivalent
dollars.
Since
it
was
a
simple
multiple
of
the
estimates
and
would
not
change
the
distributions,
they
were
not
converted
for
these
tables.

These
results
indicate
that
it
is
possible
to
use
a
generalized
form
of
the
travel
cost
model
to
estimate
the
benefits
from
a
water
quality
change.
By
u
s
i
n
g
the
recreation
use
patterns
for
a
number
of
sites,
it
was
possible
to
develop
a
general
model
that,
in
principle,
is
capable
of
being
used
to
estimate
the
recreation
benefits
associated
with
water
quality
changes
at
any
site
Providing
similar
water­
based
recreation
activities.

7­
61
L
Table
7­
18.
Consumer
Surplus
Loss
Due
to
Loss
of
Use
of
the
Monongahela
River
by
Survey
Users'
Income
Consumers
surplus
loss
(
1977
dollars)
a
Income
(
1981
dollars)
0­
10
10­
20
20­
30
30­
40
40­
50
50­
60
60­
70
70­
80
Total
0­
5,000
­
­
­
­
1
2
5,000­
10,000
­­
­­
­­
­­

10,000­
15,000
­­
­­
­­
­­

1
5
,
0
0
0
­
2
0
,
0
0
0
­
­
1
1
2
20,000­
25,000
1
1
1
­
­

25,000­
30,000
1
­
­
2
2
30,000­
35,000
­­
­­
­­
3
35,000­
40,000
­­
­­
­­
­­

40,000­
45,000
­­
­­
­­
1
45,000­
50,000
­­
­­
­­
­­

50,000
and
above
­­
­­
­­
­­

Total
2
2
5
10
­­

­­

­­

­­

1
3
­­

­­

1
­­

­­

5
4
3
­
­

2
7
2
2
6
­
­

3
11
­­

1
1
­­

8
6
­
­

2
2
­
­

2
1
­
­

­.
1
­­

3
1
­
­

2
­
­
­
­

29
39
2
10
11
8
18
6
22
7
3
3
4
2
94
a
To
convert
to
1981
dollars
multiply
the
endpoints
of
the
benefit
scale
by
1.55.

Table
7­
19.
Consumer
Surplus
Loss
Due
to
Loss
of
Use
of
the
Monongahela
River
by
Survey
Users'
Travel
Cost
Travel
cost
(
1
977
Consumer
surplus
loss
(
1977
dollars)

dollars)
o­
1o
10­
20
20­
30
30­
40
40­
50
50­
60
60­
70
70­
80
Total
o­
5
19
39
2
60
5­
1o
­
­
­
2
5
10
­
­
17
10­
15
­
­
4
8
­
12
15­
20
­
2
1
3
20­
25
2
­
­
2
Tots
I
2
2
5
10
5
29
39
2
94
7­
62
7
Table
7­
20.
Consumer
Surplus
Increments
Due
to
Water
Quality
lmprovement­­
Boatable
to
Fishable
by
Survey
Users'
Income
Consumer
surplus
increment
(
1977
dollars)
a
Income
(
lg81
dollars)
o­
1o
10­
20
20­
30
30­
40
Total
()­
5,000
5,000­
10,000
10,000­
15,000
15,000­
20,00()
20,000­
25,000
25,000­
30,000
30,000­
35,000
35,000­
40,000
40,000­
45,000
45,000­
50,000
50,000
and
above
3
7
11
8
10
11
8
18
6
22
7
3
3
4
2
2
2
3
4
3
3
4
16
3
18
3
1
1
2
Total
4
43
26
94
a
To
convert
to
1981
dollars
multiply
the
endpoints
of
the
benefit
scale
by
1.55.

Table
7­
21.
Consumer
Surplus
Increment
Due
to
Water
Quality
lmprovement­­
Boatable
to
Swimmable
by
Survey
Users'
Income
Consumer
surplus
increment
(
1977
dollars)
a
Income
(
1981
dollars)
o­
1o
10"
20
20­
30
30­
40
40­
50
50­
60
60­
70
Total
0­
5,000
5,000­
10,000
10,000­
15,000
15,000­
20,000
20,000­
25,000
25,000­
30,000
30,000­
35,000
35,000­
40,000
40,000­
45,000
45,000­
50,000
50,000
and
above
1
1
228
8
69
10
11
8
18
6
22
7
3
3
4
2
3
1
18
4
1
2
33
3
3
62
1
1
4
2
Total
3
5
15
27
9
20
15
94
a
To
convert
to
1981
dollars,
multiply
the
endpoints
of
the
benefit
scale
by
1.55.

7­
63
7.9
SUMMARY
The
findings
from
the
application
of
the
travel
cost
approach
are
of
equal,
if
not
greater,
importance.
The
research
in
this
project
developed
a
generalized
travel
cost
model
that
predicts
the
recreation
benefits
of
wate
r
quality
improvements
at
a
recreation
site.
Estimating
the
benefits
for
users
of
the
Monongahela
River,
the
travel
cost
model
predicted
benefits
of
$
83
per
year
for
a
user
if
a
decrease
in
water
quality
is
avoided.
Water
quality
improvements
to
swimmable
water
in
the
Monongahela
were
estimated
at
$
1s
per
year
(
in
1981
dollars).

Two
features
of
the
generalized
travel
cost
model
are
of
particular
importance
The
model
can
be
applied
to
predict
the
value
of
water
quality
improvements
for
a
substantial
range
of
sites,
and
it
is
especially
relevant
for
a
large
number
of
water
quality
standards
applications.
Including
the
effect
of
key
site
features
in
addition
to
water
quality
­­
like
access
and
facilities­­
and
relying
on
data
frequently
available
in
the
public
domain
makes
the
model
a
viable
tool
for
future
benefits
applications.

7­
64
CHAPTER
8
A
COMPARISON
OF
THE
ALTERNATIVE
APPROACHES
FOR
ESTIMATING
RECREATION
AND
RELATED
BENEFITS
8.1
INTRODUCTION
One
of
the
primary
objectives
of
this
research
has
been
to
compare
available
methods
for
measuring
benefits
of
water
quality
improvement.
Of
course,
the
"
true"
value
of
benefits
associated
with
a
specific
increment
of
water
quality
can
never
be
known,
and
a
comparison
of
measurement
methods
cannot
be
interpreted
as
a
validation
of
any
one
of
them.
Nonetheless,
it
is
important
to
recognize
that
contingent
valuation
methods
for
estimating
the
benefits
of
environmental
quality
improvements
are
viewed
with
considerable
skepticism
by
many
(
if
not
most)
economists.
Presumably,
these
economists
assume
that
individuals
will
experience
difficulty
in
responding
to
valuation
questions
for
nonpriced
goods
and
that
their
responses
will
exhibit
significant
strategic
bias.
By
contrast,
indirect
methods
have
been
more
favorably
regarded
by
most
economists,
and
this
study's
use
of
benefit
estimates
derived
from
one
indirect
method­­
the
travel
cost
recreation
demand
model­­
as
a
benchmark
for
the
contingent
valuation
estimates
ref
Iects
this
perspective.
Of
course,
it
should
be
recognized
that
indirect
and
direct
benefits
measurement
approaches
can
be
distinguished
according
to
the
assumptions
each
makes
and
that
a
comparison
of
them
reflects
in
part
the
plausibility
of
their
assumptions
as
descriptions
of
real­
world
behavior
and
constraints.

To
aid
in
the
interpretation
of
the
comparisons
of
benefit
estimation
approaches
this
chapter
highlights
the
specific
features
of
the
approaches
and
how
they
are
applied
in
this
study.
The
Monongahela
River
case
study
provides
the
basis
for
the
evaluation
of
the
approaches.
The
types
of
possible
evaluations
are
bounded
by
its
scope.
More
specifically,
Section
8.2
of
this
chapter
introduces
the
conceptual
basis
for
a
comparative
evaluation
of
benefit
estimation
approaches.
Following
this
discussion,
Section
8.2
also
relates
the
evaluation
scheme
used
in
this
chapter
to
that
used
in
earlier
comparisons,
including
those
of
Knetsch
and
Davis
[
1966],
Bishop
and
Heberletn
[
1979]
,
and
Brookshire
et
al.
[
1982].
Section
8.3
discusses
the
results
of
the
comparison
of
approaches,
including
the
findings
of
a
numerical
comparison
of
the
mean
estimates
of
the
user
and
intrinsic
components
of
benefits
for
specific
water
quality
changes
by
methodology.
This
discussion
is
followed
by
pairwise
comparisons
of
the
contingent
valuation
and
travel
cost
methods
and
of
the
contingent
valuation
and
contingent
ranking
methods,
Finally,
Section
8.4
summarizes
the
findings
and
discusses
their
implications
for
the
practical
use
of
benefit
measurement
approaches.

8­
1
8.2
THE
CONCEPTUAL
FRAMEWORK
FOR
A
COMPAR
BENEFIT
ESTIMATION
APPROACHES
8.2.1
Background
SON
OF
RECREATION
Improvements
in
water
quality
associated
with
water
bodies
that
support
recreation
activities
can
lead
to
both
user
and
intrinsic
nonuse
benefits.
User
benefits
arise
because
water
quality
can
be
expected
to
affect
the
types
of
recreation
activities
at
the
site
experiencing
the
changes.
Individuals
who
wish
to
participate
in
activities
made
possible
by
the
improvement
will
be
able
to,
thus
enhancing
their
levels
of
economic
well­
being.
User
benefit
estimates
of
water
quality
improvements
attempt
to
measure
the
magnitude
of
these
changes
in
well­
being.
Intrinsic
benefits,
on
the
other
hand,
arise
either
because
individuals
are
uncertain
of
their
potential
use
of
a
site
or
because
they
experience
enhanced
utility
merely
from
knowing
of
improved
site
conditions.
The
first
recognition
of
the
importance
of
intrinsic
benefits
has
most
often
been
associated
with
Krutiila's
[
1967]
discussion
of
the
rationale
for
public
involvement
in
the
management
of
natural
environments.
Intrinsic
benefits
have
been
identified
under
a
variety
of
classification
schemes
to
include
option
and
existence
values.

Because
preceding
chapters
have
presented
detailed
discussions
of
both
user
and
intrinsic
benefits,
the
definitions
of
each
are
not
repeated
here.
Rather,
this
chapter
considers
the
relationship
between
benefit
estimation
approaches
and
the
two
benefit
classes.
This
relationship
is
important
because
it
affects
the
types
of
comparisons
that
can
be
undertaken
across
approaches.

The
measurement
of
the
economic
benefits
of
water
quality
improvement
requires
a
mechanism
for
linking
the
water
quality
change
to
a
consistent
measure
of
benefits.
As
noted
in
Chapter
2,
this
linkage
provides
one
basis
for
classifying
methods
used
to
measure
benefits
of
a
change
in
any
environmental
amenity
not
exchanged
in
an
organized
market.
While
Chapter
2
identifies
several
types
of
assumptions
that
provide
these
links,
two
classes
of
assumptions
are
especially
relevant
to
the
approaches
considered
in
this
project
for
benefit
measurement.

The
first
relevant
class
of
assumptions
involves
the
use
of
the
technical
association
between
water
quality
and
recreation
site
services.
Use
of
a
water
body's
recreation
services
involves
a
corresponding
(
and,
indeed,
simultaneous)
use
of
the
water
quality
at
the
site.
Thus,
the
types
of
activities
that
can
be
undertaken
at
a
particular
site
are
affected
by
the
site's
water
quality
(
a
point
explicitly
made
throughout
the
analysis
in
Chapters
4
through
7).
Given
both
a
behavioral
model
to
describe
how
individuals
allocate
their
resources
and
exogenous
measures
of
their
use
of
recreation
sites
with
differing
levels
of
water
quality,
this
approach
maintains
that
it
may
be
possible
to
estimate
individuals'
willingness
to
pay
for
water
quality
indirectly.
This
recognition
is,
of
courso,
the
basis
for
the
approach
used
in
the
generalized
travel
cost
model
developod
in
Chapter
7.*
However
,
more
important
for
comparing
measurement
approaches
*
This
model
assumes
that
each
set
of
users
for
each
of
the
sites
included
in
our
sample
from
the
Federal
Estate
Survey
acts
as
the
"
representative"
individual
would
under
the
circumstances
defined
by
the
site's
availability
l
 
n
d
the
survey
respondent's
economic
characteristics.

.

8
­
2
.
.
.­
­.­
 
.
.__.
_­_.
__..
_..
.
 
.
.
______________
this
approach­­
using
"
indirect"
technical
linkages
between
water
quality
is
that
~
ecreatiOn
site
services
­­
only
measures
user
values.
and
The
second.
relevant
class
of
assumptions,
identified
in
Chapter
2
as
insti
­
assumption
S,
explicitly
recognizes
that
ideal
markets
would
provide
the
~
Utional
benefit
measures
required
for
any
good
or
service,
providing
the
good
could
be
exchanged
i
n
t
h
e
m
.
However,
attempts
to
estimate
the
valuation
of
such
amenities
as
water
quality
face
difficulties
because
ideal
markets
~
nvironmental
available.
Thus,
the
contingent
valuation
approach­­
using
"
direct"
are
not
­­
assumes
that,
if
individuals
are
confronted
with
a
hypo­
institutional
linkages
thetical
market
(
in
the
form
a
survey
questionnaire)
for
these
amenities,
their
~
e~
ponses
will
measure
their
true
valuation
of
the
resources
(
or
amenities)

involved
Thus,
the
contingent
valuation
approach
assumes
it
is
possible
to
mimic
the
outcomes
of
ideal
markets
by
completely
describing
the
conditions
of
exchange
in
a
hypothetical
market
for
the
service
to
be
valued.
As
a
result,

these
methods
assume
`
hat
an.
individual's
resPonses
to
the
conditions
presented
in
this
hypothetical
market
WIli
be
equivalent
to
the
actual
responses
that
would
be
made
if.
the
exchanges
took
place
in
actual
markets.
Since
the
market
i
s
simply
an
Institution
,
a
hypothetical
market
can
be
defined
to
suit
any
particular
nonmarketed
service
and
does
not
require
that
it
actually
be
feasible
to
exchange
the
services
described.
Thus,
contingent
valuation
methods
can
tneasure
both
user
and
intrinsic
benefits.

in
comparing
the
two
classes
of
assumptions
and
the
approaches
for
benefit
measurement
arising
from
them,
it
is
important
not
to
confuse
the
flexibility
of
the
approaches
using
institutional
restrictions
with
judgments
that
these
approaches
require
less
stringent
assumption
s.*
Alternative
approaches
require
different
assumptions.
Therefore,
appraisals
of
the
severity
of
one
approach's
assumptions
relative
to
another's
should
be
regarded
as
individual
judgments,
not
necessarily
as
objective
comparisons.

8.2.2
Research
Design
and
Comparative
Analysis
The
research
design
of
this
project
permits
several
types
of
comparisons.
Chapters
reporting
each
approach's
estimates
have
discussed
the
first
type­­
those
within
a
benefit
estimation
framework.
For
example,
the
contingent
valuation
survey
was
designed
to
consider
five
different
approaches
for
eliciting
an
individual's
valuation
of
water
quality
changes.
In
four
of
these
approaches,
only
the
valuation
question
differed:

.
A
direction
question
.
A
question
using
a
payment
card
*
The
classification
scheme
for
benefit
estimation
methods
given
by
Schulze,
d'Arge,
and
Brookshire
[
1981],
pp.
154­
155,
is
somewhat
misleading
in
that
it
implies
the
contingent
valuation
approach
has
the
least
a
priori
assumptions.
While
this
is
true
as
a
description
of
the
assumptions
con~
erning
constraints
to
actual
behavior,
it
ignores
the
implicit
assumption
that
responses
to
hypothetical
institutions
will
provide
a
good
guide
to
the
responses
made
to
the
actual
institutional
arrangements
implied
by
their
"
constructed"
markets.

8
­
3
.
The
conventional
iterative
bidding
framework
with
a
$
25
starting
point
l
The
conventional
iterative
bidding
framework
with
a
$
12S
starting
point.

Each
questioning
format
was
applied
to
an
approximately
equal
proportion
of
the
sample
and
provides
independent
estimates
of
an
individual's
valuation
of
the
specified
water
quality
changes.
Because
the
design
of
the
questions
elicited
the
individual's
option
price
and
user
values,
comparisons
of
these
questioning
formats
were
undertaken
for
the
estimates
of
option
price,
user
value,
and
option
value
with
the
results
described
in
detail
in
Chapters
4
and
5.

This
chapter
focuses
on
comparisons
between
benefit
estimates
across
methodologies
­­
e.
g.
,
travel
cost
vs.
contingent
valuation.
These
comparisons
will
also
involve
the
effect
of
question
format,
but
the
effect
of
format
may
differ
from
the
within
methodologies
comparison
because
the
standards
for
the
comparisons
are
different.
Equally
important,
the
comparisons
across
methods
cannot
consider
each
method's
performance
in
measuring
combined
user
plus
intrinsic
benefit
(
i.
e.
,
option
price)
as
well
as
their
separate
estimates
(
e.
g.,
in
form
of
option
value).
The
travel
cost
method
measures
only
user
value,
and
the
contingent
ranking
only
a
composite
of
the
two.

The
specific
details
of
the
within
method
comparison
involved
two
types
of
evaluations:

l
.
The
Statistical
tests
for
the
differences
in
means
between
all
pairs
of
question
formats
for
the
full
sample
and
for
users
and
nonusers
of
the
Monongahela
River.

Multivariate
regression
analysis,
including
dummy
variables
for
the
question
formats
along
with
other
prospective
determinants
of
the
relevant
dependent
variables.

option
price
results
exhibit
the
most
differences
among
question
formats
with
some
evidence
of
a
starting
point
bias.
The
regression
models
also
exhibit
the
most
cases
of
significant
effects
for
the
question
format
variables
in
this
case.
This
finding
contrasts
with
several
(
but
not
all
)
of
the
past
contingent
valuation
studies.
*
With
the
option
value
estimates
there
is
also
some
evidence
of
starting
point
bias,
but
these
findings
are
not
as
pronounced
as
in
the
analysis
of
the
option
price
estimates.
These
differences
are
not
necessarily
surprising
since
only
the
first
stage
of
the
individual's
response
(
i
.
e.
,
the
option
price)
had
distinct
questioning
formats.
Thereafter,
the
questions
calling
for
separation
of
the
option
price
into
components
(
i.
e.,
user
values)
were
(
by
practical
necessity)
direct
questions.

*
The
Schulze,
d'Arge,
and
Brookshire
[
1981]
summary
concludes,
based
on
an
analysis
of
several
contingent
valuation
experiments,
that
starting
point
bias
is
not
a
serious
problem.
Our
results
do
not
conform
to
this
conclusion
and
indicate
that
the
prescreening
of
data
used
to
eliminate
inconsistent
observations
may
affect
their
conclusions.
Of
course,
it
should
also
be
emphasized
that
our
results
relate
only
to
a
single
experiment.

8
­
4
Finally,
the
reSUltS
are
qUite
SenSitiVe
­
to
the
screening
of
observations
to
be
refusal
to
Partl~
l
Pate
in,
or
Inconsistent
with,
the
contingent
judged
framework.
As
noted
in
Chapter
4,
while
procedures
used
to
identify
valuation
observations
are
based
on
a
statistical
index
of
the
influence
of
each
these
observation
(
and
are
therefore
capable
of
replication),
the
effects
individual
socioeconomic
characteristics
of
the
survey
respondents
cannot
be
of
specific
distinguished
from
the
question
format
(
see
Table
4­
8
in
Chapter
4).
Thus,
for
starting
point
bias
and
for
other
pairs
of
question
formats
the
result
s
(
iterative
bidding
with
$
125
starting
point)
would
have
been
more
pronounced
with
the
inclusion
of
the
observations
judged
to
be
inconsistent
with
the
contingent
val
UatiOn
framework.

Comparisons
across
approaches
are
limited
because
the
methods
do
not
uniformly
measure
the
same
components
of
the
benefits
associated
with
a
water
quality
improvement.
As
we
noted
earlier,
the
contingent
valuation
method
design
measures
both
user
and
intrinsic
benefits
and
permits
these
estimates
to
be
separated.
By
contrast,
the
travel
cost
and
contingent
ranking
meth
­
Ocjs
are
more
limited.
The
travel
cost
approach
measures
only
user
values
(
i.
e.
ordinary
consumer
surplus).
The
contingent
ranking
design
measures
Option
price
but
does
not
divide
the
estimates
into
the
user
value
and
option
Value.
Therefore,
comparisons
here
are
limited
to
examining
the
relationship
between
the
user
value
estimates
of
the
contingent
valuation
and
travel
cost
approaches
and
the
option
price
estimates
for
contingent
valuation
and
cont
ngent
ran
king.*

The
comparison
of
gent
valuation
approach
plus
estimates
derived
interesting
comparison.
shire
et
al.
[
1982]
for
the
estimated
user
values
derived
using
the
contin
­
(
with
all
four
question
formats)
and
the
consumer
surfrom
the
generalized
travel
cost
model
is
the
most
It
provides
an
extension
to
the
recent
work
of
Brookthe
valuation
of
air
quality
using
hedonic
property
value
and
contingent
valuation
methods.

Using
a
subset
of
the
survey
respondents
who
visited
specific
Mononga
­
hela
River
sites
to
derive
consumer
surplus
estimates
from
the
generalized
travel
cost
model
(
presented
in
Chapter
7)
allowed
a
matching
of
each
respondent's
expressed
user
value
for
a
comparable
water
quality
change
with
the
values
predicted
from
the
travel
cost
model.
This
comparison
of
the
travel
cost
and
contingent
valuation
methods
can
be
made
for
each
user
in
this
survey,
in
contrast
to
the
Brookshire
et
al.
[
1
9
8
2
]
analysis.
~
Thus,
both
the
mean
estimates
derived
from
the
two
approaches
and
the
association
in
the
estimates
can
be
compared
across
individual
users.

*
For
the
sake
of
simplicity
in
the
use
of
terms
in
this
chapter
contingent
valuation
is
used
to
refer
to
the
four
question
formats
in
the
contingent
valuation
survey.
While
contingent
ranking
is
a
subset
of
contingent
valuation
(
and
this
distinction
was
made
in
Chapter
1),
the
easier
terminology
of
contingent
ranking
vs.
contingent
valuation
is
used
in
this
comparison
chapter.

l'This
is
one
of
the
aspects
of
our
extension
over
this
work.
A
second
involves
replacing
the
broad
bounds
for
contingent
valuation
estimates
with
a
potentially
more
restrictive
upper
threshold.

8­
5
Several
features
limit
the
ability
to
compare
the
estimates
derived
from
the
travel
cost
and
contingent
valuation
methods.
The
simplest
of
these
features
is
different
dollar
values
in
each
method
because
the
travel
cost
model
was
developed
with
1977
dollars
and
the
contingent
valuation
estimates
was
developed
with
1981
dollars.
Using
the
consumer
price
index,
an
adjustment
can
approximately
account
for
this
difference.
A
more
important
reason
for
differences
stems
from
what
is
being
measured.
The
user
values
derived
using
contingent
valuation
methods
estimate
an
individual's
expected
willingness
to
pay
or
compensating
surplus
(
for
improvements
in
water
quality),
while
the
generalized
travel
cost
model
estimates
ordinary
consumer
surplus.
A
long
literature
on
the
theoretical
foundations
of
consumer
surplus
estimates
has
suggested
that
there
are
good
reasons
why
these
two
measures
should
di
­
verge.
*
However,
for
price
(
Willig
[
1976])
and
quantity
(
Randall
and
Stoll
[
1980]
)
changes,
the
difference
between
the
two
measures
can
be
bounded
under
specific
conditions
(
see
Chapter
2
for
a
brief
review).

At
first,
the
comparison
of
welfare
measures
in
this
project
might
seem
to
involve
a
case
that
falls
outside
the
scope
of
the
bounds,
because
it
involves
a
change
in
water
quality
rather
than
a
price
or
quantity
change.
Fortunately,
this
conclusion
is
premature.
One
of
the
assumptions
used
to
develop
the
generalized
travel
cost
model
­­
that
water
quality
augments
the
effect
of
a
recreation
site's
services
in
the
production
of
recreation
activities
(
see
Equations
(
7.11)
and
(
7.12)
in
Chapter
7)­­
implies
that
a
water
quality
change
can
be
translated
into
an
equivalent
change
in
either
the
quantity
of
a
site's
services
or
in
the
"
effective"
price
of
using
the
site
(
see
Equations
(
7.12)
and
(
7.13),
respectively).
~
Therefore,
for
changes
in
water
quality
*
See
Just,
Hueth,
and
Schmitz
[
1982]
for
further
discussion.

_#'
in
general
terms
the
consumer
surplus
increment
due
to
ity
change
r
w
,
with
a
demand
function
Q
=
F(
P,
w)
(
P
=
price,
is
given
as
I
p
*

[
P*
Cs
i
=
F(
pi,
w2)
dp
­
F(
pi,
wl)
dp
,
Pi
i
where
a
water
qual
­
Q
=
quantity)

Cs
i
=
consumer
surplus
for
individual
facing
price
Pi
p*
.
price
at
which
the
quantity
demanded
would
be
zero
W2
=
improved
level
of
water
quality
WI
=
existing
level
of
water
quality.

The
form
of
the
household
production
technology
assumed
in
the
development
of
our
travel
cost
sidered
equivalent
ices.
This
implies
lent
to
some
change
model
implies
that
a
change
in
water
quality
can
be
conto
a
change
in
the
quantity
of
or
price
of
a
site's
serv
­
that
the
change
from
WI
to
W2
can
be
treated
as
equivain
the
price
of
a
site's
services
from
P(
wl)
to
P(
w2).

8
­
6
that
translate
into
relatively
small
pr~
Ce
(
quantity)
changes,
the
Willig
(
Randall
and
Stoll)
bounds
can
be
applled
to
judge
the
relationship
between
Marsha
llian
consumer
surPlus
and
the
willingness
to
pay
for
a
water
quality
change.

From
a
practical
perspective,
one
might
assume
that
the
discrepancies
between
the
Marshal
lian
consumer
surplus
and
the
willingness
to
pay
for
a
water
quality
change
associated
with
recreation
water
sites
would
be
small.
Most
households:
expenditures
on
water­
based
recreation
activities
would
be
a
very
small
fraction
of
their
income.
This
judgment
is
also
supported
by
the
estimated
tra~
el
cost
demands
developed
for
this
study
in
that
they
imply
income
is
not
a
significant
determinant
of
the
demand
for
the
services
of
water­
based
sites
comparable
to
sites
on
the
Monongahela
River.
Thus,
the
difference
between
the
willingness
to
pay
and
the
consumer
surplus
for
a
comparable
change
in
water
quality
can
be
expected
to
be
less
than
5
percent.
*
The
evidence
necessary
for
judging
the
implications
of
income
for
survey
respondents
who
were
users
of
the
Monongahela
River
can
be
derived
using
the
same
type
of
information
required
by
the
travel
cost
model.
~
That
is,
because
individual
estimates
of
the
ordinary
consumer
surplus
require
travel
cost
and
income
information
these
variables
were
combined
with
the
respondents'
reported
use
patterns
for
the
Monongahela
sites,
thus
treating
all
13
sites
as
if
they
shared
a
common
demand
function
,
even
though
the
generalized
travel
cost
model
does
not
require
this
assumption.
These
data
permit
the
estimation
of
a
travel
cost
model
for
the
Monongahela
in
its
current
state.
The
results
are
given
in
Equation
(
8.1
)
below:

In
V
=
0.7983
­
0.0195
(
T+
M)
cost
+
0.000015
income
(
3
.
1
5
3
)
(
­
0
.
7
8
5
)
(
8
.
1
)
(
1
.636)

R
2
=
0
.
0
3
2
The
numbers
in
parentheses
are
the
t­
ratios
for
the
null
hypothesis
of
no
association.
These
results
indicate
that
income
is
not
a
significant
determinant
of
user
trips
to
the
Monongahela
sites.
Therefore,
these
findings
would
be
consistent
with
judgments
based
on
the
generalized
travel
cost
model,
and
willingness
to
pay
would
be
expected
to
be
less
than
the
Marsh
allian
consumer
surplus
for
water
quality
improvements
(
the
equivalent,
in
the
generalized
*
See
Freeman
[
1979a]
or
Just,
Hueth,
and
Schmitz
[
1982]
for
a
complete
discussion
of
the
implications
of
the
Willig
[
1976]
bounds
for
applied
benefit
analysis.

l'The
generalized
travel
cost
model
assumes
that
a
water
quality
change
can
be
translated
into
either
an
equivalent
price
or
quantity
change.
Thus,
the
site
demand
equation
is
the
relevant
basis
for
judging
income
responsiveness
Survey
responses
for
compensating
surplus
(
referred
to
as
user
value
in
Chapter
5)
are
expected
to
provide
equivalent
results
if
these
two
sets
of
information
provide
consistent
descriptions
of
the
individuals'
demand
characteristics
An
examination
of
the
role
of
income
in
the
user
value
equations
confirms
this
q
priori
expectation.
The
coefficients
estimated
for
income
are
never
judged
to
be
statistically
significant
determinants
of
user
values.

8­
7
travel
cost
model,
of
price
decreases
for
the
site's
services
or
quantity
increases
).
However,
these
estimates
of
income
effects
imply
that
the
difference
between
willingness
to
pay
and
consumer
surplus
should
be
small.

These
results
can
also
be
compared
with
the
predicted
demands
for
each
of
the
13
Monongahela
sites
based
on
the
generalized
travel
cost
model
and
the
characteristics
of
each
of
these
sites.
Of
course,
this
comparison
cannot
be
treated
as
an
evaluation.
The
estimates
given
in
Equation
(
8.1)
are
pooled
across
sites
and
assume
the
demand
parameters
are
invariant
with
respect
to
site
attributes.
Nonetheless,
the
comparison
may
serve
to
identify
whether
the
implied
demand
features
are
completely
incompatible
with
these
crude
estimates
available
as
a
byproduct
of
the
survey
data.
The
focus
is
on
the
parameters
of
greatest
influence
for
estimates
of
consumer
surplus
change
in
response
to
a
water
quality
change.
Table
8­
1
reports
these
predicted
parameters
for
the
intercept
and
coefficient
of
(
T+
M)
cost
for
each
of
the
13
sites
under
the
assumption
of
boatable
water
quality.
The
absolute
magnitude
of
the
price
coefficient
is
in
all
cases
smaller
than
any
estimates
based
on
the
survey,
but
they
are
reasonably
close
to
the
survey
estimates.
The
intercept
predictions
are
substantially
larger
than
the
survey
estimates.

The
second
comparison
across
benefit
methodologies
involves
the
contingent
valuation
and
contingent
ranking
approaches.
Because
all
survey
respondents
were
asked
one
of
the
four
types
of
contingent
valuation
questions
and
the
contingent
ranking,
the
estimates
from
these
approaches
are
not
independent
estimates
of
the
option
prices
for
water
quality
changes.
Indeed,
it
is
possible
that
an
individual's
responses
to
the
contingent
valuation
questions,
which
preceded
the
ranking
questions
on
the
survey
instrument,
influenced
the
rankings.
Therefore,
this
comparison
ref
Iects
both
the
effects
of
the
methods
used
to
estimate
benefits
and
an
individual's
consistency
in
responding
to
comparable
water
quality
increments
in
different
formats.

Table
8­
1.
Predicted
Demand
Parameters
for
Monongahela
Sites
Coefficient
for
Site
Intercept
T+
M
cost
Pittsburgh
area
Confluence
of
the
Youghiogheny
and
Monongahela
Rivers
Elrama
Town
of
Monongahela
Donora
and
Webster
Near
Charleroi
California
and
Brownsville
Maxwell
Lock
and
Dam
Point
Marion
Morgantown
Fairmont
9th
Street
Bridge
Cooper's
Rock
1.323
1.317
1.317
1.306
1.323
1.317
1.308
1.311
1.323
1.404
1.449
1.323
1.323
­
0.0133
­
0.0132
­
0.0132
­
0.0131
­
0.0133
­
0.0132
­
0.0131
­
0.0130
­
0.0733
­
0.0140
­
0.0144
­
0.0133
­
0.0133
8­
8
arisons
of
Benefit
Estimation
Methods
8.2.3
Past
Comp
comparisons
`
f
`
he
results
of
benefit
estimation
methodologies
within
the
of
a
common
problem
have
been
quite
limited.
The
first
such
com­
~
ontext
~
arison
was
undertaken
by
Knetsch
and
Davis
[
1966]
and
involved
a
bidding
game
version
of
what
is
now
commonl
Y
referred
to
as
contingent
valuation
and
of
the
travel
cost
model.
The
survey
was
based
on
a
sample
of
185
a
fOrm
Users
of
a
forest
recreation
area
in
northern
Maine.
With
the
iterative
bidding
game,
respondents
were
asked
their
willingness
to
pay
(
as
increased
cost
to
visit
the
area).
A
similar
format
was
used
to
elicit
willingness
to
drive
to
the
area.
Individuals
were
also
asked
the
actual
distance
they
traveled
to
the
site.

Knetsch
and
Davis
compared
three
approaches
for
estimating
the
aggregate
benefits
from
the
site.
The
first
used
a
willingness­
to­
pay
equation
based
on
the
survey
results
to
estimate
a
willingness­
to­
pay
schedule
for
the
user
population
in
the
area
surrounding
the
site.
The
two
sets
of
distance
measures
were
each
valued
at
$.
05
per
mile
and
used
to
derive
aggregate
schedules
for
the
user
population.
The
aggregate
benefit
estimates
derived
for
each
approach
provided
the
basis
for
comparing
the
methods:.

Contingent
valuation
$
71,461
Willingness
to
drive
$
63,690
Travel
cost
$
69,450
Because
the
contingent
valuation
approach
measures
willingness
to
pay
and
travel
cost
measures
the
ordinary
consumer
surplus,
the
latter
would
be
expected
to
exceed
the
former
at
an
individual
level,
However,
it
is
difficult
to
gauge
the
expected
nature
of
the
differences
between
the
two
methods
for
these
calculations
because
they
involve
the
aggregate
schedule
over
all
individuals
and
relate
to
changes
in
the
price
of
the
site
comparable
to
a
loss
of
its
availability
for
this
population.
As
Bockstael
and
McConnell
[
1980]
observe,
the
Willig
bounds
may
not
hold
where
the
analysis
involves
the
removal
of
the
site.
They
observed
that:

it
is
difficult
to
find
single
valued
functions,
x
=
f(
p,
m)
[
where
x
=
quantity
demanded,
p
=
price
and
m
=
income],
decreasing
in
p
and
increasing
in
m,
such
that:

ma)
J
1.
~
m
is
finite
for
all
values
of
p
and
x
2.
the
function
creases
in
p
evaluted
as
p
f(
p,
m)
must
tend
to
that
the
integral
of
goes
to
infinity.
(
p.
zero
rapidly
enough
with
inf
p,
m)
will
be
hounded
when
61)

Because
Knetch
and
Davis
do
not
present
demand
equation
estimates
with
their
travel
cost
findings,
it
is
difficult
to
evaluate
the
relationship
between
their
willingness
to
pay
and
consumer
surplus
estimates
on
an
individual
basis.
Their
benefit
estimates
based
on
the
willingness­
to­
travel
responses
are
diffi
­

8­
9
Table
8­
2.
Bishop­
Heberlein
Comparative
Results
for
Benefit
Approaches
a
Average
benefit
estimate
Method
per
permit
1.
Actual
case
offers
$
6
3
I
l
.
Hypothetical
responses
(
a)
willingness
to
sell
$
101
(
b)
willingness
to
pay
$
2
1
I
l
l
.
Travel
cost
ordinary
consumer
$
11
to
$
45
surplus
(
variation
associated
with
valuation
of
travel
time
from
O
to
%
median
income
rate)

a
These
estimates
are
taken
from
Table
1
in
Bishop
and
Heberlein
[
1979],
p.
929.

cult
to
interpret
within
the
conventional
welfare
economics
framework
and
thus
cannot
be
directly
associated
with
either
of
the
other
benefit
estimates.
Thus,
while
this
study
offered
the
first
evaluation
of
benefit
estimation
approaches,
it
did
not
permit
a
detailed
comparative
analysis
of
them.

The
second
comparative
analysis
was
conducted
by
Bishop
and
Heberlein
[
1979]
and
was
primarily
intended
to
evaluate
the
relationship
between
hypothetical
and
actual
responses
to
willingness­
to­
sell
questions.
*
Their
analysis
was
conducted
using
goose
hunting
permits
for
hunters
in
Wisconsin.
Three
samples
of
hunters
were
used
in
their
analysis.
The
first
sample
received
actual
cash
offers
for
their
permits
(
ranging
from
$
1
to
$
200);
a
second
sample
received
questionnaires
asking
the
individual's
willingness
to
pay
for
(
and
willingness
to
sell)
their
permits;
and
a
third
sample
received
questionnaires
designed
to
permit
the
estimation
of
a
traval
cost
demand
equation.
Table
8­
2
summarizes
the
Bishop
and
Heberlein
estimates
per
permit
for
each
of
the
ap­

*
Bishop
and
Heberlein
describe
a
number
of
potential
biases
that
might
distinguish
hypothetical
and
actual
responses
to
willingness­
to­
pay
questions.
Some
of
these
problems
conform
to
the
definitions
used
in
the
papers
reporting
contingent
valuation
survey
results.
The
most
directly
comparable
case
is
strategic
bias.
However,
the
Bishop­
Heberlein
approach
does
not
attempt
W
induce
differential
responses
from
individuals,
by
giving
them,
for
examPle~
different
information
about
the
uses
that
will
be
made
of
their
bids
to
hYW
thetical
changes.
This
approach
has
been
the
most
common
method
for
i
nvcstl
­

gating
the
potential
for
strategic
bias
in
the
contingent
valuation
experiM@
(
s
e
e
Schulze,
d'Arge,
a
n
d
Brookshire
[
1981]).
R
a
t
h
e
r
,
t
h
e
i
r
compari
­
of
actual
and
hypothetical
responses
will
reflect
a
composite
of
any
such
bias
due
to
the
"
framing"
of
their
hypothetical
survey
instrument
and
to
the
distinction
between
hypothetical
and
real
conditions.

.

8­
10
considered.
Their
findings
suggest
that
hypothetical
willingness­
to­
~
roaches
sell
estimates
overstate
actual
responses.
Moreover,
Bishop
and
Heberlein
~
rgue
that
hypothetical
willingness
to
PaY
and
ordinary
consumer
surplus
esti
­
~
ith
t
h
e
travel
cost
demand
model
understate
the
actual
willingness­
to­
~
ated
the
Willig
bounds
would
imply.
sell
by
more
than
The
Bishop
and
Heberlein
results,
w
h
i
l
e
l
i
m
i
t
e
d
to
a
single
experiment,
important
implications
for
the
relationship
between
hypotheti
­
~
ave
potential
IY
Cal
and
aCtual
estimates
of
willingness
to
sell.
They
do
not
offer
as
much
on
the
comparative
properties
of
the
benefit
estimation
methodologies
guidance
The
authors'
benefit
estimates
made
with
the
travel
cost
model
~
he~
sel
Ves.
can
be
interpreted
(
for
one
Value
fOr
the
opportunity
cost
of
travel
time)
as
quite
close
to
the
hypothetical
willingness
to
pay.
However,
because
the
~
election
of
an
opportunity
cost
for
travel
time
is
treated
as
judgmental,
more
~
pecifiC
conclusions
are
not
possible.
Finally,
the
Bishop
­
Heberlein
research
design
(
i.
e.
,
the
selection
of
independent
samples
for
the
hypothetical
and
travel
cost
su~
veys)
did
not
permit
comparison
of
the
hypothetical
willingness
to
pay
and
ordinary
consumer
surplus
on
an
individual
basis.

Most
recently,
Brookshire
et
al.
[
1982]
provided
comparative
analysis
of
benefit
estimation
methods,
maintaining
it
as
a
validation
analysis
of
the
contingent
valuation
methodology.
As
observed
earlier,
this
reflects
the
interpretation
given
to
contingent
valuation
versus
indirect
benefit
estimation
methods
by
many
economists
and
is
somewhat
unfortunate.
Each
of
the
methods
involved
in
the
Brookshire
et
al.
[
1982]
comparative
evaluation
is
based
on
different
assumptions
concerning
the
economic
behavior
of
households
and
the
role
of
environmental
amenities
(
i.
e.
,
air
quality)
in
their
decisionmaking.
Neither
method
provides
the
"
true"
benefit
estimates
for
air
quality
improvements

The
Brookshire
et
al.
[
1982]
analysis
compares
a
hedonic
property
value
model
to
a
contingent
valuation
approach
for
measuring
the
willingness
to
pay
for
reductions
in
air
pollution.
The
authors
interpret
the
hedonic
model
as
providing
an
upper
bound
for
willingness
to
pay
and
argue
that
the
assumptions
of
the
model
are
approximately
satisfied
for
the
Los
Angeles
area.
At
issue
in
their
comparison,
however,
is
whether
direct
questions
can
be
believed
They
demonstrate
if
each
method
conforms
to
its
respective
assumptions
the
annual
rent
differential
for
pollution
should
exceed
estimates
of
the
annual
willingness
to
pay.

Using
paired
areas
in
Los
Angeles
selected
to
be
homogeneous
with
respect
to
socioeconomic,
housing,
and
community
characteristics
but
with
variation
in
air
pollution,
Brookshire
et
al
.
[
1982]
tested
two
hypotheses:

l
The
rent
differential
for
pollution
should
exceed
estimates
of
annual
willingness
to
pay.

.
Willingness
to
pay
estimated
from
the
contingent
valuation
survey
bids
are
different
from
zero.

The
design
for
the
test
used
a
hedonic
property
model
that
was
estimated
with
sales
of
single­
family
houses
in
these
areas
and
the
contingent
v
a
l
u
a
t
i
o
n
8­
11
experiment
conducted
with
households
selected
from
the
same
areas.
Overall,
the
Brookshire
et
al.
[
1982]
findings
supported
the
presence
of
positive
bids
for
air
pollution
reductions
in
all
areas
,
as
well
as
the
ranking
of
rent
differ.
entials
over
bids
in
10
of
the
11
communities.
Thus,
the
Brookshire
et
al.
[
1982]
analysis
provides
the
first
evidence
that
benefit
estimates
derived
fro
m
survey
procedures
fall
within
the
theoretical
bounds
for
willingness
to
pay.
Nonetheless,
the
comparison
is
based
on
average
responses
within
the
selected
communities
and
not
estimates
at
an
individual
level.

In
summary,
past
efforts
(
especially
those
of
Bishop
and
Heberlein
[
1979]
a
n
d
Brookshire
et
al.
[
1982]
)
directed
toward
comparative
evaluations
of
benefit
methodologies
are
complementary
to
those
available
from
the
comparative
analysis
of
this
study.
The
comparison
of
the
travel
cost
and
contingent
valuation
is
especially
important
because
of
the
ability
to
compare
benefits
estimated
for
the
same
users.

8.3
A
COMPARATIVE
EVALUATION
OF
THE
CONTINGENT
VALUATION,
TRAVEL
COST,
AND
CONTINGENT
RANKING
BENEFIT
ESTIMATION
METHODS
Mean
estimates
are
provided
in
Table
8­
3
for
each
component
of
the
benefits
associated
with
three
water
quality
changes:

.
Deterioration
in
water
quality
leading
to
the
loss
of
the
recreational
use
of
the
area
for
water­
based
activities
.
Improvement
in
water
quality
from
its
present
state
(
beatable
conditions)
to
fishable
conditions
.
Improvement
from
beatable
to
swimmable
conditions.

The
estimates
include
the
option
price
and
its
components­­
user
value
and
option
value.
These
results
are
based
on
different
subsets
of
the
Monongahela
survey
respondents
and
are
measured
in
1981
dollars.
The
contingent
valuation
estimates
are
based
on
the
full
sample,
excluding
protest
bids
and
those
respondents
identified
as
out!
iers
in
the
survey
(
i.
e.
,
using
the
Bels­
Iey,
Kuh,
and
Welsch
[
1980]
regression
diagnostics,
as
detailed
in
Chapter
4).
The
travel
cost
estimates
were
derived
for
the
survey
respondents
who
were
users
of
sites
along
the
Monongahela
River.
*
Finally,
the
contingent
ranking
estimates
relate
to
those
survey
respondents
who
reported
complete
ranking
information
and
income.
Thus,
this
group
includes
some
individuals
who
were
judged
outliers
in
the
contingent
valuation
survey.

*
The
travel
cost
results
include
all
survey
respondents
who
were
users
of
sites
along
the
Monongahela
River,
whether
or
not
they
were
identified
as
p
r
o
t
e
s
t
b
i
d
s
o
r
Belsley,
K
u
h
,
a
n
d
Welsch
[
1
9
8
0
]
outliers.
T
a
b
l
e
C
­
1
8
i
n
Appendix
C
provides
the
regression
comparisons
of
contingent
valuation
and
travel
cost
estimates
with
these
individuals
deleted
from
the
sample.
The
deletion
of
these
respondents
does
change
any
of
our
conclusions.

8­
12
Table
8­
3.
A
Comparison
of
Benefit
Estimates
for
Water
Quality
Improvements
(
1981
Dollars)

AWQ
=
Loss
of
use
AWQ
=
Boatable
to
fishable
AWQ
=
Boatable
to
swimmable
Option
.
User
Option
Option
User
Option
Option
User
valuea
Option
Methodology
y
price
value
price
value
a
value
price
valuea
value
I.
Contingent
valuation
Direct
question
24.55
6.57
17.98
17.65
7.06
10.59
31.20
13.61
20.80
(
19.71)
(
21.18)
(
31.18)

Payment
card
51.00
6.20
44.82
29.26
9.72
19.54
42.87
15.92
26.76
(
19.71)
(
30.88)
(
51.18)

Iterative
bidding
($
25)
28.97
2.16
26.81
15.95
1.38
14.57
`
25.09
3.12
21.64
(
6.58)
(
4.21)
(
10.53)

y
Iterative
bidding
($
125)
57.40
12.08
45.31
36.88
6.77
30.10
60.20
2
(
36.25)
13.43
43.96
(
20.31)
w
(
48.75)

II.
Contingent
rankingc
Ordered
Iogit
60.03
­
108.06
Ordered
normal
62.12
­
­
111.81
I
l
l
.
Generalized
travel
costd
82.65
­
7.01
­
14.71
­

a
The
numbers
in
parentheses
below
the
estimated
user
values
report
average
user
values
for
users
only.
Since
nonusers
have
a
zero
user
value,
the
combined
mean
understates
user
values.

bThese
estimates
are
for
the
combined
sample
including
users
and
nonusers.
It
excludes
protest
bids
and
outliers
detected
using
the
Belsley,
Kuh,
and
Welsch
regression
diagnostics.

Cthese
estimates
are
for
the
sample
of
respondents
with
usable
ranks
and
reported
family
income.
Estimates
evaluated
at
the
intermediate
payment
level.

dThese
estimates
are
for
survey
respondents
using
Monongahela
sites
and
have
been
converted
to
1981
dollars
using
the
consumer
price
index.
Table
8­
3
clearly
illustrates
the
pairwise
comparisons
possible
with
these
three
methods.
Because
contingent
valuation
provides
the
most
complete
set
of
estimates,
it
can
be
compared
to
both
of
the
other
methods
for
several
components
of
the
benefits
from
a
water
quality
change.

Simple
comparisons
of
the
means
in
Table
8­
3
indicate
that
the
relationship
between
the
methods
depends
on
the
type
of
change
in
water
quality
being
considered.
For
example,
in
the
case
of
user
values,
contingent
valuation
estimates
would
be
expected
to
be
less
than
the
travel
cost
estimates
of
ordinary
consumer
surplus
for
improvements
in
water
quality.
However,
based
on
the
arguments
developed
in
the
previous
section
of
this
chapter,
these
differences
would
likely
be
slight.
This
relationship
does
not
seem
to
have
been
upheld
for
improvements
in
water
quality
when
the
mean
willingness
to
pay
for
users
(
reported
in
parentheses
in
Table
8­
3)
is
compared
with
the
ordinary
consumer
surplus
increments.
Three
of
the
four
contingent
valuation
approaches
contrast
with
this
expectation
for
both
of
the
water
changes.
Only
the
mean
for
the
iterative
bidding
format
with
the
$
25
starting
point
is
I.
ess
than
the
ordinary
consumer
surplus
estimate.
Moreover,
the
differences
in
some
cases
are
greater
than
the
theoretical
arguments
would
have
implied.
Because
the
largest
of
these
estimates
is
not
associated
with
the
iterative
bidding
framework
with
a
$
125
starting
point,
the
discrepancy
cannot
be
attributed
to
starting
point
bias.
These
comparisons
are
not
statistical
tests,
and
the
contingent
valuation
estimates
exhibit
considerab~
variabi
lity.
Indeed,
the
travel
cost
estimates
do
fall,
for
both
levels
of
improvement
in
water
quality,
in
the
range
of
estimates
provided
by
the
various
approaches
to
contingent
valuation.

The
comparison
between
the
means
for
the
contingent
valuation
and
travel
cost
estimates
is
consistent
with
theoretical
expectations
for
a
reduction
in
water
quality
that
leads
to
the
loss
of
the
area.
In
this
case,
the
ordinary
consumer
surplus
is
more
than
twice
the
size
of
the
largest
of
the
contingent
valuation
estimates.
The
size
of
this
difference
was
somewhat
unexpected
based
on
the
simple
theoretical
arguments
discussed
earlier.
Accordingly,
it
serves
to
highlight
the
potential
importance
of
each
methodology's
assumptions
in
comparing
their
respective
estimates.
One
explanation
of
this
large
difference
arises
from
an
assumption
implicit
in
the
travel
cost
model.
The
data
required
that
the
travel
cost
demand
model
ignore
the
effects
of
substitute
sites
as
determinants
of
the
demand
for
any
one
site's
services.
However,
judging
the
potential
effects
of
this
limitation
on
the
estimates
from
the
generalized
travel
cost
model
are
difficult.
The
model
developed
in
Chapter
7
assumes
that
each
individual
considered
only
site
attributes
in
judging
the
degree
of
substitutability
between
sites.
Indeed,
it
was
based
on
the
assumption
that
all
sites'
services
could
be
measured
on
a
common
scale
reflecting
these
attributes.
To
the
extent
this
assumption
is
either
inappropriate
or
a
relatively
weak
approximation
of
each
individual's
perceptions
of
the
relationship
between
sites,
there
will
be
two
types
of
effects
on
the
demand
model.
First,
the
omission
of
variables
reflecting
the
prospective
role
of
these
substitution
effects
in
any
site's
demand
function
is
a
specification
error
that
may
bias
estimates
of
the
other
variables'
effects
on
demand.
Equally
important,
the
differential
accessibility
of
substitute
sites
of
comparable
or
higher
quality
will
tend
to
mitigate
the
impact
of
any
deterioration
in
water
quality
at
a
given
8­
14
`
uce
the
incremental
benefits
from
improvements.
Thus,
it
is
difficult
to
predict
with
certainty
the
impacts
of
the
treatment
of
the
role
of
substitutes
for
benefit
estimates
derived
from
the
generalized
travel
cost
model.

Nonetheless,
it
does
seem
reasonable
to
expect
that
the
use
of
a
model
that
ignores
the
role
of
substitutes
may
not
seriously
affect
the
benefit
estimates
associated
with
the
increments
to
water
quality
that
serve
to
enhance
the
activities
supported
by
a
recreation
site.
By
contrast,
this
judgment
is
not
as
readily
accepted
for
the
loss
of
a
site.
I
n
this
case,
the
presence
of
substitute
facilities
can
be
expected
to
mitigate
the
loss.
Thus,
the
generalized
travel
cost
model
(
which
ignores
the
role
of
substitute
sites)
may
overestimate
the
consumer
surplus
associated
with
the
loss
of
the
use
of
the
Monongaheia
River
for
boating
recreation.

The
second
comparison
that
can
be
made
is
between
the
contingent
valuation
and
contingent
ranking
estimates
of
the
option
price.
Regardless
of
the
technique
used
to
estimate
the
random
utility
function,
the
contingent
ranking
approximation
of
option
price
consistently
exceeds
the
contingent
valuation
estimates.
Because
both
methods
focus
on
the
same
benefit
concept,
the
explanations
for
it
must
arise
from
the
assumptions
of
each
approach.
The
approximations
used
to
derive
the
contingent
ranking
benefit
estimates
may
be
especially
important
to
such
an
explanation.
*
However,
in
the
final
analysis,
there
is
little
additional
information
that
can
be
gleaned
from
a
comparison
of
means.

The
most
interesting
comparisons
of
contingent
valuation
and
travel
cost
estimates
are
based
on
the
subsample
of
users;
the
most
interesting
comparisons
of
contingent
valuation
and
contingent
ranking
are
based
on
the
sub­
­

sample
of
respondents
with
complete
information
on
the
ranking
of
water
quality
and
payment
alternatives.
Both
sets
of
comparisons
use
individual
benefit
estimates.

The
comparison
of
contingent
valuation
and
travel
cost
estimates
of
user
values
is
presented
in
Table
8­
4.
The
objective
of
this
comparison
is
to
judge
how
the
benefit
measures
derived
using
the
two
approaches
compared
across
individuals.
Accordingly,
a
common
set
of
procedures
was
used
to
evaluate
the
accuracy
of
a
set
of
forecasts
(
see
Theil
[
1961],
pp.
31­
33,
for
discussion
of
this
type
of
application).
In
this
comparison,
the
contingent
valuation
measure
of
user
value
was
regressed
on
the
travel
cost
estimate.
Because
this
comparison
may
be
affected
by
the
question
format
used
with
the
contingent
valuation
approach,
qualitative
variables
for
three
of
the
four
modes
were
also
included
as
determinants
of
the
level
of
the
contingent
valuation
estimates.

*
This
benefit
measure
is
described
as
approximate
because
of
its
definition
as
an
increment
to
the
payment
required
to
hold
an
individual's
utility
constant
in
the
presence
of
a
water
quality
improvement
and
because
of
the
theoretical
inconsistency
in
the
functions
function
(
see
Chapter
6
for
details).

8­
15
`
form
used
for
the
indirect
utility
.
.
.
.
.
.
.
.
.
.
.
.
..
 
 
.
.
 
.
.
 
.
.
 
 
.
 
.
­
 
 
 
..
 
­
 
.­
 
­
.
 
 
.
 
 
.
 
Table
8­
4.
A
Comparison
of
Contingent
Valuation
and
l_
ARIC
Q­
A
Generalized
Travel
Cost
Benefit
Estimates
a
AWQ
=
Loss
of
area
AWQ
=
Boatable
to
fishable
A
W
Q
=
Boatable
to
swimmable
Model
Testb
Model
Testb
Model
T
e
s
t
b
Independent
variable
Intercept
Travel
cost
benefit
estimate
Qualitative
variables
Payment
card
Direct
question
Iterative
bid
($
25)

R2
n
F
21.862
(
1
.
3
7
1
)

.328
­
4.357
(
1.169)

­
32.640
(­
2.551)
­

­
14.602
(­
1
.270)

­
31.817
(­
2.549)
­

.099
93
2.42
(
0.05)
C
33.985
(
1.900)

­
3.670
(­
1
.204)

51.757
(
2
.
6
3
9
)

12.957
(
0
.
7
4
8
)

­
1
1
.
2
4
4
(
­
0
.
5
9
5
)

.120
93
3
.
0
0
(
0.02)
C
59.574
(
2.017)

­
1
.
7
1
2
­
2.713
­
1
.
7
9
3
(
­
1
.
1
4
1
)

77.010
(
2
.
3
5
9
)

21.001
(
0.729)

­
21.819
(
­
0
.
6
9
3
)

.107
93
2.62
(
0.04)
C
a
The
numbers
in
parentheses
below
the
estimated
coefficients
are
t­
ratios
for
the
null
hypothesis
of
no
association.

b
This
column
reports
the
t­
ratio
for
the
hypothesis
that
the
coefficient
for
the
travel
cost
variable
was
1,55.
The
travel
cost
model
measures
consumer
surplus
in
1977
dollars.
The
contingent
valuation
experiments
were
conducted
in
1981.
Using
the
consumer
price
index
to
adjust
the
travel
cost
benefit
estimates
to
1981
dollars
would
require
multiplying
each
estimate
by
1.55.
Since
the
estimated
regression
coefficients
(
and
standard
errors)
will
correspondingly
adjust
to
reflect
this
scale
change,
a
test
of
the
null
hypothesis
that
the
coefficient
of
travel
cost
was
equal
to
unity
is
equivalent
to
a
test
that
is
equal
to
1.55
w
h
e
n
t
h
e
travel
cost
benefit
estimates
are
measured
in
1977
dollars
and
user
values
estimates
(
the
dependent
variable
are
in
1981
dollars.

cThis
number
in
parentheses
below
the
reported
F­
statistic
is
the
level
of
significance
for
rejection
of
the
null
hypothesis
of
no
association
between
the
dependent
and
independent
variables.

The
analysis
was
considered
for
each
of
three
water
quality
changes:

.
Deterioration
in
water
quality
leading
to
.
Improvement
in
water
quality
from
its
conditions)
to
fishable
conditions
.
Improvement
from
boatable
to
swimmable
the
loss
of
the
areas
present
state
(
boatable
conditions.

The
results
generally
reinforce
the
earlier
judgments
from
comparing
the
estimated
mean
user
values
from
each
method.
Theory
suggests
contingent
valuation
estimates
would
be
less
than
the
ordinary
consumer
surplus
estimates
from
the
travel
cost
model
for
water
quality
improvements,
but
these
differences
should
be
rather
small.
This
g
priori
expectation
can
be
evaluated
by
testing
the
null
hypothesis
that
the
intercept
for
the
model
is
zero.
Equally
important,
if
the
two
methods
provide
comparable
estimates
of
user
values
that
CIOSCIY
8­
16
each
individual's
willingness
to
pay,
the
slope
parameter
for
the
~
pproximate
traVel
cost
consumer
surplus
would
be
expected
to
be
insignificantly
different
from
unitY.
Finally,
if
the
question
mode
does
not
influence
the
responses
deriVe~
with
contingent
valuation
surveys,
the
dummy
variables
for
question
likely
not
be
significantly
different
from
zero.
mode
would
More
formally,
it
has
been
maintained
that
the
contingent
valuation
esti
­

@
eS
of
an
individual's
willingness
to
pay
for
water
quality
changes
"
a",
CV
,
~
ill
be
approximately
a
homogeneous
function
of
the
conditional
expectati~
n
for
the
Marshal
lian
consumer
surplus,
MSa
(
i.
e.
,
the
predicted
consumer
sur­

~
luS
from
the
generalized
travel
cost
model
for
water
quality
change
"
a").
This
function
will
exhibit
a
slope
of
unity.
This
model
is
to
be
distinguished
from
an
errors­
in­
variables
framework
in
which
it
would
be
maintained
that
"
either
benefit
measure
describes
what
it
is
purported
to
measure.
this
study's
interpretation,
Under
the
travel
cost
estimates
of
consumer
surplus
play
the
same
role
as
the
estimates
of
the
conditional
expectation
of
endogenous
variables
in
a
deterministic
simulation
of
an
econometric
model
(
see
H
o
w
r
e
y
and
Kelejian
[
1969]
and
Aigner
[
1972]).
Hence,
large
sample
evaluations
o
f
the
parameters
in
the
model
­­
testing
the
hypotheses
of
zero
intercept
and
uni
­
t
a
ry
slope­­
do
provide
some
guidance
as
to
the
relationship
between
methods.

The
results
provide
some
interesting
insights
for
each
of
these
issues.
considering
the
relationship
between
the
level
of
the
contingent
valuation
estimates
and
those
of
the
travel
cost
model,
there
is
some
evidence
for
a
difference
between
the
levels
of
the
two
approaches
for
improvements
in
water
quality
that
contradicts
q
priori
expectations.
The
intercepts
for
the
e
q
u
a
­
tions
associated
with
both
levels
of
water
quality
increments
(
i.
e.
,
from
beatable
to
fishable
and
from
boatable
to
swimmable)
are
positive
and
statistically
significant
at
the
90­
percent
significance
level.
However,
there
are
at
least
two
reasons
for
interpreting
these
results
cautiously.
The
generalized
travel
cost
model
does
not
permit
the
effect
of
the
intercept
to
be
distinguished
from
at
least
one
of
the
questioning
formats.
In
the
models
reported
in
Table
8­
4,
the
intercept
reflects
the
effects
of
the
iterative
bidding
format
with
a
$
125
starting
point.
Testing
whether
the
sum
of
the
intercept
and
any
one
of
the
coefficients
for
other
models
was
nonzero
would
simply
change
the
format
included
Ignoring
the
effects
of
question
format
by
eliminating
these
variables
from
the
models
simply
reinforces
the
conclusion
that
the
intercept
for
these
cases
is
positive
and
significantly
different
from
zero.

Thus,
there
is
some
evidence
to
support
the
conclusion
that
contingent
valuation
methods
may
overstate
willingness
to
pay
for
water
quality
improvements
It
is
not
unambiguous
evidence,
because
the
tests
are
based
on
large
sample
behavior
and
have
been
applied
using
the
conventional
t­
distributions.
These
findings
are
not
necessarily
at
variance
with
the
Brookshire
et
al.
[
1982]
conclusions.
Their
evaluation
concluded
that
contingent
valuation
estimates
fall
within
the
bounds
which
can
be
established
by
theory.
It
does
not
indicate
how
close
the
estimates
fall
to
the
"
true"
value
of
individual
willingness
to
pay.
An
appraisal
suggests
that,
for
increments
(
improvements)
to
water
quality,
contingent
valuation
estimates
may
well
overstate
the
u
s
e
r
benefits.

8­
17
L
The
conclusion
for
reductions
in
water
quality
that
would
be
associated
with
the
loss
of
the
area
is
less
clearcut.
In
this
case,
the
contingent
valuation
estimates
are
less
than
ordinary
consumer
surplus,
as
theory
would
imply.
However,
they
are
substantially
less,
and
the
reasons
may
be
associated
with
the
travel
cost
model
and
not
the
survey
approach
to
benefit
estimation.
Based
on
the
association
between
estimates
across
individuals,
there
is
support
for
the
conclusion
that
the
travel
cost
model
overstates
the
benefits
associated
with
avoiding
the
loss
of
the
area.
The
slope
coefficient
is
significantly
different
from
theoretical
expectations.
Since
the
travel
cost
benefits
are
measured
in
1977
dollars,
the
correct
null
hypothesis
for
the
slope
coefficient
when
1977
dollars
are
not
converted
to
1981
is
that
the
coefficient
equals
the
adjustment
factor
(
in
this
case,
1
.55).*
For
improvements
in
water
quality,
the
coefficients
are
numerically
large
and
have
an
incorrect
sign,
but
they
are
not
significantly
different
from
1.55.

Thus,
for
changes
in
water
quality,
the
models
do
seem
to
move
together
(
with
the
contingent
valuation
potentially
exhibiting
a
positive
bias
in
estimating
willingness
to
pay).
The
performance
of
the
contingent
valuation
method
does
appear
to
depend
on
the
mode
of
questioning
used­­
with
the
clearest
distinctions
found
between
the
payment
card
and
iterative
bid
with
a
$
125
starting
point.
While
the
explanatory
power
of
the
model
is
not
high,
reflecting
the
variability
in
the
contingent
valuation
responses
for
user
values,
the
null
hypothesis
of
no
association
between
these
measures
of
user
values
(
along
with
the
qualitative
variables)
is
clearly
rejected
at
high
levels
of
significance
based
on
the
F­
statistics,
reported
at
the
bottom
of
the
table.

The
second
individual
level
comparison
involves
estimates
of
the
option
price
using
contingent
valuation
and
contingent
ranking
methods.
Table
8­
5
reports
a
comparable
set
of
regression
models
comparing
these
estimates.
However
two
further
distinctions
are
possible
in
this
comparison.
Given
the
functional
form
specified
for
the
indirect
utility
function,
the
contingent
ranking
estimate
of
option
price
will
depend
on
the
level
of
the
payment
suggested
to
the
individual.
Consequently,
the
benefits
were
calculated
at
all
three
levels
and
the
regressions
were
replicated
for
each
of
them.
In
addition,
two
econometric
estimators
were
used
with
the
contingent
ranking
models
so
that
each
was
also
considered.
Table
8­
5
reports
all
of
the
comparisons
for
two
increments
in
water
quality
­­
improvements
from
boatable
to
fishable
and
from
b
o
a
t
a
b
l
e
t
o
s
w
i
m
m
a
b
l
e
.
"
­

*
Scaling
all
the
values
of
an
independent
ordinary
least­
squares
estimate
of
the
parameter
model)
and
its
estimated
standard
esis
of
unity
for
such
a
parameter
~­
k
1
variable
by
k
will
scale
the
for
this
variable
(
in
a
linear
error
by
~.
Thus,
to
test
the
null
hypothwould
imply
using
1
A
I
­
orbSo
`
b
`
b
T
T
a
b
l
e
8
­
5
.
A
C
o
m
p
a
r
i
s
o
n
o
f
C
o
n
t
i
n
g
e
n
t
V
a
l
u
a
t
i
o
n
a
n
d
C
o
n
t
i
n
g
e
n
t
Ranking
E3enetit
E
s
t
i
m
a
t
e
s
,.

AWQ
=
Beatable
to
fishable
AWQ
=
Boatablo
to
swimmab\
e
Payment
=
$
50
Payment
=
$
100
Payment
=
$
175
Payment
=
$
50
Payment
=
$
100
Payment
=
$
175
Independent
variable
Model
Test
Model
Test
Model
Test
Model
Test
Model
Test
Model
Test
I
co
I
co
ORDERED
LOGIT
Intercept
A
Payment
Qualitative
variables
Payment
card
Direct
question
Iterative
bidding
R2
n
.
r
ORDERED
NORMAL
Intercept
A
Payment
qualitative
variables
Payment
card
Direct
question
Iterative
bidding
R
2
n
F
($
25)

($
2s)
­
20.141
(­
1.095)

1.209
(
4.279)

­
22.486
(­
2.424)

­
35.267
(­
3.7s1)

­
38.045
(­
4.067)

.165
184
8.67
(
0.0001)

­
13.467
(­
0.839)

1.073
(
4.554)

­
22.642
(­
2.457)

­
34.934
(­
3.745)

­
37.541
(­
4.014)

.176
184b
9.53
(
0.0001)
b
0.741
0.309
­
23.647
(­
1.223)

1.315
(
4.237)

­
22.070
(­
2.360)

­
34.595
(­
3.683)

­
37.562
(­
4.015)

.164
184
8.77
(
0.
OQO1)

­
15.565
(­
0.940)

1.140
(
4.528)

­
22.357
(­
2.426)

­
34.458
(­
3.696)

­
37.196
(­
4.004)

.175
184b
9.47
(
0.0001)
1.016
­

0.554
­
23.927
(­
1.227)

1.330
(
4.214)

­
21.960
(­
2.367)

­
34.425
(­
3.665)

­
37.446
(­
4.001)

.163
184
8.72
(
0.0001
­
15.832
(­
0.951)

1.151
(
4.516)

­
22.286
(­
2.418)

­
34.344
(­
3.683)

­
37.116
(­
3.994)

.174
184b
9.43
(
0.0001)
1.048
0.592
­
25.661
(­
0.795)

1.081
(
3.925)

­
46.842
(­
2.877)

­
55.327
(­
3.353)

­
68.611
(­
4.178)

.153
184
8.06
(
0.0001)

­
15.153
(­
0.537)

.962
(
4.182)

­
47.108
(­
2.910)

­
54.808
(­
3.345)

­
67.808
(­
4.156)

.162
184b
8.63
(
0.0001)
0.283
­
0.165
­
30.734
(­
0.905)

1.170
(
3.867)

­
46.145
(­
2.834)

­
54.215
(­
3.288)

­
67.817
(­
4.128)

.151
184
7.94
(
0.0001)

­
18.212
(­
0.626)

1.018
(
4.146)

­
46.630
(­
2.880)

­
54.020
(­
3.298)

­
67.242
(­
4.120)

.160
184b
8.54
(
0.0001)
0.561
0.073
­
31.032
(­
0.906)

1.183
(
3.841)

­
45.961
(­
2.822)

­
53.935
(­
3.270)

­
67.626
(­
4.115)

.150
184
7.88
(
0.0001)

­
18.559
(­
0.634)

1
.
0
2
8
(
4.131)

­
46.510
(­
2.872)

­
53.832
(­
3.286)

­
67.112
(­
4.111)

.160
184b
8.51
(
0.0001)
0.594
0.113
l
 
These
esllmates
are
for
regression
diagnostics.
b
These
l
 
stlmates
are
for
the
combined
sample
including
users
and
nonusers.
It
excludes
protest
bids
and
outliers
detected
using
the
Kuh­
Welsch
the
sample
of
respondents
w!
th
usable
ranks
and
reported
family
income.
The
interpretation
of
these
results
is
somewhat
different
from
the
earlier
comparison
with
travel
cost
estimates.
I
n
this
case,
both
methods
seek
to
estimate
the
same
benefit
concept.
However,
they
are
not
independent.
Each
survey
respondent
was
asked
to
engage
in
both
activities­­
one
of
four
types
of
contingent
valuation
experiment
and
a
contingent
ranking.
Thus,
these
results
reflect
the
consistency
in
individuals'
responses
and
the
potential
effects
of
how
the
valuation
exercise
is
undertaken
(
i.
e.
,
requests
for
bids
or
ranks).
Despite
the
fairly
substantial
differences
in
the
means
for
the
two
approaches
as
reported
in
Table
8­
3,
these
results
exhibit
remarkable
consistency.
Once
again,
the
relevant
hypotheses
are
for
zero
intercept
and
unitary
slope
coefficients
Both
hypotheses
cannot
be
rejected
across
all
possible
variants
of
the
contingent
ranking
and
changes
in
water
quality.
Indeed,
the
numerical
estimates
of
the
slope
coefficient
exhibit
rather
considerable
agreement
between
the
direction
of
the
movements
in
the
two
estimates
of
option
price.
The
estimated
coefficients
for
the
question
format
used
are
especially
interesting.
They
indicate
that
the
association
between
the
two
approaches
depends
quite
importantly
on
the
question
format,
with
the
iterative
bidding
format
with
a
$
125
starting.
point
providing
larger
estimates
than
any
of
the
other
three
formats.

Overall,
these
findings
suggest
that
even
though
the
models
used
to
derive
benefit
estimates
from
the
contingent
ranking
models
were
somewhat
arbitrary
(
and
in
some
cases
inconsistent
with
a
strict
interpretation
of
the
relevant
theory),
the
results
move
closely
with
the
contingent
valuation
estimates
Indeed,
one
of
the
primary
sources
of
divergence
between
the
two
arises
in
the
format
used
with
the
contingent
valuation
questions.

8.4
IMPLICATIONS
This
chapter
has
developed
comparisons
of
three
methods
for
estimating
the
benefits
from
water
quality
improvements.
Each
method
has
involved
a
fairly
detailed
set
of
assumptions
and,
in
some
cases,
a
complex
model.
Overall
the
results
are
remarkably
consistent
across
methods
for
comparable
changes
in
water
quality.
While
this
discussion
has
been
devoted
to
the
types
of
discrepancies
between
each
method's
estimates,
the
consistency
in
these
estimates
should
be
interpreted
as
offering
strong
support
for
the
feasibility
of
performing
benefit
analyses
for
water
quality
changes.
The
range
of
variation
in
estimates
across
methods
is
generally
less
than
the
variation
expected
in
models
seeking
to
translate
the
effects
of
effluent
in
a
water
body
into
the
corresponding
measures
of
water
quality
parameters.

Nonetheless,
this
conclusion
does
not
imply
that
there
is
not
room
for
improvement
in
benefit
estimation
methods.
I
n
most
cases,
the
indirect
methods
for
benefit
measurement,
such
as
the
travel
cost
framework,
have
been
limited
by
the
data
availability.
While
this
study's
analysis
was
greatly
enhanced
by
the
existence
of
the
Federal
Estate
Survey,
the
form
of
the
data
nonetheless
imposed
limitations
on
the
character
of
the
travel
cost
demand
models
that
could
be
formulated.
Survey
approaches
do
not
face
the
same
types
of
limitations.
However,
this
study's
findings
do
suggest
that
the
question
format
used
is
an
important
factor
in
the
benefit
estimates
derived
from
the
survey.
They
also
suggest
that
greater
attention
to
the
nature
and
form
8­
20
of
the
information
provided
to
survey
respondents
will
be
needed
if
this
approach
is
to
seek
to
develop
detailed
measures
of
the,
components
of
benefits.

The
analysis
performed
for
this
study
had
the
advantage
of
a
well­
defined
valuation
problem
that
was
easily
explained
and,
according
to
interviewer
feedback
after
the
survey,
readily
understood
by
the
survey
respondents.
Many
of
the
most
complex
environmental
valuation
problems
do
not
share
this
characteristic
and
therefore
may
not
have
the
same
successes
reported
here.

The
specific
findings
of
the
comparison
indicated
that
contingent
valuation
methods
may
overstate
the
willingness
to
pay
for
water
quality
improvements
Theory
would
suggest
that
ordinary
consumer
surplus
should
provide
an
upper
bound
for
these
estimates
and
this
study's
findings
indicate
it
dOeS
not.
Nonetheless,
these
differences
are
not
substantial
and
fall
within
the
range
of
variation
of
the
contingent
valuation
estimates
across
the
question
formatS.
For
the
case
of
the
loss
of
the
use
of
the
area,
the
association
adheres
to
theoretical
anticipations.
Indeed,
there
are
reasons
to
believe
that
the
travel
cost
estimates
overstate
the
benefits
provided
by
the
area.

Comparison
between
the
contingent
ranking
and
contingent
valuation
estimates
indicate
a
remarkable
degree
of
consistency.
While
the
mean
benefit
estimates
derived
from
the
contingent
ranking
framework
appear
larger
than
the
contingent
valuation
estimates,
there
is
not
a
statistically
significant
displacement
between
the
two.
Moreover,
the
benefit
estimates
move
in
close
agreement
across
individuals.

8­
21
.
.
.
.
.
.
.
.
.
.
.
..
­
 
­
­
 
.
.
.
.
­
 
 
­
..
 
.
 
.
 
 
 
.
.
 
.
..
.
 
.
­
 
 
I
.
CHAPTER
9
REFERENCES
Aignerf
Dennis
J.
,
1972,
"
A
Note
on
Verification
of
Computer
Simulation
Model
s,"
Management
Science,
Vol.
18,
July
1972,
pp.
615­
19.

Aizen,
I
.
r
and
M.
Fishbien,
1977,
"
Attitude­
Behavior
Relations:
A
Theoretical
Analysis
and
Review
of
Empirical
Research,
"
Psychological
Bulletin,

vol.
8
4
,
1
9
7
7
,
p
p
.
8
8
8
­
9
1
8
.
l
Allen,
P.
Geoffrey,
Thomas
H.
Stevens,
and
Scott
A.
Barrett,
1981,
"
The
Effects
of
Variable
Omission
in
the
Travel
Cost
Technique,
"
Land
Economics,
Vol.
57,
No.
2,
May
1981,
pp.
173­
80.

Amemiya,
Takeshi,
1981,
"
Qualitative
Response
Models:
A
Survey,
"
Journal
~
Economic
Literature,
Vol.
19,
No.
4.,
December
1981,
pp.
1483­
536.

Anderson,
R
.
J
.
,
1
9
8
1
,
"
A
Note
on
Option
Value
and
the
Expected
Value
of
Consumer's
Surplus,
"
Journal
q
Environmental
Economics
and
Management
Vol.
8,
June
1981,
pp.
187­
91.

A
r
r
o
w
,
K.
J.,
and
A.
C.
F
i
s
h
e
r
,
1
9
7
4
,
"
Environmental
Preservation,
Uncertainty
and
Irreversibility,
"
Quarterly
Journal
of
Economics,
Vol.
88,
May
1974,
pp.
313­
19.
 
Barker,
Mary
L.
,
1971,
"
Beach
Pollution
in
the
Toronto
Region,
"
in
W.
R.
Sewell
and
Ian
Burton,
eds.
,
Perceptions
and
Attitudes
~
Resource
Management,
Ottawa,
Canada:
D
e
p
a
r
t
m
e
n
t
o~
Energy,
Mines,
and
Resources
1971,
pp.
37­
48.

Becker,
Gary.
S.
,
1965,
"
A
Theory
of
the
Allocation
of
Time,
"
Economic
Journal,
Vol.
75,
September
1965,
pp.
493­
517.

Becker,
Gary
S.
,
1974,
"
A
Theory
of
Social
Interactions,
"
Journal
~
Political
Economy,
Voi
.
82,
1974,
pp.
1063­
93.

Becker,
Gary
S.
,
1981,
"
Altruism
in
the
Family
and
Selfishness
in
the
Market
Place,
"
Economics,
Vol.
48,
February
1981,
pp.
1­
16.

Beggs,
S.
,
S.
Cardell,
a
n
d
J
.
Hausman,
1
9
8
1
,
"
A
s
s
e
s
s
i
n
g
t
h
e
P
o
t
e
n
t
i
a
l
Demand
for
Electric
Cars,
"
Journal
of
Econometrics,
Vol.
16,
September
1981,
pp.
1­
19.
 
9­
1
Belsley,
David
A.
,
Edwin
Kuh,
and
Roy
E.
Welsch,
1980,
Regression
Dia~­
nostics,
New
York:
John
Wiley
and
Sons,
1980.

Berndt,
Ernst,
R.
,
1983,
"
Quality
Adjustment
in
Empirical
Demand
Analysis,
"
Working
Paper
1397­
83,
Sloan
Schools,
Massachusetts
Institute
of
Technology
Cambridge,
Massachusetts:
January
1983.

Binkley,
Clark
S.
,
and
W.
Michael
Hanemann,
1978,
The
Recreation
Benefits
q
Water
Quality
Improvement:
A
n
a
l
y
s
i
s
g
D~
Trips
i
n
a
n
U
r
b
a
n
S_#_
Qt
W
a
s
h
i
n
g
t
o
n
,
D
.
C
.
:
U
.
S
.
Environmental
Protect~
on
`
A
g
e
n
c
y
,
.

Bishop,
R.
C.,
and
T.
A.
Heberlein,
1
9
7
9
,
"
Measuring
Values
of
Extra­
Market
Goods
:
Are
Indirect
Measures
Biased?"
American
Journal
g
Agricultural
Economics,
Vol.
6,
December
1979,
pp.
926­
30.

Bishop,
Richard
C.
,
1982,
"
Option
Value:
An
Exposition
and
Extension,
"
Land
Economics,
Vol.
58,
February
1982,
pp.
1­
15.

Blackorby,
Charles,
Daniel
Primont,
and
R.
Robert
Russell,
1978,
Separability,
and
Functional
Structure:
I@
XY
@
Economic
tions,
New
Yor~
North
Holland,
1978.
Duality,

L@!!!
22­

Bockstael,
Nancy,
1982,
Discussion
comments
at
the
Association
of
Environmental
and
Resource
Economists
Session,
American
Economic
Association
Annual
Meeting,
1982.

Bockstael,
Nancy
E.
,
and
Kenneth
E.
McConnell,
1980,
"
Calculating
Equivalent
and
Compensating
Variation
for
Natural
Resource
Environments,
"
Land
Economics,
Vol.
56,
No.
1,
February
1980,
pp.
56­
63.

Bockstael,
Nancy
E.
,4
and
Kenneth
E.
McConnell,
1981,
"
Theory
and
Estimation
of
the
Household
Production
Function
for
Wildlife
Recreation,
"
Journal
of
Environmental
Economics
and
Management,
Vol.
8,
September
1981,
 
 
pp.
199­
214.

Bockstael,
Nancy
E.
,
and
Kenneth
E.
McConnell,
1982,
"
Welfare
Measurement
in
the
Household
Production
Framework,
"
unpublished
paper,
Department
of
Agricultural
and
Resource
Economics,
University
of
Maryland,
College
Park,
Maryland,
February
1982.

Bohm,
Peter,
1971,
"
An
Approach
to
the
Problem
of
Estimating
Demand
for
Public
Goods,
"
Swedish
Journal
~
Economics,
Vol.
73,
March
1971,
pp.
55­
66.

Bohm,
Peter,
1975,
"
Option
Demand
and
Consumer
Surplus:
Comment,
"
American
Economic
Review,
Vol.
65,
September
1975,
pp.
733­
36.

Bouwes,
Nicolaas
W.
,
Sr.
,
and
Robert
Schneider,
1979,
"
Procedures
in
Estimating
Benefits
of
Water
Quality
Change,
"
American
Journal
~
Agricultural
Economics,
August
1979,
pp.
535­
39.

9
­
2
.
.
~
r@
k5hire,
David
S.
,
Ronald
G
.
C
u
m
m
i
n
g
s
,
M
o
r
t
e
z
a
R
a
h
m
a
t
i
a
n
,
William
D.
Sctlulzef
and
Mark
A.
Thayer,
1982,
Experimental
AP
P
~
nt~~
vironmental
roaches
for
Valu
­
Commodities,
draft
report
prepared
for
U.
S.
~
nvfiprotection
Agency,
University
of
Wyoming,
Laramie,
Wyoming,

April
1982.

kshire,
David
S.
,
Ralph
C.
d'Arge,
William
D.
Schulze,
a
n
d
M
a
r
k
A
.

Thayerl
1979,
Methods
Development
for
Assessinq
Air
Pollution
Control
Benefits,
Volume
I
I,
Experiments
@
V=
uin~
Non­
Ma=
et
Goods:
A
Case
Study
~
Alternative
Benefit
Measures
of
$&
Pollution
Control
`&
~
south
Coast
Air
Bas~
n
~
Southern
Calif=
rnla,
EPA­
600
/
5­
79
­
OOlb,
U
.
S.
fionmental
Protection
Agency,
Washington,
D.
C.,
1979.

Brookshirel
David
S.,
B.
Ives,
and
William
D.
Schulze,
1976,
"
The
Valuation
of
Aesthetic
Preferences,
"
Journal
~
Environmental
Economics
and
Management
Vol.
3,
December
1976,
pp.
325­
46.
 
 
SrOokshire,
David
S.,
a
n
d
A
l
a
n
R
a
n
d
a
l
l
,
1
9
7
8
,
"
Public
Policy
Alternatives,
public
Goods,
and
Contingent
Valuation
Mechanisms,
"
paper
presented
at
the
Western
Economic
Association
Meeting,
Honolulu,
Hawaii,
June
1978,
pp.
20­
26.

Brookshire,
David
S.,
Mark
A.
Thayer,
William
D.
Schulze,
a
n
d
R
a
l
p
h
C
.
d'Arge,
1982,
"
valuing
public
G
o
o
d
s
:
A
Comparison
of
Survey
and
Hedonic
Approaches,
"
American
Economic
Review,
Vol.
72,
March
1982,
pp.
165­
77.

Brown,
Gardner,
Jr.
,
and
Robert
Mendelssohn,
1980,
"
The
Hedonic
Travel
Cost
Method,
"
unpublished
paper,
Department
of
Economics,
University
of
Washington,
Seattle,
Washington,
December
1980.

6rown,
William
G.,
and
Farid
Nawas,
1973,
"
Impact
of
Aggregation
on
the
Estimation
of
Outdoor
Recreation
Demand
Functions,
"
American
Journal
~
Agricultural
Economics,
Vol.
55,
May
1973,
pp.
246­
49.

Burt,
O.
R.,
and
D.
Brewer,
1971,
"
Estimation
of
Net
Social
Benefits
from
Outdoor
Recreation,
"
Econometrics,
Vol.

Cesario,
Frank
J.
,
1976,
"
Value
of
Time
in
Economics,
Vol.
52,
February
1976,
pp.

Cesario,
Frank
J.,
and
Jack
L.
Knetsch,
Benefit
Estimates,
"
Water
Resources
pp.
700­
04.
39,
October
1971,
pp.
813­
27.

Recreation
Benefit
Studies,
"
Land
32­
41.

1970,
"
Time
Bias
in
Recreation
Research,
Vol.
6,
June
1970,

Cesario,
Frank
J.
,
and
Jack
L.
Knetsch,
1976,
"
A
Recreation
Site
Demand
and
Benefit
Estimation
Model,
"
Regional
Studies,
Vol.
10,
1976,
pp.
97­
104.

Cicchetti,
Charles
J.
,
Anthony
C.
Fisher,
and
V.
Kerry
Smith,
1976,
"
An
Economic
Evaluation
of
a
Generalized
Consumer
Surplus
Measure:
The
Mineral
King
Controversy,
"
Econometrics,
Vol.
44,
November
1976,
pp.
1259­
76.

9
­
3
..
­.,
"
 
_
____
.=
.­.
 
 
­
.
_______
Cicchetti,
Charles
J.
,
and
A.
Myrick
Freeman,
111,
1971,
"
Option
Demand
and
the
Consumer
Surplus:
Further
Comment,
"
Quarterly
Journal
of
Economics
Vol.
85,
August
1971,
pp.
528­
39.
 
Cicchetti,
Charles
J.,
Joseph
J.
Seneca,
and
Paul
Davidson,
1969,
The
Demand
and
Supply
~
Outdoor
Recreation,
New
Brunswick,
New
Jersey:
 
.
 
 
Bureau
of
Economic
Research,
Rutgers
University,
1969.

Cicchetti,
Charles
J.,
and
V.
Kerry
Smith,
1976,
The
Costs
of
Congestion,
Cambridge,
Massachusetts:
Ballinger
Publishing
Co.
,
1976.
 
 
.

Clawson,
M.,
1959,
"
Methods
of
Measuring
the
Demand
Outdoor
Recreation,
"
Reprint
No.
10,
Resources
for
Washington,
D.
C.,
1959.

Clawson,
M.
,
and
J.
L.
Knetsch,
1966,
Washington,
D.
C.
:
Resources
for
the
Conrad,
J.
M.
,
1980,
"
Quasi
Option
Value
tion,
"
Quarterly
Journal
of
Economics,
 
for
and
Value
of
the
Future,
Inc.
,

Economics
of
Outdoor
F
u
t
u
r
e
,
Inc
.
;
1
966.

and
the
Expected
Value
VOI.
94,
June
1980,
pp.
Recreation,

of
Informa­
813­
20.

C
o
o
k
,
PhiliD
J.,
and
Daniel
A.
Graham,
1977,
"
The
Demand
for
Insurance
and
`
Protection;
The
Case
of
I
replaceable
Commodities,
"
Quarterly
Journal
of
Economics,
Vol.
91,
February
1977,
pp.
143­
56.
 
 
 
Council
of
Economic
Advisors,
1982,
"
Economic
Report
of
the
President,
"
Washington,
D.
C.
:
U.
S.
Government
Printing
Office,
1982.

COX,
D
.
R
.
,
1
9
7
2
,
"
Regression
Models
and
Life­
Tables,
"
Journal
of
the
Royal
Statistical
Society,
Series
B,
Vol.
34,
1972,
pp.
187­
202.
 
 
Cronin,
F
r
a
n
c
i
s
J
.
,
1
9
8
2
,
V
a
l
u
i
n
g
Nonmarket
Goods
Throuqh
Contingent
Markets,
prepared
for
U.
S.
Environmental
Protection
Agency,
Pacific
Northwest
Laboratory,
Richland,
Washington,
September
1982.

Davidon,
Fletcher
R.,
and
M.
Powell,
1963,
"
A
Rapidly
Convergent
Descent
Method
for
Minimization,
"
The
Computer
Journal,
Vol.
6,
1963,
P
P.
163­
68.

Davis,
Robert
K.
,
1963,
"
Recreation
Planning
as
an'
Economic
Problem,
"
Natural
Resources
Journal,
Vol.
3.,
No.
2,
October
1963,
pp.
239­
49.

Deaton,
Angus,
and
John
Muellbauer,
1980,
Economics
and
Consumer
Behavior,
Cambridge:
Cambridge
University
Press,
1980.

Deyak,
Timothy
A.,
and
V.
Kerry
Smith,
1978,
"
Congestion
and
Participation
in
Outdoor
Recreation:
A
Household
Production
Approach,
"
Journal
~
Environmental
Economics
and
Manaqement,
Vol.
5,
March
1978,
pp.
63­
80.

Ditton,
B.
,
and
Recreational
No.
3,
1973,
T
.
L.
Goodale,
1973,
"
Water
Users
of
Green
Bay,
"
Water
pp.
569­
79.

9
­
4
Quality
Perceptions
and
the
Resources
Research,
Vol.
9,
Dwyer,
J.
CF.
,
J.
R.
Kelly,
and
M.
D.
Bowes,
1977,
Improved
Procedures
for
valuation
of
the
Contribution
~
Recreation
to
National
Economic
Develo~
 
 
 
ment~
Urbana­
champaign:
University
of
Illinas,
Ig77.

Feenber9,
Daniel,
and
Edwin
S.
Mills,
1980,
Measuring
the
Benefits
of
Water
Pollution
Abatement,
New
York:
Academic
Press,
1980­
 
 
Fisher,
A.
C.
,
and
V.
Kerry
Smith,
1982,
"
Economic
Evaluation
of
Energy's
Environmental
Costs,
with
Special
References
to
Air
Pollution,
"
Annual
Review
of
Eneray,
voi.
7,
1982,
pp.
1­
35.
 
 
Fisher,
Ann,
and
Robert
Raucher,
1982,
"
Comparison
of
Alternative
Methods
of
Evaluating
the
Intrinsic
Benefits
of
Improved
Water
Quality,
"
paper
presented
at
the
American
Economics
Association
Annual
Meeting,
New
York,
New
York,
December
1982.

F
i
s
h
e
r
,
Frapklin
M.
,
and
Karl
Shell,
1968,
"
Taste
and
Quality
Changes
in
the
Pure
Theory
of
the
True
Cost­
of­
Living
Index,
"
in
J.
N
.
Wolfe,
ed.
,
Value,
w!
.
and
Growth:
@
XZ2
@
HOnOUr
g
w
J
o
h
n
H
i
c
k
s
,
mgo
:
Aldine
Publishing
Co.
,
1968.

Freeman,
A.
Myrick,
I
I
1,
1979a,
The
Benefits
of
Environmental
Improvement:

=
2!
2s!
practicer
B
a
l
t
i
m
o
r
e
:
Johns
Ho~
kins
Press
for
Resources
for
the
Future,
Inc.
,
1979.

Freeman,
A.
Myrick,
I
II,
1979b,
The
Benefits
of
Air
and
Water
Pollution
Control:
~
R
e
v
i
e
w
a
n
d
Synt=
is
of
Rece;
t
=
tim=
s,
p
r
e
p
a
r
e
d
f
o
r
Council
on
Env=
ental
Quality,
Wash~
ngton,
D.
C.
,
December
1979.

Freeman,
A
.
Myrick,
II
1
,
1979c,
"
Hedonic
prices,
property
values
and
Mea~­

uring
Environmental
Benefits:
A
Survey
of
the
Issues,
"
Scandinavian
Journal
g
Economics,
Vol.
81,
No.
2,
1979,
pp.
154­
73.

Freeman,
A
.
Myrick,
I
l
l
,
1
9
8
1
,
"
Notes
on
Defining
and
Measuring
Existence
Values,
"
unpublished
manuscript,
Department
of
Economics,
Bowdoin
College,
Brunswick,
Maine,
June
1981.

Freeman,
A.
Myrick,
Ill,
1982,
"
The
Size
and
Sign
of
Option
Value,
"
unpublished
paper,
Bowdoin
College,
Brunswick,
Maine,
1982.

Giles,
D.
E.
A.,
1982,
"
The
Interpretation
of
Dummy
Variables
in
Semilogar
­
ithmic
Equations:
Unbiased
Estimation,
"
Economic
Letters,
VOI
.
I(),
1982,
pp.
77­
79.

Goldberger,
A.
S.
,
1968,
`
lThe
lnterDretation
and
Estimation
of
Cobb­
Douglas
Function
s,"
Econometrics,
Vol
.
36,
July
to
October
1968,
pp.
464­
72.
­

G
r
a
h
a
m
,
D.
A.
,
1981,
"
C
o
s
t
­
B
e
n
e
f
i
t
Analysis
Economic
Review,
Vol.
71,
September
1981,
Under
Uncertainty,
"
American
pp.
715­
25.

9­
5
Greene,
William
H.,
1981,
"
On
the
Asymptotic
Bias
of
the
Ordinary
Least
Squares
Estimates
of
the
Tobit
Model,
"
Econometrics,
Vol.
49,
March
1981,
pp.
505­
14.

Greene,
William
H.
,
1983,
"
Estimation
of
Limited
Dependent
Variable
Models
by
Ordinary
Least
Squares
and
the
Method
of
Moments,
"
Journal
of
Econometrics
Vol.
21,
February
1983,
pp.
195­
212.
.
 
G
r
e
e
n
l
e
y
,
D
.
A
.
,
R
i
c
h
a
r
d
G
.
Walsh,
a
n
d
R
o
b
e
r
t
A.
young,
1981,
"
option
Value:
Empirical
Evidence
from
Quality,
"
Quarterly
J
o
u
r
n
a
l
~
pp.
657­
74.

Greenley,
D.
A.
,
Richard
G.
Walsh,
Mitchell
and
Carson's
Proposed
Evidence
From
a
Case
Studv
of
a
Case
Study
of
Recreation
and
Water
Economics,
Vol.
96,
November
1981,

and
Robert
A.
Young,
1983,
"
Notes
on
Comment
on
`
Option
Value:
Empirical
Recreation
and
Water
Quality,
"
unpub­
.
Iished
manuscript,
Department
of
Economics,
Colorado
State
`
University,
Fort
Collins,
Colorado,
January
1983,
pp.
1­
12.

Gum,
R.,
and
W.
E.
Martin,
1975,
"
Problems
and
Solutions
in
Estimating
the
Demand
for
the
Value
of
Rural
Outdoor
Recreation,
"
American
Journal
of
Agricultural
Economics,
Vol.
57,
November
1975,
pp.
558­
66.
 
Haspel,
Abraham,
E.
,
and
F.
Reed
Johnson,
1982,
"
Multiple
Destination
Trip
Bias
in
Recreation
Benefit
Estimation,
"
Land
Economics,
Vol.
58,
August
1982,
pp.
364­
72.

Hause,
John
C.,
1975,
"
The
Theory
of
Welfare
Cost
Measurement,
"
Journal
of
Political
Economy,
Vol.
83,
December
1975,
pp.
1145­
82.
 
Hausman,
Jerry
A.
,
1978,
"
Specification
Error
Tests
in
Econometrics,
"
Econometrics
Vol.
46,
November
1978,
pp.
1251­
72.

Hausman,
Jerry
A.
,
1981,
"
Exact
Consumer's
Surplus
and
Deadweight
Loss,
"
American
Economic
Review,
Vol.
71,
No.
4,
September
1981,
pp.
662­
76.

Hausman,
Jerry
A.
,
and
David
A.
Wise,
1978,
"
A
Conditional
Profit
Model
for
Qualitative
Choice:
Discrete
Decisions
Recognizing
Interdependence
and
Heterogeneous
Preferences,
"
Econometrics,
Vol.
42,
March
1978,
pp.
403­
26.

Henry,
C.
,
1974,
"
Option
Values
in
the
Economics
of
Irreplaceable
Assets,
"
Review
of
Economic
Studies,
Vol.
64,
1974,
pp.
89­
104.
 
 
Hicks,
John
R.,
1943,
Ii
The
Four
Consumers'
Surplus,
"
Review
of
Economic
Studies,
Vol.
11,
Winter
1943,
pp.
31­
41.
 
Hirshleifer,
J.
,
1970,
Investment,
Interest,
and
Capital,
Englewood
Cliffs,
New
Jersey:
Prentice
Hall,
1970.

9­
6
I
t
t
Y
1
r
!

­
1
II
t
e
f
>
t
f
I
.
=
~
owre~
l
E
.
Phillip,
and
Harry
H.
Kelejian,
1
9
6
9
,
"
Simulation
versus
Analytical
solutions,
"
in
T.
H.
Naylor,
ed.
,
The
Design
of
Computer
Simulation
E~
perimentsl
Durham,
North
Carolina:
Duke
University
Press,
1969.

J@
ftson
/
Norman
L.
,
and
Samuel
Kotz,
1970,
Continuous
Univariate
Distributions
l,
New
York:
Houghton
Mifflin,
1970.

JuSt/
Richard
E.
,
Darell
L.
Hueth,
and
Andrew
Schmitz,
1982,
Applied
Welfare
Economics
~
Public
Policy,
Englewood
Cliffs,
New
Jersey:
Pren~
e
~
1,
1982.

Keener,
Robert,
and
Donald
M.
Waldman,
1981,
"
Maximum
Likelihood
Regression
of
Rank­
Censored
Data,
"
unpublished
paper,
Department
of
Economics,
university
of
North
Carolina
at
Chapel
Hill,
Chapel
Hill,
North
Carolina,
October
1981.

Kelejian,
Harry
H
.,
1971,
"
Two
Stage
Least
Squares
and
Econometric
Systems
Linear
in
parameters
but
Nonlinear
in
Endogenous
Variables,
"
Journal
of
the
American
Statistical
Association,
Vol.
66,
June
1971,
pp.
373­
74.
 
Klein,
R
.
W.,
L.
G.
Rafsky,
D.
F.
Sibley,
and
R.
D.
Willig,
1978,
"
Decisions
with
Estimation
Uncertainty,
"
Econometrics,
Vol.
46,
November
1978,
pp.
1363­
88.

Knetsch,
Jack
L.
,
and
Robert
K.
Davis,
1966,
"
Comparison
of
Methods
for
Recreation
Evaluation,
"
in
A.
V.
Kneese
and
S.
C.
Smith,
eds.
,
Water
Research,
Baltimore:
Johns
Hopkins,
1966.

Krutilla,
J.
V.,
1967,
"
Conservation
Reconsidered,
"
American
Economic
Review,
Vol.
57,
September
1967,
777­
86.

Krutilla,
J
.
V
.
,
and
A.
C.
Fisher,
1975,
~
Economics
~
Natural
Environments
Baltimore:
Johns
Hopkins
Press
for
Resources
for
the
Future,
Inc.
,
1975.

Lau,
Lawrence,
J.
,
1982,
"
The
Measurement
of
Raw
Materials
Inputs,
"
in
V.
Kerry
Smith
and
John
V.
Krutilla,
e
d
s
.
,
E
x
p
l
o
r
a
t
i
o
n
s
i
n
Natural
Resource
Economics,
Baltimore:
Johns
Hopkins,
1982,
pp.
167­
2~
0.

Mater,
Karl
G.
,
1974,
Environmental
Economics:
~
T
h
e
o
r
e
t
i
c
a
l
I
n
q
u
i
r
y
,
Baltimore:
Johns
Hopkins
Press
for
Resources
for
the
Future,
Inc.
,
1974.

Malinvaud,
E
.
,
1972,
Lectures
in
Macroeconomic
Theory,
Amsterdam:
North
H
o
l
l
a
n
d
,
1
9
7
2
.
 
 
McConnell,
Kenneth
E.
,
and
Ivan
Strand,
1981,
"
Measuring
the
Cost
of
Tim@
in
Recreation
Demand
Analysis;
An
Application
to
Sport
Fishing,
"
American
Journal
of
Agricultural
Economics,
Vol.
63,
February
1981,
 
~
56
.

9­
7
.

McConnell,
Kenneth
E.,
and
Jon
G.
Sutinen,
1983,
"
A
Conceptual
Analysis
of
j
Congested
Recreation
Sites,
"
in
V.
Kerry
Smith,
ed.
,
Advances
&
i
Ap
plied
Micro­
Economics,
Greenwich,
Connecticut:
J
A
I
P
r
e
s
s
,
f
o
r
t
h
­
coming,
1983.

McFadden,
Daniel,
1974,
"
Conditional
Logit
Analysis
of
Qualitative
Choice
Behavior,
"
in
P.
Zarembka,
ed.
,
Frontiers
~
Econometrics,
New
York:
Academic
Press,
1974.

McFadden,
Daniel,
1981,
"
Econometric
Models
of
Probabilistic
Choice,
"
in
Charles
F.
Manski
and
Daniel
McFadden,
eds.
,
Structural
Analysis
~
Discrete
Data
with
Econometric
Ap
placations,
Cambridge,
Massachusetts:
 
 
MIT
Press,
1981.

McKenzie,
G.
W.
,
and
I
.
F.
Pearce,
1982,
"
Welfare
Measurement­­
A
Synthesis
"
American
Economic
Review,
vol.
72,
No.
4,
September
1982,
pp.
669­
82.

Miller,
Jon
R.
,
and
Frederic
C.
Menz,
1979,
"
Some
Economic
Considerations
for
Wildlife
Preservation,
"
Southern
Economic
Journal,
Vol.
45,
January
1979,
pp.
718­
29.

Mitchell,
Robert
Cameron,
and
Richard
T.
Carson,
1981,
An
Experiment
in
Determining
Willingness
~
~
~
National
Water
Qualiv
Improvement;
d
r
a
f
t
r
e
p
o
r
t
p
r
e
p
a
r
e
d
f
o
r
U
.
S
.
Environmental
Protection
Agency,
Resources
for
the
Future,
Inc.
,
Washington,
D.
C.
,
June
1981.

Mitchell,
Robert
Cameron,
and
Richard
T.
Carson,
1982,
"
Comment
on
Option
Value:
Empirical
Evidence
from
a
Case
Study
of
Recreation
and
Water
Quality,
"
unpublished
paper,
Resources
for
the
Future,
Inc.
,
Washington,
D.
C.,
1982.

Morey,
Edward
R.
,
1981,
"
The
Demand
for
Site­
Specific
Recreational
Activities:
A
Characteristics
Approach,
"
Journal
~
Environmental
Economics
and
Management,
Vol.
8,
December
1981,
pp.
345­
71.

Muellbauer,
J
o
h
n
,
1
9
7
4
,
"
Household
Production
Theory,
Quality
and
the
`
Hedonic
Technique,'"
American
Economic
Review,
Vol.
64,
December
1974,
pp.
977­
94.

Nielsen,
Larry,
1980,
"
Water
Quality
Criteria
and
Angler
Preference
for
important
Recreational
Fishes,
"
EPA
Benefits
Project
Recreation
Working
Paper
3,
unpublished,
Washington,
D.
C.
,
Resources
for
the
Future,
Inc.
,
1980.

Olsen,
Randall
J.
,
the
Method
of
1980,
"
Approximating
a
Truncated
Normal
Moments,
"
Econometrics,
Vol.
48,
July
1980,

9
­
8
Regression
With
pp.
1099­
106.
I
 
­
.
 
­
 
 
page,
Talbot,
Robert
Harris,
and
Judith
Bruser,
1981,
"
waterborne
Carcino.
gens:
An
Economist's
View,
"
in
Robert
W.
Crandall
and
Lester
B.
Lave,
eds.
,
~
Scientific
Basis
of
Health
and
Safety
Regulation,
Washington,
 
.
D.
C.:
The
Brookings
lnstit~
tion,
1981,
pp.
197­
228.

p
a
g
e
,
Talbot,
Robert
Harris,
and
Judith
Bruser,
1982,
"
An
Economists
View
of
Waterborne
Carcinogen
s,"
in
R.
W
.
Crandall
and
L.
B.
Lave,
eds.
,
~
Scientific
Basis
of
Health
and
Safety
Regulation,
Washington,
D.
C.
:
 
.
The
Brookings
Institution,
1982.

pollak,
Robert
A.,
1969,
"
Conditional
Demand
Functions
and
Consumption
Theory,
"
Quarterly
J
o
u
r
n
a
l
~
pp.
60­
78.

pollak,
Robert
A.
,
1971,
"
Conditional
of
Separable
Utility,
"
Southern
pp.
423­
33.

Pollak,
Robert
A.
,
and
M.
L.
Wachter,
Economics,
Vol.
83,
February
1969,

Demand
Functions
and
the
Implications
Economic
Journal,
Vol.
37,
April
1971,

1975,
"
The
Relevance
of
the
Household
Production
Function
and
Its
Implications
for
the
Allocation
of
Time,
"
Journal
of
Political
Economy,
Vol.
83,
April
1975,
pp.
255­
77.
 
.

porter,
Richard
D.
,
1973,
"
on
the
Use
of
Survey
Sample
Weights
in
the
Linear
Model,
"
Annals
of
Economic
and
Social
Measurement,
Vol
.
2,
February
1973,
p~­
5~
 
 
Rae,

Rae,

Rae,
Douglas
A.
,
1981a,
Visibility
Impairment
at
Mesa
Verde
National
Park:
An
Analysis
~
Benefits
and
Costs
of
ControIlinQ
.
 
 
Emissions
in
the
Four
=
rners
Area,
.
 
 
 
 
prepared
for
the
Elect~
c
Power
Research
institute,
Charles
River
Associates,
Boston,
Massachusetts,
1981.

Douglas
A.,
1981b,
Benefits
of
Improving
Visibility
~
Great
Smoky
National
Park,
draft
report
pre~
ared
for
Electric
Power
Research
I
nsti
­
tute,
Cha=
River
Associates,
Boston,
Massachusetts,
December
1981.

Douglas
A.
,
1982,
Benefits
of
Visual
Air
Quality
@
Cincinnati,
prepared
for
the
Electric
Power
Research
Institute,
Charles
River
Associates,
.
 
.

Boston,
Massachusetts,
1982.

Rand
McNally
and
Company,
1978,
Standard
Highway
McNally:
!!!!@$@
­
f
Rand
Chicago,
1978.

Randall,
Alan,
Orlen
Grunewaldf
Angeles
Pagoulatos,
Richard
Ausness,
and
sue
Johnson,
1978,
"
Reclaiming
Coal
Surface
Mines
in
Central
Appalachia:
A
Case
Study
of
the
Benefits
and
Costs,
"
Land
Economics,
Vol.
54,
No.
4,
November
1978,
pp.
472­
89.

Randall,
Alan,
John
P.
Hoehn,
and
George
S.
Tolley,
1981,
"
The
Structure
of
Contingent
Markets:
Some
Results
of
a
Recent
Experiment,
"
paper
presented
at
the
American
Economic
Association
Annual
Meeting,
Washington
D.
C.,
1981.

9­
9
\
Randall,
Alan,
Berry
Ives,
and
Clyde
Eastman,
1974,
"
Bidding
Games
fop
Valuation
of
Aesthetic
Environmental
Improvements,
"
Journal_
of
Enviro~
l
mental
Economics
and
Management,
Vol.
1,
1974,
pp.
132­
49.
 
\

~
Randall,
Alan,
and
John
R.
Stoll,
`
1980,
"
Consumer's
Surplus
in
Commodity
Space,
"
American
Economic
Review,
Vol.
70,
June
1980,
pp.
449­
55.

Raven
scraft
r
David
J.
,
and
John
F.
Dwyer,
1978,
"
Reflecting
Site
Attractive.
ness
in
Travel
Cost­
Based
Models
for
Recreation
Benefit
Estimation,
ll
Forestry
Research
Report
78­
6,
Department
of
Forestry,
University
of
Illinois
at
Urbana­
Champaign,
Urbana,
Illinois,
July
1978.

Rosen,
Sherwin,
1974,
"
Hedonic
Prices
ferentiation
in
Perfect
Competition,
"
January/
February
1974,
pp.
34­
55.

Rowe,
Robert
D.,
and
L.
G.
Chestnut,
and
Implicit
Markets:
Product
Dif
­
Journal
~
Political
Economy,
Vol.
82,

1981,
Visibility
Benefits
Assessment'
Guidebook,
prepared
for
U.
S.
Environmental
Protection
Agency,
Abt
West,
Denver,
Colorado,
March
1981.

Rowe,
Robert
D.
,
Ralph
C.
d'Arge,
and
David
S.
Brookshire,
1980,
"
An
Experiment
on
the
Economic
Value
of
Visibility,
"
Journal
of
Environmental
Economics
and
Management,
Vol.
7,
March
1980;
pp.
l­
19.
 
Samuel
son,
Paul,
1954,
"
The
Pure
Theory
of
Public
Expend
iture,
"
Review
of
Economics
and
Statistics,
Vol.
36,
1954,
pp.
387­
89.
 
Saxonhouse,
Gary
R.
,
1977,
"
Regression
from
Samples
Having
Different
Characteristics,
"
Review
~
Economics
and
Statistics,
Vol.
59,
May
1977,
pp.
234­
37.

Schmalensee,
R
.
,
1
9
7
2
,
"
Option
Demand
and
Consumer
Surplus:
Valuing
Price
Changes
Under
Uncertainty,
"
American
Economic
Review,
Vol.
62,
December
1972,
pp.
813­
24.

Schmalensee,
R.
,
1975,
"
Option
Demand
and
Consumer
Surplus:
Reply,
"
American
Economic
Review,
Vol.
65,
September
1975,
pp.
737­
39.

Schmidt,
P.
,
1977,
"
Estimation
of
Seemingly
Unrelated
Regressions
With
Un
­
equal
Numbers
of
Observations,
"
Journal
q
Econometrics,
Vol.
5,
May
1977,
pp.
365­
78.

Schulze,
W
.
D
.
,
D.
S.
Brookshire,
E.
G.
Walter,
and
K.
Kelley,
1
9
8
1
,
T
h
e
Benefits
g
Preserving
Visibility
in
the
National
Parklands
of
the
Soufi
west,
Volume
8
of
Methods
Development
for
Environmental
Control
Bene­
 
 
 
.
 
~
Assessment,
prepared
for
U.
S.
Envir=
mental
Protection
Agency,
Re­
=
rce
and
Environmental
Economics
Laboratory,
University
of
Wyoming,
Laramie,
Wyoming,
1981.

.
Schulze,
W
.
D
.
,
R
.
C
.
d'Arge,
and
D.
S.
Brookshire,
1981,
"
Valuing
Environmental
Commodities:
Some
Recent
Experiments,
"
Land
Economics,
Vol.
57,
No.
2,
May
1981,
pp.
151­
73.

9­
1o
r,.

cheDhard~
Ronald
W.,
1953,
Cost
and
Production
Functions,
Princeton:
 
 
"
"­'
princeton
University
Press,
1953.

Smith,
V.
Kerry,
1975a,
"
The
Estimation
and
Use
of
Models
of
the
Demand
for
Outdoor
Recreation,
"
in
Assessing
the
Demand
for
Outdoor
Recreation,
Washington,
D.
C.
:
National
Academy=
Sciences,
=
75.

Smith,
V
.
K
e
r
r
y
,
1975b,
"
Travel
Cost
Demand
Models
for
Wilderness
Recreation
A
Problem
of
Non­
Nested
Hypotheses,
"
Land
Economics,
Vol.
51,
May
1975,
pp.
103­
1
1
.

smith,
V.
Kerry,
1983,
"
Option
Value:
A
Conceptual
Overview,
"
Southern
Economic
Journal,
Vol.
49,
January
1983,
pp.
654­
68.

Smith,
V.
Kerry,
William
H.
Desvousges,
and
Matthew
P.
"
The
Opportunity
Cost
of
Travel
Time
in
Recreation
Land
Economics,
forthcoming,
August
1983.
McGivney,
1983,
Demand
Model
s,"

smith,
V.
Kerry,
and
Raymond
J.
Kopp,
1980,
"
The
Spatial
Limits
of
the
Travel
Cost
Recreation
Demand
Model,
"
Land
Economics,
Vol.
56,
February
1980,
pp.
6
4
­
7
2
.

Smith,
V.
Kerry,
and
J.
V
.
Krutilla,
1
9
8
2
,
"
Toward
Formulating
the
Role
of
National
Resources
in
Economic
Model
s,"
in
V.
K.
Smith
and
J.
V.
Krutilla,
eds.
,
Explorations
in
Natural
Resource
Economics,
Baltimore:
Johns
Hopkins,
1982,
pp.
1­
43:

Smith,
V.
Kerry,
and
D.
Waldman,
1982,
"
A
Comparison
of
Ordered
Logit
and
Ordered
Probit:
Some
Monte
Carlo
Experiment
Results,
"
unpublished
manuscript,
Department
of
Economics,
University
of
North
Carolina
at
Chapel
Hill,
Chapel
Hill,
North
Carolina,
1982.

Takayama,
Akira,
1982,
"
On
Consumer's
Surplus,
"
Economic
1982,
pp.
35­
42.

Talhelm,
Daniel
R.
,
1978,
"
A
General
Theory
of
Supply
Outdoor
Recreation
in
Recreation
Systems,
"
unpublished
manuscript,
Department
of
Agricultural
Economics,
Michigan
State
University,
East
Lansing,
Michigan,
July
1,
1978.
Letters,
Vol.
10,

and
Demand
for
Thayer,
Mark
A.
,
1981,
"
Contingent
Valuation
Techniques
for
Assessing
Environmental
Impacts:
Further
Evidence,
"
Journal
~
Environmental
Economics
and
Management,
Vol.
8,
1981,
pp.
27­
44.

Thayer,
Mark
A.
,
and
W.
Schulze,
1977,
"
Valuing
Environmental
Quality:
A
Contingent
Substitution
and
Expenditure
Approach,
"
unpublished
paper,
Department
of
Economics,
University
of
Southern
California,
Los
Angeles,
California,
1977.

Theil,
Henri,
1961,
Economic
Forecasts
and
Policy,
Amsterdam:
North
Holland,
1961.

9­
11
U
s
.

U
s
.

U
s
.

U
s
.
Bureau
Summary
Bureau
Abstract
of
the
Census,
Department
Tape,
File
A,
1970.

of
the
Census,
Department
of
the
United
States:
1976,
"
Washington,
D.
C,
1976.
of
Commerce,
1970,
F
of
Commerce,
1976,
U.
S.
Government
Print
Bureau
of
the
Census,
Department
of
Commerce,
1982,
1980
r
s
t
Count
`
Statistical
ng
Office,

Census
of
the
Population
and
Housinq",
preliminary
data
tape,
Was~
ton,
D.
C:
m2
.

Environmental
Protection
Agency,
Office
of
Water
Regulations
and
Standards
1982,
Handbook,
Water
Quality
Standards,
d;
aft,
U.
S.
Environmental
Protection
Agency,
Washington,
D.
C.
,
October
1982.

Varian,
Hal
R.
,
1978,
Macroeconomic
Analysis,
New
York:
W.
W.
`
Norton
and
Co.
,
1978.

Vaughan,
W.
J.,
and
C.
S.
Russell,
1982,
Freshwater
Recreational
Fishing:
T=
National
Benefits
of
Water
Pollution
Control,
Washington,
D.
C.
:
Resources
for
the
Future~
lnc.
,
November
1982.

von
Neumann,
J.
,
and
O.
Morgenstern,
1947,
The
Theory
of
Games
and
Economic
Behavior,
2nd
Edition,
Princeton:
~
nceton
Uni~
ersity
pre~
1947.

Walsh,
R.
G.,
D
.
G
.
Greenley,
R.
A.
Young,
J.
R.
McKean,
and
A.
A.
Prato,
1978,
Option
Values,
Preservation
Values
and
Recreational
Benefits
of
Improved
Water
Quality:
 
 
A
Case
Study
of
the
South
Platte
River
=
si
ni
Colorado,
 
 
EPA­
600/
5­
78­
001
,
U.
S.
 
 
 
E;
v=
nmenta
I
Protection
Agency,
Office
of
Research
and
Development,
January
1978.

Water
Resources
Counci
1,
1979,
"
Procedures
for
Evaluation
of
National
Economic
Development
(
NED)
Benefits
and
Costs
in
Water
Resources
Plannina
(
Level
C),
Final
Rule,
l'
Federal
Register,
Vol.
44,
No.
242,
December
14~
1979,
pp.
72892­
977.

Willig,
Robert
D.
,
1976,
"
Consumer's
Surplus
Without
Apology,
"
American
Economic
Review,
Vol.
66,
September
1976,
pp.
587­
97.

Wilman,
Elizabeth
A.
,
1980,
"
The
Value
of
Time
in
Recreation
Benefit
Studies,
"
Journal
~
Environmental
Economics
and
Management,
Vol.
7,
September
1980,
pp.
272­
86.

Zellner,
A
.
,
1
9
6
2
,
"
An
Efficient
Method
of
Estimating
Seemingly
Unrelated
Regressions
and
Tests
for
Aggregation
Bias,
"
Journal
of
the
American
Statistical
Association,
Vol.
57,
June
1962,
pp.
348­
68.
 
 
Ziemer,
R.
,
W.
N.
Musser,
and
R
.
C.
Hill,
1980,
"
Recreational
Demand
Equations:
Functional
Form
and
Consumer
Surplus,
"
American
Journal
of
Agricultural
Economics,
Vol.
62,
February
1980,
pp.
136­
41.
 
9­
12
APPENDIX
A
SAMPLE
DESIGN
This
aPPendix
provides
a
justification
for
the
~
ling
protocol
empioyed
in
the
project.

~.
1
.5
AMPLE
SIZE
JUSTIFICATION
sampling
sizes
and
the
sam
­

One
aPProach
for
usin9
the
survey
information
requires
that
many
of
the
~
arameters
to
be
estimated
in
this
study
be
treated
as
proportions­­
for
example
the
proportion
of
adults
who
participate
in
water­
related
recreation
activities
Accordingly,
the
proposed
sample
sizes
were
determined
by
computing
the
sample
size
required
to
estimate
proportions
of
the
underlying
population
(
i.
e.,
households
in
the
Monongahela
River
basin).

The
required
sample
size
depends
upon
the
desired
precision
of
the
proportion
estimates.
The
sample
size
required
to
produce
an
estimate,
~,
within
8
units
of
the
true
population
proportion,
p,
with
u
percent
certainty
depends
Upon
51
PI
and
a.
Obviously,
it
is
desirable
to
make
6
small
and
a
large.
l+
owever,
decreasing
6
and
increasing
o!
each
requires
an
increase
in
the
required
sample
size.
Additionally,
a
6
value
considered
precise
for
large
p
values
is
not
necessarily
precise
for
small
p
values.
For
example,
let
6
=
0.10,
b~
=
0.85,
and
~
z
=
0
.
0
5
.
T
h
e
n
,
@
l
t
6
is
equal
to
0.85
t
0.10,
which
is
relatively
precise.
However,
i+
*
6,
which
is
equal
to
0.05
f
0.10,
is
not
very
precise.

Table
A­
1
shows
the
sample
sizes
needed
to
detect
a
specific
difference
with
power
1
­
~.
The
crucial
specific
differences
for
this
project
were
those
in
estimated
values
for
the
willingness
to
pay
for
different
levels
of
water
quality
and
differences
in
estimates
of
option
and
existence
values
for
the
Monongahela
River.

An
example
using
estimated
coefficients
of
variation
(
which
are
equal
to
t
h
e
standard
error
of
the
estimate
divided
by
the
mean
e
s
t
i
m
a
t
e
,
o
r
s
i
m
p
l
y
a
Method
of
comparing
the
variation
in
the
measured
benefits)
from
related
studieS
shown
in
Table
A­
2,
will
explain
Table
A­
1
.
If
the
coefficient
of
variation
is
equal
to
0.2
(
as
was
the
case
in
the
Walsh
et
al.
[
1978]
South
Platte
River
Basin
Study
for
Denver
residents'
willingness
to
pay
for
existence
values),
a
sample
size
of
68
is
necessary
to
detect
a
10
percent
difference
in
the
mean
value
with
95
percent
confidence
that
the
difference
is
different
from
zero
and
a
l
o
­
p
e
r
c
e
n
t
chance
of
not
rejecting
the
null
hypothesis
(
A
=
O
)
w
h
e
n
it
iS
false.
If
there
is
little
or
no
variation
in
the
estimates,
small
differences
can
be
detected
with
minimal
sample
size.
However,
considerable
variation
in
esti
­
Mated
values
will
mean
that
the
sample
size
at
384
may
not
be
able
to
detect
small
differences
in
the
estimates.
Thus,
when
proportions
are
estimated,

A­
1
Table
A­
1.
Sample
Sizes
Needed
to
Detect
a
Specified
Difference
With
Power
1
­
P
Cv
=
coefficient
of
variation
(
ue/
pc)
a
Detection
level
(
A)
0.1
0
.
2
0
.
3
0
.
4
0
.
5
0.06
PC
0.08
pc
0.10
p
0.15
p:

0.20
pc
0.25
pc
0.06
pc
0.08
PC
0.10
pc
0.15
pc
0.20
pc
0.25
pe
(
a)
a
=
T
y
p
e
I
error
=
0.05,
p
=
Type
II
error
=
().
1
48
190
428
760
1,189
27
107
241
428
669
17
68
154
274
428
8
30
68
122
190
4
17
39
68
107
3
11
25
44
68
(
b)
a
=
0.05,
~
=
0.25
30
120
269
478
748
17
67
151
269
421
11
43
97
192
269
5
19
43
77
120
3
11
24
43
67
2
7
16
28
43
a
a
is
the
common
standard
deviation
for
both
the
treatment
and
control
e
responses
under
the
model,
and
p­
is
the
mean
response
(
usage
level)
for
the
control.
The
sample
size
is
c~
lculated
as
n
=
2(
CV/
A)
2(
a~
~
+
a
z
1­$)
2'
where
z
is
the
standard
normal
variate.

relative
precision
is
often
considered
as
the
most
appropriate
basis
for
determining
sample
size.
This
is
accomplished
by
requiring
that
~
lie
within
pa
units
of
the
true
p
value
with
a
percent
certainty
for
smallest
proportion
of
interest.
In
the
above
example,
the
estimate
of
the
small
p
value
would
change
from
0.05
*
0.10
to
0.05
f
0.005,
which
is
a
much
more
precise
estimate
Obviously,
this
method
significantly
increases
the
required
sample
sizes
for
small
p
values.

Table
A­
3
contains
minimum
sample
sizes
for
~
to
be
within
p6
units
of
p
with
95
percent
certainty
(
in
the
sense
of
repeated
sampling)
for
various
values
of
p
and
8,
assuming
simple
random
sampling.
The
p
values
to
be
estimated
in
the
study
are
unknown
and
will
probably
vary
considerably
from
one
activity
to
another.
Therefore,
it
is
impossible
to
determine
exactly
the
appropriate
sample
size.
Based
on
past
work
it
is
reasonable
to
assume
t
h
a
t
A
­
2
7
j>
f
Id
i
­
?
s
P
JS
3e
]
m
he
at
A
.
.
..­
Table
A­
2.
Coefficients
of
Variation
for
Selected
Benefits
Estimates
/
Study
la
Study
2
b
Study
3=
/
Measured
Measured
Measured
benefit
Cv
q
benefit
Cv
q
benefit
Cv
rl
Beatable
0.05
748
Existence
0.20
88
Aesthetic
and
0.38
10
water
value
health
quality
(
user)

Fishable
0.05
748
Existence
0.33
88
Aesthetic
and
0.34
10
Water
value
health
quality
(
user)

swimmable
0.0s
748
Existence
0.63
15
Aesthetic
and
0.43
9
water
value
health
quality
(
nonuser)

Bequest
0.93
15
Aesthetic
and
0.05
7
value
health
(
nonuser)

Aesthetic
and
0.61
8
health
 
 
~
See
Mitchell
and
Carson
[
1981].
 
See
Walsh
et
al.
[
1978].
c
See
Brookshire
et
al.
[
1979].

Table
A­
3.
Required
Sample
Size
for
Estimates
of
p
to
be
Within
p6
Units
of
p,
Assuming
Simple
Random
Sampling
6
P
0.05
0.10
0.15
0.20
0.25
0.01
152,127
38,032
16,903
9,508
6,085
0.05
29,196
7,299
3,244
1,825
1,168
0.10
13,830
3,457
1,537
864
553
0.25
4,610
1,152
512
288
184
0.35
2,854
713
317
178
114
0.40
2,305
576
256
144
92
0.50
1,537
384
171
96
61
0.75
512
129
57
33
21
0.95
81
21
9
6
4
A
­
3
most
p
values
will
be
in
the
range
of
0.35
to
0.40
or
higher.
Ditton
and
Goodale
[
1973]
found
that
69.2
percent
of
the
residents
in
the
Green
Bay,
Wisconsin,
area
had
engaged
in
water­
related
outdoor
recreation
within
the
last
year.
The
1977
outdoor
recreation
survey
conducted
by
the
Department
of
the
Interior
determined
that,
with
this
assumption,
a
reasonably
precise
estimate
can
be
formed
by
requiring
that
5
=
0.20
(
i.
e.
,
p6
=
(
0.35)(
0.20)
=
0.07
or
pa
=
(
0.40)(
0.20)
=
0.08).
These
values
of
p
and
6
produce
a
required
sample
size
in
the
range
of
144
to
178.
These
estimates
are
based
on
simple
random
sampling
and
need
to
be
increased
because
of
the
effects
of
a
cluster
sample
design.
That
is,
the
area
sampling
design
requires
expansion
of
the
recommended
sample
size.
The
recommended
sample
size
also
assumed
a
20­
percent
nonresponse
rate.
It
should
be
recognized
that
the
proposed
sample
size
will
give
less
precise
estimates
for
p
values
below
the
0.35
to
0.40
range
and
more
precise
estimates
for
p
values
above
the
range.
Since
the
coefficients
of
variations
for
p
shown
in
Table
A­
1
are
approximately
one
and
one­
half
times
larger
than
the
coefficients
of
variations
in
Table
A­
2,
the
recommended
sample
size
should
yield
adequate
power
for
detecting
differences
in
the
willingness
to
pay
and
option
and
existence
values.

A.
2
SAMPLING
PROTOCOL
Using
1970
census
computer
data
tapes
(
more
up­
to­
date
data
were
not
available
at
the
time
of
the
study
since
the
1980
census
computer
data
tapes
had
not
been
released)
for
Enumeration
Districts
and
Block
Groups
(
ED/
BGs),
noncompact
clusters
of
approximately
seven
households
were
constructed.
The
1970
data
were
adjusted
by
county
using
preliminary
1980
census
data
to
more
accurately
reflect
the
present.
Additionally,
the
1970
occupancy
rate
and
the
estimated
response
rate
were
taken
into
account
in
determining
the
cluster
size.

The
clusters
were
constructed
into
three
groups
once
they
were
stratified.
The
groups
are
those
households
in
(
1)
Pittsburgh,
(
2)
a
place
other
than
Pittsburgh,
and
(
3)
not
in
a
place.
Fifty­
one
clusters
were
selected.
The
number
of
clusters
selected
from
each
stratum
were
proportional
to
the
number
of
households
in
that
stratum.
For
example
,
since
61
percent
of
the
households
in
the
five­
county
area
are
located
in
Pittsburgh,
51(
0.61
)
=
31
clusters
were
selected
from
Pittsburgh.
The
clusters
were
selected
with
equal
probabilities
within
each
stratum.
Because
of
the
proportional
allocation
of
the
sample
to
the
strata,
the
probabilities
of
selection
for
all
clusters
were
nearly
equal.

Because
the
clusters
were
contained
in
ED/
BGs,
the
general
physical
location
of
the
cluster
is
known.
Interviewers
were
sent
to
the
field
to
count
and
list
all
households
in
the
ED/
BGs
that
contain
the
selected
clusters.
The
lists
produced
during
the
counting
and
listing
exercise
were
used
to
identify
the
specific
households
in
the
selected
cluster.
If
the
number
of
households
did
not
exceed
a
predetermined
number,
all
households
in
the
cluster
were
contacted
For
those
clusters
that
were
too
large,
the
list
was
used
to
determine
a
subsample
of
the
cluster
to
be
contacted.

Once
the
households
to
be
contacted
were
identified,
the
interviewers
conducted
a
preliminary
visit
and
compiled
a
roster
of
all
adults
living
in
the
household.
One
of
the
adults
was
randomly
selected
(
with
equal
probabilities)
for
interview.

A
­
4
APPENDIX
B
SURVEY
FORMS
AND
PROCEDURES
PART
1
HOUSEHOLD
CONTROL
FORM
Part
1
of
this
appendix
contains
the
household
control
form
used
by
field
interviewers
to
provide
assignment
and
other
background
information.

B­
1
 
.
 
­.
 
 
Estim.
lling
Rf,
lw[
ils
0[
hlrc
N'
11
I'rojccl
No.
41
U­
2222­
2
I
.
ASS
ltiNllllN'
1
I
NFfJNNA'
1
10}
1
II
CD
IL
Iv.
.
 
r@
iIl
iLy
lNNfStHIOLII
CONI'NOL
Form
No.
01
Otlll
No.
2000­
0381
AIIIIrOVAl
Expires:
!)/
3(
J/
8z
FOlttl
k.
Atld
I
(>
SS
(
NI,
UI~
I;;/
S~.
Zj%
i%~
 
 
 
.
 
_
_
..
 
 
 
 
 
 
­
(
Apartmcnt
No.
)
 
(
Ci
Ly)
­
 
 
.
.
.
­
 
L...
l_
l_
1.
­
1
.1.
 
­
J
`.,
 
[
.­.
l..,
_,

(
stilt'?)
(
Zill)

Ntxxll{
ll
I)
F
coNrA(
xi
­
IINIINMMTION
ANI)
SAtN'
1.
E
INNIVINUAL
l
i
l
y
of
W(.(.
ti
.
­­­

 
 
 
.
 
 
­
 
nil
1
c
.
 
.
 
 
.
.
 
 
 
.
­

.
 
­
­
 
 
`
Tism
­.­
 
`
1
­
i:.
am
­.
.
 
1!!
am
py
#
am
 
1!!!!
am
 
P!!
;
1
m
.
 
­
­­
l!?
aul
.4?­.
<
a
ml
­.
.
.._.
_
I!!!
 
­
.
.
 
 
 
 
 
F
 
 
 
Rcsul
t
Notes
code
 
.
 
 
 
 
I
 
.
 
 
 
 
 
­
 
­
l
 
 
 
 
 
.
.

I
 
 
I
 
 
.
 
 
 
.

I
.
..­
 
.
­.
 
_
­_
L
 
­
­­
FI
­
.
 
 
­
.

 
 
­
 
.
.
 
­
 
1[
1.
CONTAC1'
RESULT
CODES
(
CIRCLE
Ok'
CONTACT)

Nouschold
Enumeration
Con
Lact
.
 
 
codes
01
02
03
04
05
i@
6
*(
J
1
08
Enumeration
Lhnpleted
No
Rnumr?
ration
Eligible
a
l
]
Ioum
F.
nurneration
Respondent
Ureakoff;
ParLial
I)
ala
Enumeration
Respondent
Refused
Language
flarrier
VacanL
Rousing
Unit
Not
a
Rousing
Unil;
e.
g.
,
tiergcd,
Dem~
lisho~
l,
­

Group
QtlurLccs,
NonffesidenLia
L
BELOW
`
f'Nli
k'l
NAI.
RESULT
COINi
F(
JR
EACH
TYI'K
~~@
e
I
n
d
i
v
i
d
u
a
l
Con
Lacl
Codos
 
.
 
­
 
.
­
 
 
.
 
 
.
.
 
.­..

O
t
h
e
r
(
EXI'I.
AIN
IN
"
C(
JINIIIN'IS")
20
21
22
23
24
25
26
InLerview
Complc
Le4
(
CIRCLE
VIN!;
1ON
AIININISIENIIU
IN
SI;
C'IION
VI.
N)
Appointment
N:
JIIC
lnlervicw
Rreakoff;
P;)
rlial
Ihta
Sowple
]
odividw.,
1
nol
Rumt:
Rcfusd]
LanKuage
lhrricr
Other
(
EXPI.
AIN
lN
"
CONNLN'
13")

(
21­
22)

(
18­
19)
,.,.
So~
lrc{,
01
Illformiltion
f
o
r
Resul
L
Codes
0
6
,
07
 
 
 
­
 
 
 
N.
lmr
`
Nl$,,
ll,,.
r/
ktr,.(.
t/
l{
Fl)
 
 
 
.
.
 
 
 
 
 
­
 
 
 
 
 
.
 
 
 
 
lLily/
sl.
ilc/
zlp
`­

(
)
14
1.,.
1,..,,.
N,
nd,,
(
.
 
­
 
.
.
.
_
 
.
 
­
 
.
 
­
.
 
 
V
.
COtWNIS
.
 
­
 
V1.
llOUStJIOLfJ
ENUNERATION
ANI)
SANPLK
INDIVIDUAL
SELECTION
Nello,
['
n
(
tJAtlE)
With
the
R
e
s
e
a
r
c
h
T
r
i
a
n
g
l
e
Institule
of
North
Carolina.
We
are
doing
a
household
survey
for
a
government
agency
to
study
levels
of
wtiter
qUtiii
Ly
andsorue
outdnor
recrerr
Lional
a
c
t
i
v
i
t
i
e
s
people
take
part
in
krolh
near
and
on
pouds,
lakes,
sLreauw.
,
and
r
i
v
e
r
s
in
Llm
Pittsburgh
area.
Your
household
has
been
randomly
aelec
Led
ulorig
wiLh
other's
in
this
area
to
he
interviewed.
I
n
order
to
dcterurine
who
in
your
Immehold
shoul
II
be
i
n
t
e
r
v
i
e
w
e
d
,
I
would
like
to
ask
a
few
q~
estious
shout
the
reaidenta
of
your
household.
I
am
required
to
talk
with
a
household
urewber
who
is
16
years
of
age
or
older.
(
ASK
IF
NECESSARY,
ARS
YOU
16
YEARS
ON
OMER?)

1.

2.

m
&
`
i"
4.

5.

6
.

7.

u
t'irtit,
are
Lhere
any
occup!
ed
or
vacant
living
quarters
other
than
your
own
(
FOR
SINGLE
UNl"
f
STRUCTURE)
in
this
sLrwcture
or
on
this
property?
(
FOR
tkULTI­
UNIT
STRUCTURES?)
in
this
unit?

(
CIRCLE
NUtlUER
BELOW
FOR
RESPONSE)

1
m
(
A
D
D
T
O
L
I
S
T
OF
ADDED
IIOUSING
UNITS
mftmyfmfm
ur
tmstmm
tfuLEs)

2
NO
N
O
W,
I
would
like
to
ask
some
general
questions
about
you
and
all
of
Lhe
oLher
people
WIIO
live
in
Lhis
h
o
u
s
e
h
o
l
d
,
including
friends
and
roomers.
Let's
list
the
people
who
live
here
iu
order
o
f
a
g
e
,
b
e
g
i
n
n
i
n
g
w
i
t
h
Lhe
o
l
d
e
s
t
f
i
r
s
t
.
(
ENf'ER
AGES
IN
DkX$
CENDING
AGE
ORDER
IN
COLUIIN
B.)
r
I,
dve
ltsted
ages
for
persona
who
are
(
RltAfJ
AGES).
Is
there
anyone
e
l
s
e
liviwg
Ilrre
n
o
w
?
(
It'
YES,
ENJ'ER
A
G
E
(
S
)
A
N
D
CONtfECT
AGE
ORDER
IN
COLUtfN
U,
IF
NECESSARY
)

ASK
TIIE
SEX
FOR
EACN
PERSON
LISTED
AND
CIRCLE
TNE
CORRECT
CATEGORY
IN
COLUMN
C.

Which
person
ia
Lhe
head
of
the
household?
OJRJTE
TIIE
WORD
"
UEADI'
I
N
coLytiN
D
F
O
R
TNE
LJNE
NUNflER
OF
TIIE
t'tM30N
CONSIDERED
TNE
READ
OF
UOUSEROLD.
)

FOR
OMER
PERSONS
LISTED
ASK
TIJSIR
RELAT'IONSIIJP
TO
TRE
NEAD
OF
UOUSEI[
OLJ3
ANJI
ENTER
IN
Ilouwllol.
11
ROS'I'I:
N
A
D
c
 
 
_
­­...
!!
­­­
.
.
SEX
M
­
­
F
AGE.
12
IIOUSEIIOI.
11
rlllAl)/
ltl:
l.
Ar
lf)
li>
llll'
 
 
.
 
.
 
 
­­
 
.
­­.­­
 
­
 
 
 
.­­.
 
­
 
­
.

0
1
._
_
__
_
_
_
_
_
_
_
_
_
_
_
_
_
12
02
12
 
.
.
.
 
03
1
2
 
 
.
 
.
.
 
.

04
12
 
 
 
.
 
.
­
 
_
 
­
 
05
1
2
 
.
.
_
.
_
 
.
 
 
 
 
 
.­
 
 
 
 
 
 
 
 
 
.
.

06
1
2
 
 
.
 
 
.
 
_
 
 
.
 
 
.

01
1
2
_
.
.
.
.
.
.
.
.
.
.
.
.

OB
1
2
 
 
.
 
­..
.
.

09
1
2
..
 
 
 
.

10
1
2
 
.
 
 
.
.
 
 
.
 
._

11
12
 
­
 
 
 
.
__
 
 
 
­
.
 
.
.
.
 
­
 
 
 
.
 
26­
27)

COLUHN
D.
CARD
1.

sELEcT
mE
INNISEINmI
tIEtJtIttR
TO
BE
IN
T
E
R
V
I
E
W
E
D
FRON
AMONG
ONLY
mm
PERSONS
I
8
Y
E
A
R
S
O
R
cm.
80
=
1
OJ.
ftER
(
ELIGJ
R
L
E
tfOUStUIOLD
NENBKRS)
.
REFERRING
TO
TRE
AGES
LISTED
IN
TNE
ROSTER,
MTMN
I
NE
`
INK
NUtlJIER
OF
PERSONS
WNO
ARE
18
YEANS
OR
OLUEN
ANO
ftNAW
A
1.1
NE
ACROSS
TllE
ROSTtM
TO
SkltIARA"
rE
`
rNOSE
PERSONS
tlfOtt
TIIOSE
17
OR
YOIJNfXJf
.
LOCATE
ON
TIIE
TAfJJ.
E
JWLOW
4HIE
ROSTER
TNE
NUtlNEK
Ot'
ELJGIBLE
UOUSEUOLf)'
HENBEKS
.
f)
JRliCrLY
BELOW
TIIE
NUNUER
OF
NOUSENOLJ)
HOUSEHOLD
SIZE:

tlKtlDKtfs
,
FIND
"
rllE
NOSTJiK
LINE
NUINNM
SELECTED.
CIKCLE
TIIE
StlLJiCJ'tXk
LJNE
NUNONR
ON
TUE
!
2
`
4
b
~
7
e
`(
1,11
ttOSTEK
.

(
YO
U
h
a
v
e
/
~
~
~
$
~
N
hua)
been
selcctcct
as
the
person
to
be
inLerviewcd.
(
ASK
FOR
TIIE
tJANli
ll1443q5
}:
1
a
OF
TkfK
PERSOII
SELECTED
AND
ENt'LR
WERE)
RESPONC)[
Nl
NO.
:
 
 
..
 
.
 
.
 
 
.
.
.
..
 
 
.
.
.
 
 
 
.
.
PRJNT
NAM
OF
SELECf'ED
lNDIVlk)
UAL
J
F
ENUtttHJAT
ION
RESPONDENT
WAS
UI:
JtN
SELECTED,
A'IIENPT
TO
CONPLETE
INTERVIEW.
IF
ANOIWR
PEKSON
,
DK'JWUJ
INE
I
F
UE/
SllK
i
S
AVAI
LAIII,
E
OK
WREN
llE/
SllE
WILL
BE.

QtrSSTIOttMAIRt
Vt:
NSION
AIUll
NISTEREn
(
C
I
R
C
L
E
V
E
N
S
1
O
N
)
A
II
C
J)

(//,
~
AI,
8)
(:
Al(
l)
2
I
­
IL
1)
111'
11{(
111
CAI{
I)
1
(
XII.,
Iio
=
2
(
IV­
2))

(
24­
28)

(
29­
11)

(
M­
m)

(
)')­
4'))

(
44­
48)

(
49­
5'
1)

(:)
L­
sti)

(
59­(
1'
1)

((
d,
­(,
fr)

((
19­
7
J)
PART
2
COUNTING
AND
LISTING
EXAMPLES
­".
 
F
b.
~.
/~­
L.&.
.
L
&

Figure
B­
1.
Sample
segment
map.

B
­
4
\
t?
 
\

"\
L
@
\

­:

\

Figure
B­
2.
List
unit
sketch.

B­
5
Figure
B­
3.
List
of
housing
units.

.
B­
6
PART
3
DEBRIEFING
AGENDA
part
3
Of
t
h
i
s
appendix
contains
the
agenda
used
during
the
December
,­

,981
interviewer
debriefing
session.

Estimating
Recreation
and
Related
8enefits
of
Water
Quality
RTI
Project
2222­
2
DEBRIEFING
AGENDA
Thursday,
December
10,
1981
welcome
and
Introductions
Evaluation
of
Traininq
.
Effectiveness
of
home
study
materials
Effectiveness
of
classroom
sessions
.
Adequacy
of
training
time
­
Areas
encountered
in
interviewing
that
should
have
been
covered
in
training
Usefulness
of
specifications
l
 
nd
manual
­
Deficiencies
in
specifications
and
manual
Evaluation
of
Assignment
Materials
and
Procedures
Content
and
layout
of
Household
Control
Form
­
Accuracy
of
sample
member
assignment
data
(
names,
addresses,
etc.
)

­
Tracing/
locating
activities
required
.
Deficiencies
in
materials
and
procedures
Obtaining
Respondent
Cooperation
­
Gaining
access
to
sample
members
Explaining
Purposes
of
the
survey
.
Obtaining
Permission
to
complete
enumeration
­
Obtaining
permission
to
complete
the
interview
.
Intervention
by
other
household/
family
members
.
Effectiveness
of
"
Oear
Resident"
l
 
nd
other
informational
material
.
Characteristics
of
nonrespondents
and
reasons
for
nonresponse
.
Procedures
far
converting
refusing
sample
members
B­
7
Conducting
the
Interviews
.
Household
enumeration
procedures
and
problems
Usefulness
of
handout
materials
Deficiencies
of
handout
materials
.
Section­
by­
section
review
of
all
questionnaires
(
1)
What
questions
usually
worked
well
and
were
understood
by
all
respondents?

(
2)
what
questions
frequently
were
difficult
to
administer
or
were
misunderstood
by
respondents?

(
3)
What
questions
appeared
co
elicit
reliable
responses
with
minimai
probing?

(
4)
What
questions
frequently
yielded
"
Don't
Know"
responses?

(
5)
What
questions
were
respondents
reluctant
to
answer?
What
reasons,
if
any,
were
stated?

(
6)
What
category
of
respondents
(
i.
e.,
disabled,
widowed,
older
men,
etc.
)
had
the
most
difficulty
in
responding
to
the
questions

(
7)
What
category
of
respondents
were
most
reluctant
to
answer
certain
questions?

.
Problems
with
layout
or
design
of
each
instrument
Problems
in
the
interview
setting
Problems
with
interview
length
Questions
or
concerns
expressed
by
respondents
Administrative
Procedures
.
Status
reporting
Communications
with
supervisor/
central
office
Resolution
of
field
problems
Evaluation
of
callback
requirements
Recommendations
for
Future
Similar
Survevs
Respondent
informational
material
Assignment
materials
and
procedures
.
Contacting,
locating,
and
securing
cooperation
Instruments
and
handouts
.
Administrative
materials
and
praceaures
.

B­
8
PART
4
QUALITY
CONTROL
The
quality
control
procedures
used
the
survey
questionnaire,
including
both
dures,
are
described
below:

FIELD
EDITING
PROCEDURES
during
and
after
field
editing
and
administration
of
validation
proce
­

Field
interviewers
were
responsible
for
conducting
a
thorouqh
field
edit
of
each
completed
survey
instrument.
Interviewers
wer­
e
provided­
with
an
edit
instruction
for
the
instruments
to
insure
that
significant
edit
checks
were
made.
The
importance
of,
the
field
editing
process
and
procedures
to
be
followed
were
emphasized
in
the
interviewer's
manual
and
received
attention
as
part
of
interviewer
training.

Field
editing
by
interviewers
involved
two
steps.
First,
each
completed
instrument
was
scanned
for
completeness
at
the
conclusion
of
each
interview
while
the
interviewer
was
still
in
the
respondent's
presence.
If
any
incomplete
or
omitted
items
were
detected,
the
missing
data
were
obtained.
Second,
interviewers
thoroughly
edited
each
completed
instrument
before
submitting
their
work.
Any
omissions
or
problems
noted
during
this
edit
were
resolved
by
a
telephone
call
or,
if
necessary,
a
return
visit
to
the
respondent
by
the
interviewer
These
field
edit
procedures
were
especially
important
as
an
aid
to
insure
that
high
quality
and
complete
data
were
received
from
the
field.

To
insure
quality
control
of
the
interviewing
process,
each
interviewer's
completed
interviews
were
edited
at
the
Research
Triangle
Institute
(
RTI
)
during
the
fieldwork
period.
The
editor
used
edit
specifications
that
focus
on
the
key
elements
of
each
document,
and
interviewers
received
ongoing
assessments
of
the
quality
of
their
work
by
telephone.
In
addition,
where
graphic
instruction
to
an
interviewer
was
helpful
to
explain
the
nature
of
an
error,
photocopies
were
made
of
questionnaire
pages
to
show
interviewers
exactly
what
the
problem
was.

VALIDATION
A
major
quality
control
procedure
involved
validation
of
a
random
sample
of
10
percent
of
the
interviews
conducted.
This
procedure
was
accomplished
through
telephone
contacts
with
participating
sample
members.
The
validation
contact
was
designed
to
determine
whether
the
interview
actually
took
place
on
or
about
the
date
reported;
whether
the
interviewer
secured
a
complete,
current
household
roster;
whether
appropriate
sample
member
selection
procedures
were
followed;
and
whether
the
entire
interview
schedule
was
completed.
Also,
key
items
were
asked
and
responses
compared
with
original
responses
reported
by
the
interviewer.
I
n
addition,
the
contact
elicited
other
information
concerning
the
interviewer's
performance.

B
­
9
L
.
APPENDIX
C
SURVEY
ANALYSIS:
SUPPORTING
TABLES
This
appendix
provides
supporting
statistical
analysis
for
the
option
price,
user
valuel
and
option
value
results
presented
in
Chapters
4
and
5.

tables
in
general
The
focus
on
three
issues:
(
1)
estimates
with
outliers
excluded;
(
2)
estimates
with
protest
bids
excluded;
and
(
3)
t­
tests
for
differences
from
zero.

In
addition,
Table
C­
16
supports
the
analysis
in
Chapter
6.
This
table
shows
benefit
estimates
from
an
alternative
contingent­
ranking
specification.

c­
1
Table
C­
1
.
Student
t­
Statistics­
~
f
Characteristics
for
H
:
xi
=
x*

Zero
vs.
Characteristic
User
vs.
nonuser
nonzero
bids
Ownership
or
use
of
a
boat
Participation
in
any
outdoor
recreation
in
the
last
year
Numerical
rating
of
the
Monongahela
Rating
for
a
particular
site
Length
of
residence
Education
Race
Income
Age
Sex
2
.
4
7
1
b
10.
746
b
0.365
5.
988
b
0.242
1.655
­
0.804
1.124
­
5.
995
b
­
1.338
­
1.589
­
4.818b
­
1.369
­
3.
205
b
0.167
­
2.031b
1.699
­
1.713
4
.
9
4
2
b
­
0.347
at­
statistics
are
derived
from
the
results
reported
in
Chapter
4.
b
Denotes
significance
at
the
0.05
level.

c
­
2
h
­..
..
 
 
 
.
 
 
 
­­­
.

Table
C­
2.
Estimated
Option
Price
for
Changes
in
Water
Quality:
Effects
of
Instrument
and
Type
of
Respondent­­
All
Respondents
a
Type
of
respondent
User
Change
in
Nonuser
Combined
water
quality
i
s
n
i
s
n
)?
s
n
/

1.

2.

3.

4.
Iterative
bidding
framework­­
starting
point
=
$
25
(
Version
C)

D
to
E
(
avoid)
21.7
18.6
24
23.7
32.9
54
23.1
29.2
78
f)
toc
15.0
16.4
24
11.9
15.6
54
12.9
15.8
78
C
tO
Bb
9
.
4
13.7
24
5.7
10.7
54
6.9
11.7
78
Dto
B
25.4
27.5
24
41.4
51.5
54
combined:
all
levels
20.1
25.3
78
47.1
41,8
24
17.7
24.1
54
43.1
48.5
78
Iterative
bidding
framework­­
starting
point
=
$
125
(
Version
D)

D
to
E
(
avoid)
89.5
70.3
22
44.6
84.1
50
58.3
82.4
72
Dto
C
63.9
53.5
22
29.7
56.0
50
40.1
57.1
72
c
to
B
b
41.8
54.2
22
19.9
51.1
50
26.6
52.6
72
Dto
B
111.8
94.4
22
5
1
.
3
1
0
2
.
1
5
0
69.8
103.1
72
Combined:
all
levels
2
0
1
.
4
1
4
9
.
8
2
2
9
5
.
9
1
7
7
.
6
5
0
128.1
1
7
5
.
5
7
2
Direct
question
framework
(
Version
B)

D
to
E
(
avoid)
42.6
67.8
23
13.5
35.2
51
22.5
49.2
74
Dto
C
27.9
42.7
23
9.3
22.3
51
15.1
31.1
74
C
to
B
b
24.0
49.5
23
7.7
22.5
51
12.8
33.8
74
Dto
B
53.0
84.6
23
17.7
43.5
51
28.7
61.0
74
Combined:
all
levels
95.7
130.7
23
31.2
77.0
51
51.2
100.6
74
Direct
question
framework:
payment
card
(
Version
A)

D
to
E
(
avoid)
57.1
92.8
24
38.9
68.8
51
44.7
77.1
75
Dto
C
46.0
71.1
24
15.9
30.3
51
25.5
48.9
75
C
to
B
b
22.5
45.3
24
5.6
17.3
51
11.0
30.1
75
Dto
B
7
0
.
6
1
1
2
.
5
2
4
21.7
42.5
51
37.3
75.4
75
Combined:
all
levels
1
2
7
.
7
1
5
9
.
4
2
4
60.6
96.1
51
82.1
123.0
75
aThe
two
respondents
who
did
not
complete
the
questionnaire
are
excluded.
b
D
to
B
includes
respondents
who
were
wil[
ing
to
give
an
amount
only
for
fishable
or
swimmable
water
and
respondents
who
were
willing
to
pay
some
amount
to
avoid
the
decrease
in
water
quality
in
addition
to
the
improvements
in
water
quality.

c
­
3
A
Table
C­
3.
Estimated
Option
Price
for
Changes
in
Water
Quality:
Effects
of
Instrument
and
Type
of
Respondent­­
Protest
Bids
Excluded
Type
of
respondent
­

Change
in
User
Nonuser
Combined
~
water
quality
z
s
n
x
s
n
x
s
n
1.

2.

3.

4.
Iterative
bidding
framework­­
starting
point
=
$
25
(
Version
C)

D
to
E
(
avoid)
27.4
16.7
19
28.4
34.2
45
28.1
29.9
64
Dto
C
18.9
16.3
19
14.3
16.1
45
15.7
16.2
64
Cto
B
11.8
14.5
19
6.9
11.3
45
8.4
12.4
64
D
to
B
a
32.1
27.1
19
21.2
25.0
45
24.5
25.9
64
Combined:
all
levels
59.5
38.1
19
49.7
52.7
45
52.6
48.7
64
Iterative
bidding
framework­­
starting
point
=
$
125
(
Version
D)

D
to
E
(
avoid)
93.8
69.0
21
5
4
.
4
9
0
.
2
Dto
C
66.9
52.8
21
3
6
.
2
6
0
.
0
Cto
B
43.8
54.7
21
2
4
.
3
5
5
.
5
D
to
B
a
117.1
93.3
21
62.6
109.8
Combined:
all
levels
2
1
0
.
0
1
4
6
.
4
21
117.0
190.1
41
41
41
41
41
67.7
85.1
46.6
59.1
30.9
55.6
8
1
.
0
1
0
7
.
0
1
4
8
.
8
180.9
62
62
62
62
62
Direct
question
framework
(
Version
B)

D
to
E
(
avoid)
51.6
71.7
19
1
8
.
6
4
0
.
3
Dto
C
33.8
44.9
19
1
2
.
8
2
5
.
4
Cto
B
29.1
53.3
19
1
0
.
6
2
5
.
9
D
to
B
a
64.2
89.4
19
2
4
.
4
4
9
.
7
Combined:
all
levels
1
1
5
.
8
1
3
5
.
7
1
9
4
3
.
0
8
7
.
8
37
37
37
37
37
29.8
54.7
19.9
34.4
16.9
38.0
37.9
67.7
6
7
.
7
110.8
56
56
56
56
56
Direct
question
framework:
payment
card
(
Version
A)

D
to
E
(
avoid)
65.2
96.7
21
49.6
74.3
40
Dto
C
52.6
73.8
21
20.3
33.0
40
Cto
B
25.7
47.7
21
7.1
19.3
40
D
to
B
a
80.7
117.1
21
27.6
46.3
40
Combined:
all
levels
1
4
6
.
0
1
6
2
.
6
2
1
7
7
.
3
1
0
2
.
6
4
0
55.0
82.2
31.4
52.6
13.5
32.9
45.9
81.3
100.9
129.3
61
61
61
61
61
aD
to
B
includes
respondents
who
were
willing
to
give
an
amount
only
for
fishable
or
swimmable
water
and
respondents
who
were
willing
to
pay
some
amount
to
avoid
the
decrease
in
water
quality
in
addition
to
the
improvements
in
water
quality.
Table
C­
4.
Estimated
User
Values
for
Changes
in
Water
Quality:
Effects
of
Instrument
and
Type
of
Respondent­­
All
Respondents
a
Type
of
respondent
Change
in
User
Combined
water
quality
i
s
n
i
s
n
1.

2.

3.

4.
Iterative
bidding
framework­­
starting
point
=
$
25
(
Version
C
)

D
to
E
(
avoid)
5
.
2
11.5
24
1
.
6
6
.
7
Dto
C
3
.
3
7
.
0
24
1.0
4.1
C
to
Bb
4
.
0
7
.
4
24
1
.
2
4
.
4
Dto
B
8
.
3
13.5
24
2.6
8.3
Combined:
all
levels
13.5
23.3
24
4
.
2
14.2
Iterative
bidding
framework­­
starting
point
=
$
125
(
Version
D)

D
to
E
(
avoid)
38.0
58.9
22
11.6
Dto
C
31.1
50.0
22
9
.
5
C
to
B
b
32.0
52.9
22
9
.
8
Dto
B
69.3
102.1
22
21.2
Combined:
all
levels
107.3
147.3
22
32.8
Direct
question
framework
(
Version
B)

D
to
E
(
avoid)
19.1
37.6
23
5
.
9
Dto
C
18.0
37.7
23
5
.
6
C
to
B
b
11.9
31.6
23
3
.
7
Dto
B
29.9
62.3
23
9
.
3
Combined:
all
levels
49.0
81.9
23
15.2
Direct
question
framework:
payment
card
(
Version
A)

D
to
E
(
avoid)
20.2
35.0
24
6
.
5
Dto
C
30.2
73.2
24
9
.
7
C
to
B
b
16.0
42.7
24
5.1
Dto
B
46.7
113.5
24
14.9
Combined:
all
levels
66.9
121.3
24
21.4
36.5
30.8
32.4
64.1
94.3
22.5
22.3
18.2
36.9
50.4
21.7
43.2
25.0
67.0
74.6
78
78
78
78
78
72
72
72
72
72
74
74
74
74
74
75
75
75
75
75
a
The
two
respondents
who
did
not
complete
the
questionnaire
are
excluded
b
D
to
B
includes
respondents
who
were
willing
to
give
an
amount
only
for
fishable
or
swimmable
water
and
respondents
who
were
willing
to
pay
some
amount
to
avoid
the
decrease
in
water
quality
in
addition
to
the
improvements
in
water
quality.

c
­
5
Table
C­
5.
Estimated
User
Values
for
Changes
in
Water
Quality:
Effects
of
Instrument
and
Type
of
Respondent­­
Protest
Bids
Excluded
Type
of
respondent
Change
in
User
Combined
water
quality
i
s
n
i
s
n
1.
Iterative
bidding
framework­­
starting
point
=
$
25
(
Version
C)

D
to
E
(
avoid)
6
.
6
12.6
19
2.0
7.4
Dto
C
4
.
2
7.7
19
1
.
3
4
.
5
Cto
B
5
.
0
8.0
19
1.5
4.9
D
to
B
a
10.5
14.4
19
3.1
9.1
Combined:
all
levels
17.1
25.1
19
5.1
15.6
2.
Iterative
bidding
framework­­
starting
point
=
$
125
(
Version
D)

D
to
E
(
avoid)
39.8
59.7
21
13.5
Dto
C
32.6
50.7
21
11.0
Cto
B
33.6
53.7
21
11.4
D
to
B
a
72.6
103.4
21
24.6
Combined:
all
levels
112.4
148.9
21
38.1
3.
Direct
question
framework
(
Version
B)

D
to
E
(
avoid)
23.1
40.4
19
7
.
8
Dto
C
21.8
40.6
19
7
.
4
Cto
B
14.4
34.4
19
4
.
9
D
to
Ba
36.2
67.1
19
12.3
Combined:
all
levels
59.3
86.9
19
20.1
4.
Direct
question
framework:
payment
card
(
Version
A)

D
to
E
(
avoid)
23.1
36.6
21
8
.
0
Dto
C
34.5
77.5
21
11.9
Cto
B
18.3
45.3
21
6
.
3
D
to
Ba
53.3
120.2
21
18.4
Combined:
all
levels
76.4
127.1
21
26.3
39.1
32.9
34.7
68.6
100.7
25.6
25.4
20.9
42.1
57,2
23.8
47.7
27.6
73.9
82.0
64
64
64
64
64
62
62
62
62
62
56
56
56
56
56
61
61
61
61
61
aD
to
B
includes
respondents
who
were
willing
to
give
an
amount
only
for
fishable
or
swimmable
water
and
respondents
who
were
willing
to
pay
some
amount
to
avoid
the
decrease
in
water
quality
in
addition
to
the
improvements
in
water
quality.

C
­
6
.
.
.
Table
c
­
6
.
Estimated
Option
Values
for
Changes
in
Water
Qualit~
Effects
of
Instrument
and
Type
of
Respondent­­
All
Respondents
Type
of
respondent
Change
in
User
Nonuser
Combined
Water
quality
i
s
n
i
s
n
i
s
n
 
1.

2.

3.

4.
Iterative
bidding
framework­­
starting
point
=
$
25
(
Version
C)

D
to
E
(
avoid)
16.5
17.0
24
23.7
32.9
54
21.5
29.1
78
DtOc
11.7
13.8
24
11.9
15.6
54
11.9
15.0
78
c
t
o
B
b
5.4
9.9
24
5.7
10.7
54
5.6
10.4
78
Dto6
17.1
21.5
24
41.4
51.5
54
17.5
23.2
78
Combined:
all
levels
33.5
33.2
24
17.7
24.1
54
39.0
46.6
78
Iterative
bidding
framework­­
starting
point
=
$
125
(
Version
D)

D
to
E
(
avoid)
51.6
69.9
22
4
4
.
6
8
4
.
1
Dto
C
32.7
48.2
22
2
9
.
7
5
6
.
0
c
to
Bb
9
.
8
28.2
22
1
9
.
9
5
1
.
1
Dto
B
42.5
66.5
22
51.3
102.1
Combined:
all
levels
94.1
119.8
22
95.9
177.6
Direct
question
framework
(
Version
B)

D
to
E
(
avoid)
23.5
41.6
23
1
3
.
5
3
5
.
2
Dto
C
9
.
9
22.9
23
9
.
3
2
2
.
3
c
to
Bb
12.1
28.6
23
7
.
7
2
2
.
5
Dto
B
23.1
50.5
23
1
7
.
7
4
3
.
5
Combined:
al!
levels
46.7
84.5
23
3
1
.
2
7
7
.
0
Direct
question
framework:
payment
card
(
Version
A)
50
50
50
50
50
51
51
51
51
51
D
to
E
(
avoid)
36.9
73.9
24
38.9
68.8
51
Dto
C
15.8
25.4
24
15.9
30.3
51
C
to
Bb
6
.
5
21.1
24
5.6
17.3
51
Dto
B
24.0
43.6
24
21.7
42.5
51
Combined:
all
levels
6
0
.
8
1
1
5
.
2
2
4
60.6
96.1
51
46.7
79.6
72
30.6
53.4
72
16.8
45.3
72
48.6
92.3
72
95.3
161.3
72
16.6
37.3
74
9
.
5
22.4
74
9.1
24.4
74
19.4
45.5
74
36.0
79.1
74
38.3
70.0
75
15.9
28.7
75
5
.
9
18.5
75
22.4
42.6
75
60.7
101.8
75
a
The
two
respondents
who
did
not
complete
the
questionnaire
are
excluded.
b
D
to
B
includes
respondents
who
were
willing
to
give
an
amount
only
for
fishable
or
swimmable
water
and
respondents
who
were
willing
to
pay
some
amount
to
avoid
the
decrease
in
water
quality
in
addition
to
the
improvements
in
Water
quality.

c
­
7
.

Table
C­
7.
Estimated
Option
Values
for
Changes
in
Water
Quality:
Effects
of
Instrument
and
Type
of
Respondent­­
Protest
Bids
Excluded
Type
of
respondent
Change
in
User
Nonuser
Combined
water
quality
i
s
n
2
s
n
s
s
n
1.
Iterative
bidding
framework­­
starting
point
=
$
25
(
Version
C)

D
to
E
(
avoid)
20.8
16.6
19
28.4
34.2
45
26.2
30.1
64
Dto
C
14.7
14.0
19
14.3
16.1
45
14.5
15.4
64
Cto
B
6
.
8
10.7
19
6.9
11.3
45
6.9
11.1
64
D
to
Ba
21.6
22.1
19
21.2
25.0
45
21.3
24.0
64
Combined:
all
levels
42.4
31.9
19
49.7
52.7
45
47.5
47.3
64
2.
Iterative
bidding
framework­­
starting
point
=
$
125
(
Version
D)

D
to
E
(
avoid)
54.0
70.7
21
Dto
C
34.3
48.8
21
Cto
B
10.2
28.8
21
D
to
B
a
44.5
67.4
21
Combined:
all
levels
98.6
120.9
21
3.
Direct
question
framework
(
Version
B)

D
to
E
(
avoid)
28.5
44.4
19
Dto
C
12.0
24.8
19
Cto
B
14.7
31.0
19
D
to
Ba
28.0
54.5
19
Combined:
all
levels
56.5
90.2
19
4.
Direct
question
framework:
payment
card
D
to
E
(
avoid)
42.1
77.7
21
Dto
C
18.1
26.5
21
Cto
B
7
.
4
22.5
21
D
to
Ba
27.4
45.7
21
Combined:
all
levels
69.5
121.0
21
5
4
.
4
9
0
.
2
3
6
.
2
6
0
.
0
2
4
.
3
5
5
.
5
62.6
109.8
117.0
190.1
1
8
.
6
4
0
.
3
1
2
.
8
2
5
.
4
1
0
.
6
2
5
.
9
2
4
.
4
4
9
.
7
4
3
.
0
8
7
.
8
(
Version
A)

4
9
.
6
7
4
.
3
2
0
.
3
3
3
.
0
7
.
1
1
9
.
3
2
7
.
6
4
6
.
4
77.3
102.6
41
41
41
41
41
37
37
37
37
37
40
40
40
40
40
54.3
83.5
62
35.6
56.1
62
19.5
48.4
62
56.5
97.3
62
1
1
0
.
7
1
6
9
.
0
6
2
21.9
41.6
56
12.6
25.0
56
12.0
27.5
56
25.6
50.9
56
47.6
88.0
56
47.0
75.0
61
19.5
30.7
61
7
.
2
20.3
61
27.5
45.8
61
74.6
108.3
61
aD
to
B
includes
respondents
who
were
willing
to
give
an
amount
only
for
fishable
or
swimmable
water
and
respondents
who
were
willing
to
pay
some
amount
to
avoid
the
decrease
in
water
quality
in
addition
to
the
improvements
in
water
quality.

C
­
8
1
Table
C­
8.
Option
Price­
Student
t­
Statistics
for
H
O
:
X
=
O
With
Outliers
and
Protest
Bids
Excluded
a
.
e
User
Nonuser
Total
sample
~
ayment
card
Level
D
to
E
Level
D
to
C
Level
c
to
.
B
Total
D
to
B
Total
E
to
B
Direct
question
Level
D
to
E
Level
D
to
C
Level
C
to
B
Levei
D
to
B
Total
E
to
B
Iterative
Bidding
$
25
Level
D
to
E
Level
D
to
C
Level
C
to
B
Total
D
to
B
Total
E
to
B
Iterative
bidding
$
125
Level
D
to
E
Level
D
to
C
Level
C
to
B
Total
D
to
B
Total
E
to
B
4.56
2.62
2.49
4.15
2.86
2.92
2.34
3.01
3.91
7.14
5.07
3.57
5.15
6.80
5.73
4.48
2.74
4.54
5.70
4.22
3.94
2.34
3.82
4.81
3.05
2.93
2.26
2.86
3.03
5.20
5,97
3.86
5.63
6.05
4.27
3.27
3.32
4.37
5.59
4.36
2.85
4.04
6.34
3.86
3.93
3.22
4,03
4.67
7.20
7.81
5.23
7.54
8.48
6.42
5.16
3.28
5.21
6.46
aOnly
those
values
that
are
significant
at
the
0.05
level
are
reported.

c
­
9
Table
C­
9.
User
Value
­
Student
t­
Statistics
for
HO:
X=
O
With
Out!
iers
and
Protest
Bids
Excluded
a
User
Total
sample
Payment
card
Level
D
to
E
Level
D
to
C
Level
C
to
B
Tota
I
D
to
B
Total
E
to
B
Direct
question
Level
D
to
E
Level
D
to
C
Level
C
to
B
Total
D
to
B
Total
E
to
B
Iterative
bidding
$
25
Level
D
to
E
Level
D
to
C
Level
C
to
B
Total
D
to
B
Total
E
to
D
Iterative
bidding
$
125
Level
D
to
E
Level
D
to
C
Level
C
to
B
Total
D
to
B
Total
E
to
B
2.37
2.29
2.15
2.71
2.28
2.39
2.73
3.18
2.97
2.46
2.22
2.46
2.17
2.11
2.01
2.42
2.12
2.21
2.46
2.76
2.62
2.23
2.05
2.23
aOnly
those
values
that
are
significant
at
the
0.05
level
are
reported.

Table
C­
10.
Option
Value
Student
t­
Statistics
for
Differences
in
Means
Between
Bidding
Method
s­­
Outliers
and
Protest
Bids
Excluded
a
User
Total
sample
Iterative
bidding
$
25
vs.
iterative
bidding
$
125
Level
D
to
E
­
2
.
1
4
­
1
.
9
7
Total
E
to
B
­
2.11
aOnly
those
values
that
are
significant
at
the
0.05
level
are
reported.

.

c­
lo
Table
C­
11
.
Regression
Results
for
Option
Price
Estimates
of
Water
Quality
Changes­­
Protest
Bids
Excluded
*
ter
quality
changea
Total
:
Total
improvement
1=
Independent
Variables
D
to
E
(
avoid)
Dto
C
Cto
B
all
levels
only
,
fltercePt
I
Age
Education
Income
Direct
question
Iterative
biddin9
9ame
($
25)

Iterative
bidding
game
($
125)

User
(
1
if
user)

Willing
to
pay
cost
of
water
pollution
(
I
if
very
much
or
somewhat)

Interviewer
1
Interviewer
2
Interviewer
3
Interviewer
4
Interviewer
5
Interviewer
6
Interviewer
7
Interviewer
8
Interviewer
9
R
2
F
­
22.132
(­
0.510)

23.756
(
2.104)

­
0.314
(­
0.983)

3.826
(
1
.244)

0.0006
(
1
.299)

­
31.506
(­
2.208)

­
22.986
(­
1.671)

28.606
(
2.02S)

12.896
(
1.097)

18.719
(
1.601)

30.857
(
1.325)

7.754
(
0.355)

­
24.009
(­
1.32)

19.348
(
0.501)

6.982
(
0.316)

36.351
(
0.716)

42.280
(
1.815)

11.136
(
0.510)

49.806
(
1.385)

0.281
3.61
166
­
18.171
(­
0.627)

5.268
(
0.698)

­
0.283
(­
1.328)

1.968
(
0.956)

0.0002
(
0.587)

­
13.203
(­
1
.384)

­
13.455
(­
1.462)

21.775
(
2.308)

10.799
(
1
.374)

23.848
(
3.0s0)

13.435
(
0.862)

15.931
(
1.091)

21.959
(
1.547)

20.235
(
0.783)

3.354
(
0.227)

5J
.645
(
1.490)

6.505
(
0.418)

25.584
(
1
.750)

30.573
(
1.271)

0.248
2.99
166
4.690
(
0.177)

3.989
(
0.577)

­
0.239
(­
1.221)

0.306
(­
0.162)

0.0002
(
0.892)

0.777
(
0.089)

­
5.338
(­
0.634)

19.461
(
2.252)

10.288
1.430
9.538
1.332
15.658
(
1.097)

16.379
(
1
.224)

8.755
(
0.674)

32.428
(
1.370)

­
4.095
(­&
302)

27.450
(
0.882)

7.411
(
0.520)

14.498
(
1
.083)

29.078
(
1.320)

0.148
1.61
166
­
25.618
(­
0.308)

33.597
(
1.555)

­
0.869
"(­
1.423)

5.020
(
0.853)

0.001
(
1.178)

­
44.026
(­
1.613)

­
41.798
(­
1.588)

74.029
(
2.743)

35.420
1.575
53.944
(
2.411)

54.693
(
1
.227)

34.788
(
0.832)

1.571
(
0.039)

66.575
(
0.900)

4.168
(
0.099)

108.924
(
1.121)

58.627
(
1.315)

46.024
(
1.101)

101.538
(
1.476)

0.276
3.51
166
­
3.486
(­
0.069)

9.840
(
0.744)

­
0.555
(­
1.485)

1.194
(
0.331)

0.0004
(
0.815)

­
12.520
(­
0.749)

­
18.813
(­
1.168)

45.423
(
2.749)

22.523
(
1.636)

35.225
(
2.572)

23.836
(
0.874)

27.034
(
1.057)

25.580
(
1.029)

47.227
(
1.043)

­
2.814
(­
0.109)

72.572
(
1.220)

76.347
(
0.599
34.888
(
1.363
51.732
(
1.228
0.229
2.74
166
Degrees
of
freedom
aNumbers
in
parentheses
are
asymptotic
t­
ratios
for
the
null
hypothesis
of
no
association.

I
.
C
­
I
I
Table
C­
12.
Regression
Results
for
User
Value
Estimates
of
Water
Quality
Changes
­­
Protest
Bids
Excluded
s
Water
quality
change
a
Total:
Total
improvement
Independent
variables
D
to
E
(
avoid)
Dto
C
Cto
B
all
levels
only
Intercept
Sex
Age
Education
Income
Direct
question
Iterative
bidding
($
2S)

Iterative
bidding
($
12S)

Willing
to
pay
cost
Interviewer
1
Interviewer
2
Interviewer
3
Interviewer
4
Interviewer
5
Interviewer
6
Interviewer
7
Interviewer
8
Interviewer
9
R2
F
Degrees
of
freedom
26.618
(
1.408)

­
0.567
(­
0.115)

­
0.328
(­
2.512)

0.140
(
0.104)

0.000002
(
0.010)

­
1.694
(­
0.271)

­
5.195
(­
0.860)

6.214
(
1
.006)

4.790
(
0.9s0)

­
10.977
(­
1.075)

­
5.433
(­
0.567)

­
9.462
(­
1.039)

­
11.818
(­
0.697)

­
12.842
(­
1
.322)

­
10.835
(­
0.486)

4.895
(
0.482)

­
10.016
(­
1
.044)

­
2.618
(­
0.166)

0.11
1.26
167
9.513
(
0.422)

­
7.465
(­
1.273)

`
0.231
(­
1.485)

0.212
(
0.132)

0.0001
(
0.594)

­
5.944
(­
0.796)

­
11.770
(­
1.635)

­
2.406
(­
0.327)

9.560
(
1.591)

­
3.649
(­
0.300)

4.711
(
0.412)

23.386
(
2.153)

1.810
(
0.090)

­
5.401
(­
0.466)

9.970
(
0.375)

6.735
(
0.557)

6.084
(
0.532)

­
0.119
(­
0.006)

0.12
1.32
167
9.497
(
0.546)

­
5.447
(­
1.204)

­
0.172
(­
1.431)

0.253
(
0.204)

0.0001
(
0.452)

­
1.312
(­
0.228)

­
4.114
(­
0.740)

5.525
(
0.972)

4.808
(
1
.037)

­
7.453
(­
0.793)

­
1.321
(­
0.150)

8.302
(
0.990)

­
3.542
(­
0.227)

­
9.620
(­
1
.076)

­
1.871
(­
0.091)

1.162
(
0.124)

­
4.086
(­
0.463)

7.050
(
0.48S)

0.09
0.99
167
24.423
(
0.630)

­
11.303
(­
1.122)

­
0.455
(­
1
.698)

­
0.041
(­
0.015)

0.0003
(
0.667)

­
8.307
(­
0.647)

­
15.345
(­
1.240)

6.233
(
0.492)

14.834
(
1.436)

­
9.504
(­
0.454)

4.240
(
0.216)

32.793
(
1.756)

­
0.471
(­
0.014)

­
12.998
(­
0.653)

7.909
(
0.173)

15.612
(
0.750)

2.539
(
0.129)

6.722
(
0.208)

0.11
1.26
167
51.041
(
1.023)

­
11.870
(­
0.915)

­
0.783
(­
2.270)

0.098
(
0.028)

0.0003
(
0.522)

­
10.001
(­
0.605)

­
20.541
(­
1.289)

12.447
(
0.763)

19.624
(
1.475)

­
20.481
(­
0.760)

­
1.193
(­
0.047)

23.331
(
0.970)

­
12.289
(­
0.275)

­
25.840
(­
1.008)

­
2.926
(­
0.050)

20.507
(
0.765)

­
7.478
(­
0.295)

4.105
(
0.099)

0.12
1.39
167
a
Numbers
in
parentheses
are
symptotic
t­
ratios
for
the
null
hypothesis
of
no
association.

C­
12
Table
C­
13.
Regression
Results
for
Option
Value
Estimates
of
Water
Quality
Changes
­­
Protest
Bids
Excluded
Water
quality
changea
Total:

Independent
variables
improvement
D
to
E
(
avoid)
DtoC
ctoB
only
Intercept
­
3.931
1.879
(­
0.105)
(
0.091
j
Sex
18.033
7.528
(
1
.745)
(
1.324)

Age
­
0.341
­
0.302
(­
1.172)
(­
1.885)

EducatiOn
3.202
1.595
(
1.143)
(
1.035)

Income
0.0003
­
0.0001
23.017
(
1.205)

5.259
(
1
.002)

­
0.232
­
1.568)

­
0.810
­
0.569)

­.
0000
24.897
(
0.684)

12.096
(
1.209)

­
0.544
(­
1
.928)

0.888
(
0.328)

­
0.0001
(
0.830)
.
..­
(­
0.477)
(­
0.013)
(­
0.324)

Direct
question
­
25.304
­
5.552
4.980
­
0.257
(­
1.872)
(­
0.747)
(
0.725)
(­
0.020)

Iterative
bidding
($
25)
­
15.199
0.690
0.970
0.775
(­
1
.164)
(
0.096).
(
0.146)
(
0.061)

Iterative
bidding
($
125)
25.841
27.809
17.004
45.796
(
1.936)
(
3.802)
(
2.508)
(
3.544)

Willing
to
pay
cost
27.643
21.039
10.588
33.146
(
2.655)
3.673
(
2.001)
(
3.287)

User
­
18.682
­
14.078
­
9.307
(­
1.770)
­
24.071
(­
2.424)
(­
1.735)
(­
2.355)

R
2
0.179
0.217
0.090
0.777
F
4.22
5.39
1.92
4.18
Degrees
of
freedom
175
175
175
175
aNumbers
In
parentheses
are
asymptotic
t­
ratios
for
the
null
hypothesis
of
no
association.

C­
13
i
Table
C­
14.
Regression
Results
for
Option
Value
Estimates
of
Water
Quality
Changes­­
Protest
!
3ids
and
Outliers
Excluded
a
Water
auality
chanqe
Total:
improvement
Independent
variables
D
to
E
(
avoid)
Dto
C
cto
B
only
Intercept
Sex
Age
Education
Income
Direct
question
Iterative
bidding
($
25)

Iterative
bidding
($
125)

User
Willing
to
pay'
cast
Interviewer
1
Interviewer
2
Interviewer
3
Interviewer
4
Interviewer
5
Interviewer
6
Interviewer
7
Interviewer
8
­
35.228
(­
1.019)

5.779
(
8.986)

­
0.277
(­
1.066)

5.306
(
2.131)

0.0006
(
1
.532)

­
29.503
(­
2.596)

­
14.040
(­
1
.294)

13.018
(
1.084)

14.51.5
(­
1.549)

11.346
(
1.224)

20.321
(
1.100)

­
1.272
(­
0.075)

­
9.319
(­
0.563)

­
20.891
(­
0.656)

13.911
(
0.832)

54.899
(
1.063)

20.251
(
1.070)

19.014
(
1.115)

38.062
(
0.992)

0.269
2.78
136
­
24.058
(­
1.185)

­
0.172
(­
0.033)

­
0.182
(­
1.188)

2.880
(
1
.975)

0.0001
(
0.564)

­
8.628
(­
1
.292)

­
0.575
(­
0.090)

16.697
(
2.366)

­
8.312
(­
1.510)

14.134
(
2.595)

6.248
(
0.578)

­
0.279
(­
0.028)

0.349
(
0.036)

­
5.726
(­
0.306)

4.466
(
0.454)

76.817
(
2.530)

1.467
(
0.132)

18.181
(
1.814)

43.784
(
1
.942)

0.294
3.14
136
0.683
(
0.043)

­
2.209
(­
0.531)

­
0.155
(­
1.286)

0.148
(
0.128)

0.0002
(
1.39)

0.786
(
0.149)

0.160
(
0.032)

4.633
(
0.833)

­
2.763
(­
0.637)

3.666
(
0.854)

10.166
(
1.189)

6.402
(
0.818)

2.596
(
0.339)

16.615
(
1.123)

2.793
(
0.361)

55.478
(
2.318)

S.
098
(
0.582)

15.698
(
1.987)

­
3.945
(­
0.222)

0.129
1.12
136
­
17.021
(­
0.547)

­
4,046
(­
0.500)

­
0.326
(­
1.390)

3.088
(
1
.378)

0.0003
(
0.863)

­
6.927
(­
0.676)

­
1.138
(­
0.116)

23.315
(
2.153)

­
11.371
(­
1.346)

19.901
(
2.382)

Interviewer
9
R2
F
Degrees
of
freedom
136
a
Numbers
in
parentheses
are
asymptotic
t­
ratios
for
the
null
hypothesisof
no
association.
9.072
(
0.545)

­
0.745
(­
0.049)

­
3.135
(­
0.210)

2.848
(
0.099)

2.562
(
0.170)

12S
.627
(
2.698)

2.024
0.119
27.557
(
1.792)

30.263
(
0.875)

0.253
2.55
1.
C­
14
Table
C­
15.
Regression
Results
for
Option
Value
Estimates
of
Water
Quality
Changes­­
Protest
Bids
Excluded
Water
quality
changea
Total:
improvement
~~
deDendent
variables
D
to
E
(
avoid)
Dto
C
Cto
B
only
,!
.­
 
,

/
­
36.611
,
ntercePt
(­
0.890)

seX
20.914
(
1.953)

Age
Education
Income
Direct
question
Iterative
bidding
($
25)

Iterative
bidding
($
125)

User
Willing
to
pay
cost
Interviewer
1
­
0.257
(­
0.849)

4.067
(
1
.394)

0.0005
(
1
.252)

­
30.187
(­
2.230)

­
16.969
(­
1.300)

24.667
(
1
.843)

­
14.859
(­
1.333)

19.183
(
1
.730)

45.060
(
2.039)

Interviewer
2
13.174
(
0.636)

Interviewer
3
­
4.031
(­
0.200)

Interviewer
4
28.659
(
0.782)

Interviewer
5
19.815
(
0.944)

Interviewer
6
46.018
(
0.955)

Interviewer
7
44.117
(
1
.997)

Interviewer
8
19.804
(
0.955)

Interviewer
9
50.923
(
1.493)

R
2
0.241
F
2.93
Degrees
of
freedom
166
­
17.778
(­
0.770)

9.950
(
1
.655)

­
0.274
(­
1.611)

2.067
(
1
.262)

­
0.0000
(­
0.010)

­
7.565
(­
0.996)

­
1.014
(­
0.138)

26.037
(
3.467)

­
11.852
(­
1
.894)

18.577
(
2.984)

19.717
(
1
.590)

11.210
(
0.964)

7.156
(
0.633)

16.378
(
0.796)

8.747
(
0.743)

39.722
(
1.469)

5.264
(
0.425)

18.400
(
1.581)

29.468
1.539
0.247
3.03
166
2.781
(
0.132)

7.304
(
1
.329)

­
0.236
(­
1.521)

­
0.321
(­
0.214)

0.0001
(
0.610)

1.854
(
0.267)

­
0.711
(­
0.106)

15.358
(
2.237)

­
7.063
(­
1
.235)

8.014
(
1.409)

25.128
(
2.216)

17.693
(
1.665)

7.027
(
0.681)

34.403
(
1.829)

5.520
(
0.513)

28.591
(
1.156)

10.457
(
0.923)

17.741
(
1
.668)

21.089
(
1.205)

0.143
1.54
166
­
9.546
(­
0.235)

15.986
(
1.514)

­
0.511
(­
1.711)

1.811
(
0.630)

0.0001
(
0.206)

­
4.781
(­
0.358)

­
2.224
(­
0.173)

42.630
(
3.231)

­
19.465
(­
1.771)

28.340
(
2.592)

38.220
(
1
.754)

22.775
(
1.115)

8.697
(
0.438)

43.904
(
1.215)

10.171
(
0.492)

62.895
(
1.324)

10,920
(
0.501
)

30.310
(
1
.483)

42.740
(
1.271)

0.212
2.48
166
a
Numbers
in
parentheses
are
asymptotic
t­
ratios
for
the
null
hypothesis
of
no
association.

C­
15
..­
.
.
.
.
.
 
 
.
.
 
.
.
_­..,._.
.
 
 
 
.
 
.
.
.
Table
C­
16.
Benefit
Estimates
from
Contingent
Ranking
Models
Model/
estimator
Average
Range
Payment
=
5
Water
quality
change:
boatable
to
fishable
Final
Model
(
specification
I
)

Ordered
Iogit
­
8
.
7
7
­
7
3
.
7
7
t
o
1
1
5
.
8
2
Ordered
normal
­
9
.
9
0
­
1
5
7
.
0
2
t
o
2
8
7
.
8
8
II
Payment
=
50
Water
quality
change:
boatable
to
fishable
Ordered
Iogit
51.40
4
8
.
5
1
t
o
55.41
Ordered
normal
72.45
4
9
.
0
6
t
o
97.79
I
l
l
Payment
=
100
Water
quality
change:
boatable
to
fishable
Ordered
logit
49.56
4
8
.
3
1
t
o
51.70
Ordered
normal
69.39
4
8
.
9
0
t
o
85.94
Iv
Ordered
Iogit
Ordered
norms
v
Ordered
Iogit
Ordered
normal
VI
Ordered
Iogit
Ordered
normal
Vll
Ordered
Iogit
Ordered
normal
Vlll
Ordered
Iogit
Ordered
normal
Payment
=
175
Water
quality
chanqe:
boatable
to
fishable
49.17
4
8
.
2
6
t
o
50.94
68.75
4
8
.
8
6
t
o
83.67
Payment
=
5
Water
quality
chanqe:
boatable
to
swimmable
­
15.78
­
1
3
2
.
7
8
t
o
2
0
8
.
4
8
­
17.82
­
2
8
2
.
6
4
t
o
5
1
8
.
1
8
Payment
=
SO
Water
quality
change:
boatable
to
swimmable
92.52
8
7
.
3
1
t
o
99.74
130.40
88.30
to
176.02
Payment
=
100
Water
quality
change:
boatable
to
swimmable
89.21
8
6
.
9
5
t
o
93.05
124.90
88.01
to
154.70
Payment
=
175
Water
quality
change:
boatable
to
swimmable
88.51
8
6
.
8
7
t
o
91.69
123.75
87.95
to
150.60
c
­
1
6
Table
C­
17.
Estimated
Option
Values
fcr
Water
Quality
Change:
Effects
of
Instrument
and
Type
of
Respondent­­
Protest
Bids
and
Outliers
Excluded
Type
of
respondent
User
a
Change
in
Nonuser
water
quaiity
i
s
n
i
s
n
1.

2.

3.

4.
Iterative
Bidding
Framework,
Starting
Point
=
$
25
D
to
E
(
avoid)
21.43
16.81
14
28.52
Dto
C
14.64
12.32
14
14.55
C
to
Bb
8.93
11.80
14
6.48
Dto
B
23.57
22.65
14
21.02
Iterative
Bidding
Framework,
Startinq
Point
=
$
125
D
to
E
(
avoid)
62.33
Dto
C
40.33
C
to
B
b
14.00
Dto
B
54.33
Direct
Question
Framework
D
to
E
(
avoid)
18.21
Dto
C
10.50
C
to
Bb
9.86
Dto
B
22.14
Payment
Card
67.03
49.77
33.60
72.60
31.29
26.94
27.14
53.73
15
15
15
15
14
74
14
14
37.58
25.45
11.21
39.24
17.89
10.62
8.73
20.30
D
to
E
(
avoid)
27.73
30.03
11
49.19
Dto
C
15.91
21.19
11
20.47
C
to
Bb
5.00
10.00
11
6.63
Dto
B
20.91
27.46
11
28.26
34.16
15.47
11.13
23.61
50.96
44.90
32.60
68.30
34.42
20.74
20.97
39.75
72.69
32.27
18.70
44.87
44
44
44
44
33
33
33
33
37
37
37
,
37
43
43
43
43
a
These
results
are
based
on
the
narrow
definition
of
users.
b
D
to
B
represents
the
sum
of
bids
for
the
improvements
in
water
quality
and
for
some
individuals
the
payment
to
move
from
Level
D
to
Level
B
directly.

C­
17
Table
c­
18.
A
Comparison
of
Contingent
Valuation
and
Travel
Cost
Benefit
Estimates­­
Protest
Bids
and
Outllers
Excluded
a
AWQ
=
Loss
of
area
AWQ
=
Beatable
to
fishable
AWQ
=
Eoatable
to
swimmable
Model
Test
b
Model
Test
b
Model
Test
b
Independent
variable
Intercept
17.482
(
1.022)
­
35.422
(
1.672)
58.359
(
1.669)

Travel
cost
benefit
.450
3.608
­
4.923
­
1.708
­
3.166
­
1.600
estimate
(
1.475)
(­
1.298)
(­
1.076)

Qualitative
variables
Payment
card
­
34.502
69.510
109.632
(­
2.335)
­
(
2.883)
(
2.734)

Direct
question
­
27.039
­
17.831
(­
2.062)
(
0.850)

Iterative
bid
($
25)
­
28.803
­
4.740
(­
1.993)
­
(­
0.201)
17.421
(
0.499)

­
11.500
(­
0.283)

R2
.117
.158
.146
n
68
68
68
F
2.09
2.68
(
0.
oa)
c
(:%=
(
0.04)
C
.

aThe
numbers
in
parentheses
below
the
estimated
coefficients
are
t­
ratios
for
the
null
hypothesis
of
no
association.

bThis
column
reports
the
t­
ratio
for
the
hypothesis
that
the
coefficient
for
the
travel
cost
variable
was
1.55.
The
travel
cost
model
measures
consumer
surplus
In
1977
dollars.
The
contingent
valuation
experiments
were
conducted
in
1981.
Using
the
consumer
price
index
to
adjust
the
travel
cost
benefit
estimates
to
1981
dollars
would
require
multiplying
each
estimate
by
1.55.
Since
the
estimated
regression
coefficients
(
and
standard
errors)
will
correspondingly
adjust
to
reflect
this
scale
change,
atestof
the
null
hypothesis
that
the
coefficient
of
travel
cost
was
equal
to
unity
is
equivalent
to
a
test
that
Is
equal
to
1.55
when
the
travel
cost
benefit
estimates
are
measured
in
1977
dollars
and
user
values
estimates
(
the
dependent
variable
are
in
1981
dollars.

cThis
number
in
parentheses
below
the
reported
F­
statistic
is
the
level
of
significance
for
rejection
of
the
null
hypothesis
of
no
association
between
the
dependent
and
independent
variables.

C­
18
APPENDIX
D
SURVEY
QUESTIONNAIRES
This
appendix
contains
two
parts.
Part
1
contains
the
survev
auestion
­
naires
as
administered
during
the
survey
of
the
Monongahela
Part
2
contains
a
brief
summary
of
suggestions
for
improving
the
for
future
use
in
similar
surveys.

D­
1
Ri've~
basin.
questionnaire
 
 
.
.
.
.
..
 
 
­
 
.
 
 
 
.
.
PART
1
SURVEY
QUESTIONNAIRE
AS
ADMINISTERED
DURING
THE
MONONGAHELA
RIVER
BASIN
SURVEY
m
#
2000­
0381
Approval
Srpima:
9/
20/
82
ESTIMATING
BENEFITS
OF
WATER
QUALITY
QUESTIONNAIRE
Form
??
0.
02
(
1)

I.
IDENTIFICATICN
IATORMATION
A.
Study
NO.
~
­
n
B
psulsesm­
t
No.
n
­

~
1
(
2­
6)
(
8­
13)

c.
Housing
Unit
No.
[
Ill
D.

(
15­
17)
;==
W==
~
(%
iD)

E.
Sample
Individual
Roster
Line
No.
m
F.
@
lestiomnaire
Version
(
19­
20)
A
II.
INTRODUCTION
(
22)

IF
THE
ENUMERATION
RESPONDENT
IS
ALSO
TRS
SELECTED
SAMPLE
INDIVIDUAL,
CONTINUE
YOUR
INTRODUCTION
TO
TRl?
STUDY
BY
READING
TRE
SECOND
PARAGRAPH
BELOW.
IF
TNE
SANPLE
INI)
IVIDUAL
IS
SOMEONE
OTHER
TNAN
TSE
ENUMERATION
RESPONDENT
,
READ
THE
ENTIRX
INTRODUCTION
BELOW.

Hello,
I`
m
(
NAHII)
f
roa
the
Research
Triangle
Institute
in
North
Caroline.
!
de
are
doing
a
study
for
a
government
agency
to
study
levels
of
water
quality
and
some
outdoor
recreational
activities
peop"
le
take
psrt
in
both
near
­
d
on
ponds,
lakes,
streams
and
rivers
in
the
)
lonongahela
River
Baein.
You
have
been
randomly
selected
to
participate
in
the
study.

Your
participation
is
entirely
volunta~
and
you
may
refuse
to
answer
any
questions.
Because
only
l
 
small
number
of
people
are
being
selected
for
the
study,
the
participation
of
l
 
ach
person
selected
is
extremely
important.
Most
of
the
questions
have
to
do
with
your
attitudes
and
opiniona
and
there
are
no
right
or
wrong
answers.
The
information
which
you
provide
will
be
kept
strictly
confidential
and
will
be
used
only
for
oversll
statistical
results.
If
you
would
like,
we
will
send
you
a
suaaaary
of
the
results
of
the
study.

CNECX
APPROPRIATE
BOX
BELOW
AND
IS?
"
YES*'
PRINT
RESPONDENT
`
S
MAILING
ADDRESS
.
RESULTS
REQUESTED
:
YES
a
NO
a
Mail
ing
I
Number/
Street/
RFD
Apt
.
NO.

Address
11111!
City/
State
ZIP
INTERVIEW
START
TIME
:
AHf
PH
D­
2
A­
3
LEAVZ
CAM
1
IN
FRONT
OF
RESPONDENT.
GIVS
RESPONDENT
CARD
2,
"
LIST
OF
SITES.
"
Here
is
a
list
of
recreational
sites
in
the
area.
GIVE
RESPONDENT
CARD
3,
"
PICTORIAL
MAP
.
"
And
here
is
a
pictorxal
map
showing
the
location
of
theee
sites.
ALLOW
RESPONDENT
TIM
TO
LOfJK
AT
BOTH
CARDS.
THESE
THRJM
CARDS
SHOULD
REHAIN
IN
FRONT
OF
THE
RESPONDENT
THROUGHOUT
THE
INTERVIEW
.

How
many
times
within
the
past
twelve
months
did
you
visit
any
of
the
sites
listed
on
this
card
or
any
other
recreational
site
near
water?

AS
SITES
AM
MENTIONED,
RECORD
SITS
CODE
AND
NUlfBKR
OF
TIMES
THE
SITE
WAS
VISITED.
TEiN
ASK:
Which
activities
listed
on
Card
1
did
you
partLc~­
pate
in
at
that
site
during
the
last
12
months?

CIRCLS
THE
ACTIVITY
NUMBER(
S)
Iii
THE
COLUMN
ACROSS
FROM
THE
SITE(
S)
.
MNTIONED

IF
UNLISTED
SITES
ARE
MENTIONED,
ENTER
SITE
NAM
ON
LINE
AND
NUMBER(
S)
OF
VISITS
ANO
ACTIVITIES.

$
$
s
$
~
:
~
i
=
:
G$
4
1
~
g
g
~
g
s
g
g
~
g
~
i
i
;
~
~
;
:
g
3
g
5
~
g
~
g
~
site
N­
o
sit­
No.
0
:`!
a
Noc
Li
aced
Codes
visit
g
j
;
$
~
~
?$
:
#
g
j
$
;
~

01
02
03
04
05
06
07
08
09
10
11
12
13
14
15
01
02
03
06
0s
06
07
0s
09
10
11
12
13
14
15
01
02
03
06
0s
06
07
0s
09
;
0
11
12
13
14
1s
01
02
03
ok
03
06
07
08
09
10
11
12
13
16
M
01
02
03
06
0s
06
07
0s
09
10
11
12
13
M
Is
01
01
03
06
0s
06
07
08
09
10
11
12
13
1A
M
01
02
03
M
03
06
07
08
09
10
11
12
13
1~
Is
01
02
03
04
05
06
07
0s
09
10
11
12
13
16
1s
01
02
03
06
05
06
07
08
09
10
11
12
13
16
Is
01
0'
2
03
06
03
06
07
0s
09
10
11
12
13
16
15
01
02
03
06
05
06
07
08
­
09
10
11
12
13
16
1s
01
02
03
06
03
06
07
0s
09
10
11
12
13
lb
15
01
02
03
ok
0s
06
07
0s
09
10
11
12
13
16
1s
01
02
03
04
0s
06
07
0s
09
10
11
12
13
16
Is
01
02
03
06
05
06
07
0s
09
10
11
12
13
14
15
01
02
03
w
05
06
07
0s
09
lC
11
12
13
lb
M
ECORD
w
YL!
v!­
M
+­
l­+++

D­
3
A.
RIKREATIONAL
ACTIVITIES
A­
1
a.
First,
do
you
own
or
have
the
use
of
any
kind
of
boat?
NUMBER
.

TM.
.
.
.
.
.
.
.
.
.
.
.
01
(
GO
TO
A­
l
b.)

NO
.
.
.
.
.
.
.
.
.
.
.
.
02
(
GO
TO
A­
2)
CIRCLE
b.
Which
of
the
following
describes
the
boat
you
use
meet
often?
SIAO
ANSWER
CHOICES
ANO
CIRCLE
NUKBER.

SAILBOAT
.
.
.
.
.
..­.
O1
INBOARD
.
.
.
.
.
.
.
.
..
O2
OUTBOARD
.
.
.
.
.
.
.
..
O3
CANOE
.
.
.
.
.
.
.
.
.
..
O~

OTSeR
(
sPl!
CIH)
.
.
.
.
.
.
05
A­
2
The
next
few
questions
we
would
like
to
aek
deal
with
outdoor
recreational
activities
which
people
take
part
in
near
Lakee
and
rivers
in
this
area;
that
is,
the
activities
showm
on
this
card.
GIVE
RESPONDENT
u
1,
"
ACTIVITT
CARD".
Please
look
carefully
over
the
list
of
activities,
keeping
in
mind
that
all
the
activities
listed
refer
to
activities
near
lakes
or
rivers.
ALLOW
RESPONDENT
TIM
TO
LOOK
AT
TEE
LIST.

Within
the
past
12
months,
that
is
since
last
November,
did
you
take
part
in
any
of
the
activities
listed?
CIRCLE
NUMBER.

Yes.
.
.
.
.
.
.
.
.
.
.
.
01
(
GO
TO
A­
3)

NO
.
.
.
.
.
.
.
.
.
.
.
.
02
(
GOTO
B­
1)
23)

24)

(
2s)

D­
4
B.
BEN&
FITS
MEASURSS
B­
1
The
next
group
of
questions
is
about
the
quality
of
water
in
the
Monongahela
River.
Congress
passed
water
pollution
control
laws
in
1972
and
in
1977
to
improve
the
nation'
s
water
quality.
The
states
of
Pennsylvania
and
West
Virginia
have
also
been
involved
in
water
quality
improvement
programs
of
their
own.
These
programs
have
resulted
in­
cleaner
rivers
that
are
better
places
for
fishing,
boating,
and
other
outdoor
activities
which
people
take
part
in
near
water.
We
all~
y
for
these
water
quality
improvement
programs
both
as
taxpayers
and
as
consumers.

In
this
study
we
sre
concerned
with
the
water
quality
of
only
the
Monongahela
River.
Keep
in
mind
that
people
take
part
in
all
of
the
activities
on
Card
1
both
on
and
near
the
water.

Generally,
the
better
the
water
quality,
the
better
suited
the
water
is
for
recreational
activities
snd
the
q
 
ore
likely
people"
will
take
part
in
outdoor
recreational
activities
on
or
near
the
wster.
Here
is
a
picture
of
a
ladder
that
showa
various
levels
of
water
quality.
CARD
4,
GIVE
RESPONDENT
"
WAKER
QUALITY
LADDSR"
.

The
top
of
the
ladder
stands
for
the
best
uossible
quality
of
water,
The
bottom
of
the
ladder
stands
for
the
w=
possible
water
quality.
On
the
ladder
you
can
see
the
different
lev~
f
the
quality
of
the
water.
For
l
 
xample:
(
POINT
TO
S&
X
LEVSL
­­
E,
D,
C,
B,
A
­­
AS
YOU
READ
THE
STATEMENTS
BBLOW
.
)

Level
"
E"
(
POINTING)
ia
so
polluted
that
it
has
oil,
raw
sewage
and
other
things
like
traah
~
n
it;
it
haa
no
plant
or
animal
life
and
smells
bad.

Water
at
level
"
D"
is
okay
for
boating
but
not
fishing
or
swimming.

Level
"
C"
shows
where
the
water
is
clean
l
 
nough
ao
that
game
fish
like
baas
can
live
in
it.

Level
"
B"
shows
where
the
water
ia
clean
l
 
nough
so
that
people
can
swim
in
it
safely.

And
at
level
"
A",
the
quality
of
the
water
ia
so
good
that
it
would
be
possible
to
drink
directly
from
it
if
you
wanted
to.

a.
Now,
think
about
the
water
quality
of
the
Monongahels
River
on
the
whole.
In
terms
of
this
scale,
from
zero
to
ten,
how
would
you
rate
the
water
quality
of
the
?
40nongahela
River
at
the
present
time?
POINT
TO
THB
ZSRO­
TO­
TEN
SCALE
ON
TIE
LADDER
AND
CIRCLS
NUMBSR.

00
01
02
03
04
05
06
07
08
09
10
(
GCI
TO
B­
1.
h.)

EON'
T
KNOW
.
.
.
.
.
.
.
.
.
.
.
.
.
..
ll(
cQTo
B­
2)

b.
Ia
your
rating
for
a
particular
site
on
the
river?
CIRCLE
NUMBER.

Yes
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
01
(
GO
TO
B­
I.
e.)

NO
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
..
02
(
Go
TO
B­
2)
cards
(
23­
24)

(
2s)

I
D
­
5
c.
On
the
map,
please
show
me
which
river
site
your
rating
applies
to.
,
 
.

B­
2
Site
Code:
m
IF
NOT
ON
LIST
OF
RECREATIONAL
SITES,
SPECIFY:

I
Another
important
purpose
of
this
study
is
to
learn
how
much
the
quality
of
water
of
the
Monongahela
River
ia
worth
to
the
people
who
live
in
the
river
basin.
In
answering
this
question,
there
are
three
waya
of
thinking
about
water
quality
that
q
 
ight
influence
your
decision.
GIVE
RESPONDENT
CARD
5,
"
VALUE
CARD".
The
three
ways
are
shown
on
this
card.

a.
~,
you
might
think
about
how
much
water
quality
is
worth
to
you
becauae
you
use
the
river
for
recreation.
POINT
TO
PART
I
OF
VALUS
CARD
AND
GIVE
RESPONDENT
TIME
TO
MAD
THAT
PART.

How
important
a
factor
is
your
actual
use
of
the
river
in
q
 
aking
a
decision
about
how
much
clean
water
ia
worth
to
you?
CIRCLE
NUMBSR.

VERY
IMPORTANT
.
.
.
.
..
O1
SOMINNAT
IMPORTANT
.
.
.
.
02
NEITHER
IMPORTANT
NOR
UNIMPORTANT
.
.
.
.
.
..
O3
NOT
WRY
IMPORTANT
.
.
.
.
04
NOT
MPORTANT
AT
ALL
.
.
.
05
b.
Another
way
you
might
think
about
how
much
clean
water
ia
worth
to
you
is
that
it
is
worth
something
to
you
to
know
that
a
clean
water
river
is
bein8
maintained
for
your
uae
if
you
should
decide,
in
the
futhre,
that
you
want
to
use
it.
POINT
TO
PART
II
OF
VALUE
CARD
AND
GIVE
RESPONDENT
T2XE
TO
RMO
lYIAT
PART.
For
example,
You
might
buy
an
advance
ticket
for
the
Steelers
or
Pirates
just
to
be
able
to
go
to­
a
home
game
if
you
later
decide
you
want
to
go.
Likewise,
you
might
pay
some
amount
each
year
to
have
a
clean
water
river
available
to
use
if
you
should
decide
to
use
it.

In
decidtig
how
much
clean
Water
ia
worth
to
you,
how
important
a
factor
ia
knowing
that
a
clean
water
river
is
being
maintained
for
your
uae,
if
you
should
decida
to
uae
it?
CIRC2JI
NUMBER.

VERY
IMPORTANT
.
.
.
.
..
OI
SOMEWHAT
IMPORTANT
.
.
.
.
02
NEITHSR
IMPORTANT
NOR
UNIMPORTANT.......
O3
NOT
VERY
IKPORTANT
.
.
.
.
06
NOT
IMPORTANT
AT
ALL
.
.
05
(
26­
27)

!
8­
29)

30­
31)

D­
6
c.
A
third
thing
you
might
think
about
in
deciding
how
much
clean
water
is
worth
to
you
is
the
satisfaction
of
knowing
that
a
clean
water
river
is
there.
POI.
NT
TO
PART
III
OF
VALUE
CARD
AND
GIVE
RESPONDENT
TIME
TO
RMO
THAT
PART.
For
example,
you
might
be
willing
to
pay
something
to
 
q
 
aintain
a
public
park
even
though
you
know
you
won'
t
use
it.
The
same
thing
could
be
true
for
clean
water
in
the
tlonongahela
that
is,
you
q
 
ight
pay
something
just
for
the
satisfaction
of
knowing
that
it
is
clean
and
that
others
can
use
it.

In
deciding
how
much
clean
water
is
worth
to
you,
how
important
is
knowing
that
a
clean
water
river
is
being
q
 
aintained?
CIRCLE
N­
UKSER
.

VERY
IMPORTANT
.
.
.
...
01
SOMZWHAT
IMYORTANT
.
.
.
.
02
NEITHER
IMPORTANT
NOR
U
N
I
M
P
O
R
T
A
N
T
.
.
.
.
.
.
.
O
3
NOT
VERY
IMPORTANT
.
.
.
.
04
NOT
IHPORTANT
AT
ALL
.
.
.
05
INTRODUCTION
TO
QUESTION
B­
3
Now,
we
would
like
for
you
to
think
about
the
relationship
between
improving
the
quality
of
water
in
the
Monongahela
River
and
what
we
all
have
to
pay
each
year
as
taxpayers
and
as
consumers.
We
all
pay
directly
through
our
tax
dollars
each
year
for
cl~
ing
up
all
rivers.
We
also
pay
indirectly
l
 
ach
year
through
higher
prices
for
the
products
we
buy
because
it
costs
companies
money
to
clean
up
water
they
use
in
making
their
products.
Thus,
each
year,
we
are
paying
directly
and
indirectly
for
improvements
in
the
water
quality
of
the
?
lonongahela
River.

I
want
to
aak
you
a
few
questions
about
ubat
amount
of
money
you
would
be
willing
to
pay
each
year
for
different
levels
of
water
quality
in
the
Monongahela
River.
Pleaae
keep
in
mind
that­
the
amounta
you
would
pay
each
year
would
be
paid
in
the
form
of
taxes
or
m
the
form
of
higher
prices
for
the
products
that
companiea
sell.

We
are
talking
about
different
levels
of
water
quality
for
only
the
!
lonongahela
River,
with
water
quality
at
other
sites
on
Card
2
staying
the
same
as
it
is
now.

I
also
want
you
to
keep
in
mind
the
recreational
activities
that
you
now
do
and
that
you
m=
do
in
the
future
on
the
?
lonongahela
River
or
at
other
sitey
That
is,
keep
in
mind
the
first
two
parts
of
the
value
card.
(
POINT
TO
TBE
VALUE
CARD.
CARD
5.
)
Your
actual
use
or
uossible
use
can
involve
activities
in
the
water
or
near
the
water,
or
both,
a:
we
talked
about
earlier.

We
bow
that
for
the
Moaongahela
River
as
a
whole
the
current
water
quality
is
at
level
"
D",
but
that
it
may
vary
at
different
points
along
the
river.
At
level
"
D"
it
is
clean
l
 
nough
for
boating,
but
not
clean
l
 
nough
for
catching
game
fish
or
for
sw~
ing.

NAVE
RENINDER
CARD
READY.
RECORD
DOLLM
AMOUNTS
GIVEN
FOR
EACH
PART
Asm
.

D­
7
1
(
32­
33)

I
5
B­
3
a.
This
payment
card
shows
different
yearly
amounts
people
might
be
willing
to
pay
for
different
levels
of
water
quality.
HAND
RESPONDENT
CARD
6,
"
PAYMENT
CARD
,
"
AND
ALLOW
RESPONDENT
TIM
TO
LOOK
AT
IT.

What
ia
the
most
it
is
worth
to
you
(
and
your
family)
on
a
yearly
basis
to
keep
the
water
quality
in
the
Ilonongahela
River
from
slipping
back
from
level
"
Q"
to
level
"
E",
where
it
is
not
even
clean
enough
for
boating?
Please
pick
any
amount
on
the
card,
any
amount
in
­

between,
or
any
other
amount
you
thiok
ia
appropriate.

$
IF
ANY
AHOUNT,
GO
TO
B­
3.
b.;
`
u
(
IF
ZKRO
DOLLARS,
ASK
1
.
)

L
b.
Would
it
Be
worth
something
to
you
(
and
your
family)
to
raiae
the
water
quelity
level
from
level
"
D"
to
a
higher
level?
CIRCLE
NUMBSR.

YES
.
.
.
.
.
.
.
..
O1
(
CO
TO
B­
3.
b.)

NO
.
.
.
.
.
.
.
.
.
02
(
GO
TO
B­
3.
e.)'
I
(
In
addition
to
the
amount
You
just
told
me,
)
What
ia
the
most
that
you
would
be
willing
to
pay
each­
year
in
hi&
er
taxes
and
prices
for
products
that
companies
sell
to
raise
the
water
quality
from
level
"
D"
to
level
"
C",
where
game
fish
can
live
in
it
and
it
is
improved
for
other
activities?

$
IF
ANY
AMOUNT,
00
TO
B­
3.
c.;
`
u
(
IF
ZKRO
DOLLARS,
GO
TO
B­
3.
d.
)

c.
How
much
more
then
(
READ
AMOUNT
FROb
b.
~
would
you
be
willing
to
pay
each
year
in
higher
taxes
and
prices
for
products
that
companies
sell
to
raise
the
water
quality
from
level
"
C"
to
level
"
B",
where
it
is
clean
l
 
nough
for
swiming
and
it
is
improved
for
other
activities

$
DOLLARS
(
CO
TO
B­
4)

D­
8
34­
s6)

`
37­
39)

(
40­
41)
d.
What
is
the
most
that
you
would
be
willing
to
pay
each
year
in
higher
taxes
and
prices
for
products
that
companies
sell
to
raise
the
water
quality
from
level
"
D"
to
level
"
B",
where
it
is
clean
enough
for
swirming
and
it
is
improved
for
other
activities?

/
IF
ANY
Amounting.,
GO
TO
B­
4;
\

$
(
IF
ZERO
DOLLARS
IN
a.
AND:
DOLLARS
.
\
ANT
AMOUNT
IN
d.,
GO
TO
B­
4.
d.;
ZERO
DOLLARS
IN
d.,
GO
TO
B­
3.
e./

e.
We
have
found
in
different
reasons
studies
of
this
for
answerina
as
type
that
people
have
a
lot
of
they
do.
So~
people
felt
they
did
not
have
enough
informsci&
to
give
a
dollar
amount,
some
did
not
want
to
put
dollar
values
on
l
 
nvironmental
quality,
and
some
objected
to
the
way
the
question
was
presented.
Others
gave
a
zero
dollar
amount
because
that
was
what
it
was
worth
to
them.

Which
of
these
reasons
best
describes
why
you
answered
the
way
you
did?
REPEAT
REASONS
IF
NECESSARY
AND
CIRCLE
NUNBER.

NOT
ENOUGN
INFORMATION
.
.
01­

DID
NOT
WANT
TO
PLACS
DOLLAR
VALUE
.
.
.
.
..
O2
OBJECTSD
TO
WAY
QUESTION
WAS
PRESENTED
.
.
.
.
..
O3
THAT
IS
WHAT
IT
IS
WJRTR
.
04
O'TNER
(
SPECIFY)
.
.
.
.
.
.
05
(
GO
TO
B­
6)
(
43­
45;

[
45­
47,

D­
9
_
.

B­
6
REFER
TO
REMINDER
CARD.
DO
NOT
ASK
QUESTIONS
CORRESPONDING
TO
ZERO
.
AMOUNTS
ON
CARD.

a.
IIS
answering
the
next
question(
s),
keep
in
mind
your
actual
and
possible
future
use
of
the
?
lonongshela.
You
told
me
in
the
last
section
that
it
was
worth
$(
AMOUNT)
to
keep
the
water
quality
from
slipping
from
level
"
D"
to
level
"
E".
How
much
of
this
amount
was
based
on
your
actual
use
of
the
river?

$
(
48­
50)

b.
You
(
also)
told
me
that
you
would
be
willing
to
pay
$(
AMOUNT)
co
raise
the
water
quality
from
level
"
D"
to
level
"
C".
POINT
TO
LEVZLS
"
n"
AND
"
c".
How
much
of
this
amount
waa
due
to
your
actual
use
of
the
river?

$
(
51­
s3)

c.
You
(
also)
told
me
that
you
would
be
willing
to
pay
$@
MOUN'T)
to
raise
the
water
quality
from
level­
"
C"
to
level
"
B".
POINT
TO
LEVELS
"
C"
AND
"
B".
How
much
of
this
amount
was
due
to
your
actual
use
of
the
river?

$
(
GO
TO
B­
5)
(
54­
s6)

d.
You
told
me
in
the
last
question
that
you
would
be
willing
to
pay
$(
MOUNT)
to
raise
the
water
quality
from
level
"
D"
to
level
"
B".
POINT
TO
LEVFLS
"
D"
AND
"
B".
How
much
of
this
amount
was
due
to
your
actual
use
of
the
river.

$

L­
(
57­
s9)

B­
5
REFER
TO
REMINDER
CARD.
You
have
said
that
you
would
be
willing
to
pay
$
@
MOUNT~
to
keep
the
level
of
water
quality
from
slipping
from
level
"
D"
to
level
"
E"
and
you
said
that
you
would
be
willing
to
pay
$
Jb.
PLUS
c.
~
OR
d.)
to
raise
the
level
from
level
"
D".
This
is
a
total
of
@
EAD
TOTAL
m.

Let'
s
thimk
about
another
way
that
the
quality
of
water
in
the
Monongahela
River
could
affect
your
recreation
on
or
aear
water.
I
would
like
you
to
think
about
how
the
river
being
closed
for
certain
activities
for
different
periods
of
time
would
change
the
JRMD
TOTAL
$
AMOUNT)
you
would
be
willing
to
pay
per
year.
Suppose
the
government
is
considering
relaxing
the
water
pollution
control
laws,
but
aot
totally
removing
them.
This
would
mean
that
the
overall
quality
of
the
water
in
the
tfonongahela
River
would
decrease
to
level
"
E"
where
it
would
be
closed
some
weekends
for
activities
on
or
in
the
water
like
boating,
fishing
and
swismsing
and
you
would
not
know
it
was
closed
until
the
day
you
wanted
co
gO.
`
The
area,
however,
would
remain
open
all
weekends
for
activities
near
the
water,
like
jogging
or
hiking
or
picnicking.

D­
10
a.
If
the
water
pollution
laws
were
relaxed
to
the
point
that
the
water
quality
would
decrease
to
level
"
E"
and
the
area
would
be
closed
1/
4
of
the
weekends
of
the
year
for
activities
on
or
in
the
water
but
would
remain
open
for
activities
near
the
water,
how
much
would
you
change
this
(
READ
TOTAL
$
AMOUNT)
to
keep
the
area
open
all
weekends
for
all
activities?

$
00LLAR
CNANGB
b.
If
the
area
would
be
closed
for
activities
on
or
in
the
water
1/
2
of
the
weekenda,
how
much
would
you
change
this
(
READ
TOTAL
$
MOUNT)
to
keep
the
area
open
all
weekenda
for
all
activities?

s
001.
LAR
CHANGE
c.
If
the
area
would
be
closed
for
activities
on
or
in
the
water
3/
4
of
the
weekends,
how
much
would
you
change
this
(
READ
TOTAL
$
AMOUNT)
to
keep
the
ares
open
all
weekends
for
all
activities?

s
DOLLAR
CHANGE
B­
6
a.
If
the
water
quality
in
the
Monongahela
River
were
improved
from
Level
"
D!!
to
level
"
B",
where
it
is
clean
enough
for
sw~
ing
and
it
is
improved
for
other
activities,
how
would
this
affect
your
annual
use
or
future
use
of
sites
along
the
river?
CIRCLE
NUMBER.

IWREASE
USE
BY
HORS
THAN
5
VISITS
PSR
YEAR
.
.
.
01
INCREASE
USE
BY
1
TO
5
VISITS
PSR
YEAR.
.

NO
CHANGE
LO
USE
.
.
.
.
.
.
.
.
.
.
.
.
.

lN%
CRMSE
USE
ALONG
~
MONONGAEELA
RIVER.

DON'T
SNOW
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
02
.
.
.
.
03
.
.
.
.
04
.
.
.
.
05
b.
How
would
this
change
from
"
D"
to
"
B"
in
the
Honongahela
River
affect
your
annual
use
or
future
uae
of
other
recreational
sites
near
water,
but
not
along
the
tfonongahela
River?
CIRCLS
NUMBER.

DECREASE
USE
BY
~
RE
TBAN
5
VISITS
PER
YEAR
.
.
.
01
DECREASE
USE
BY
1
TO
5
VISITS
PER
TEAR.
.
.
.
.
.
02
NO
CHANGE
IN
USE
.
.
.
.
.
.
.
.
.
.
.
.
.
.
...
03
RWREMEU
SE..
.
.
.
.
.
.
.
.
,
.
.
.
.
.
...
04
CON'
TKNW
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
...
05
D
­
I
I
`
60­
62)

f63­
6S)

(
66­
68)

(
69­
70)

(
71­
72)
B­
7
B­
8
Up
to
now
we
have
talked
about
water
quality
baaed
on
your
use
and
possible
future
use
of
the
Plonongahela
River.
Let'
s
again
think
about
the
third
part
of
the
value
card.
That
is,
it
is
worth
something
just
to
know
a
river
with
clean
water
is
there
without
actually
using
it
or
planning
to
use
it.
We
want
you
to
think
~
in
terms
of
this
satisfaction
which
excludes
any
use
by
you
of
the
river.
With
this
in
mind,
suppose
the
government
were
to
remove
the
water
pollution
laws
entirely.
This
would
mean
lower
taxes
and
would
allow
companies
to
produce
their
products
at
lower
prices.
~,
it
would
also
mean
that
during
most
of
the
rest
of
your
lifetime
the
I'lonongahela
River
would
be
at
level
"
E"
and
would
not
be
usable
for
recreational
activities.
The
change
could
be
reversed
in
your
lifetime
but
it
would
cost
a
great
deal
of
money.

a.
What
is
the
most
that
you
(
and
your
family)
would
be
willing
to
pay
each
year
in
the
form
of
higher
taxes
and
prices
for
the
goods
you
buy
for
keeping
the
river
at
level
"
D"
where
it
is
okay
for
boating,

l
 
ven
if
you
would
never
use
the
river?
 
 
$
(
IF
ANY
AMOUNT,
GO
TO
B­
7.
b.;
IF
ZERO
lMLLARS,
GO
TO
B­
8)
)

b.
Suppose
the
change
could
not
be
reversed
for
an
l
 
ven
longer
period
of
time
than
your
lifetime.
How
much
more
than
@
AD
.
MIOUNT
FROH
a.~
would
you
(
and
your
family)
be
willing
to
pay
per
year
to
keep
the
river
at
level
"
D",
even
if
you
would
never
use
the
river?

GIVE
RESPONDENT
TIE
FOUR
CARDS
FROM
TRE
CARD
SET
7.
I
would
now
like
you
to
look
at
these
cards
which
show
different
combinations
of
levels
of
water
quality
and
amounts
in
higher
taxes
and
prices
it
would
coet
every
family
each
year
to
have
the
indicated
water
quality
levels.

a.
First,
I
would
like
you
to
rank
the
combinations
of
water
quality
levels
and
amounts
you
might
be
willing
to
pay
to
obtain
those
levels
in
order
from
the
card,
or
combination,
that
you
most
prefer
to
the
one
you
lesst
prefer.
I
would
like
you
to
do
this
based
only
on
your
use
l
 
nd
possible
use
in
the
future
of
the
Monongahela
River.
That
is,
keeping
in
mind
&
Parts
I
and
II
of
the
value
card.
POINT
TO
VALUE
CARD
­
PARTS
I
AND
II.
RSCORD
RANKING
OF
CAMS
BY
CIRCLED
WATER
QUALI~
LEVELS
ANTI
DOLLAR
AMOUNTS.

WATSR
RANRING
QUALIIT
$
AMOUNT
LEVEL
Host
Preferred
$

2nd
$

3rd
$

Least
Preferred
$

D­
12
73­
7s)

`
76­
78)

;
02.
?
0=
5
Card
6
1­
22
*.

(
24­
27)

[
28­
31)

(
32­
3S1
(
36­
39)

 
­
..
 
.
,
 
.
.
 
.
.
.
­
.
b.
Now,
I
would
like
you
to
repeat
this
procedure
but
assume
this
time
that
you
will
not
use
the
river
now
or
in
the
future.
That
is,
think
about
onl~
Part
III
of
the
value
card.
POINT
TO
VALUE
CARD
­
PART
III.
RECOi?
D
RANKING
OF
CARDS
BY
CIRCLED
WATER
QUALITY
LSVELS
AND
OOLLAR
AMOUNTS.

WATER
RANKING
QUALITT
$
AMOUNT
LEVSL
Most
Preferred
$

2nd
$

3rd
$

Least
Preferred
$
(
40­
43)

(
44­
47)

(
48­
51)

(
52­
5S)

i?
oz.
80=
6
c.
BACKGROUND
DATA
I
have
a
few
more
questiona
that
will
help
our
research
staff
analyze
the
results
of
the
study
properly.

C­
1
How
long
have
you
lived
in
the
Monongahela
River
basin
area?
CIRCLS
NUl!
BER
.

LESS
TRANIYEAR
.
.
.
.
..
O1
1
YEAR
OR
LONGER
BUT
LESS
TEAN3YEARS
.
.
..
O2
3
YSARS
OR
LONGER
BUT
LESS
THAN
5YEARS
.
.
..
O3
5YEARSOR
LONGSR
.
.
.
..
O.
4
C­
2
Now
I
am
going
to
read
some
phrases
that
describe,
different
kinda
of
interesta
people
have.
Aa
I
read
each
one,
please
tell
me
how
q
 
uth
the
phraae
is
like
you;
that
is,
a
lot
like
you,
somewhat
like
you,
a
little
.
 
like
you,
or
not
at
~
like
you.
.
 
CIRCLS
ONE
NUKSER
ON
EACH
LINE.
 
 
REPEAT
ANSW
CHOICSS
AS
NECESSARY.
SOHE
A
NOT
NO
A
LOT
WHAT
LITTLE
AT
ALL
OPINION
 
 
 
 
 
a.
AN
OUTDOORS
PERSON
.
.
.
..
O1
.
.
02
.
.
.
03
...
04...
05
b.
AN
ENVIRONMENTALIST
.
.
.
..
O1
.
.
02
.
.
.
03
.
..
04...
05
c.
SO?
ISONE
WNO
IS
AGAINST
NUCLXAR
mwn
FOR
ELECTRIC
PLANTS
.
.
.
.
.
..
O1
.
.
02
.
.
.
03
.
..
06...
05
d.
SOMEONE
WHO
IS
CONCERNED
ABWTWATERP
OLLUTION
.
.
..
O1
.
.
02
.
.
.
03
...
04...
05
l
 
.
SOtfBONE
WNO
IS
WILLING
TO
PAY
THE
COST
REQUIRED
TO
CONTROL
WATER
POLLUTION
.
..
O1
.
.
02
.
.
.
03
...
04...
05
C­
3
Which
of
the
following
best
describes
your
present
statua?
READ
CHOICES
AS
NECESSARY
AND
CIRCLS
`~.

SMPLOYED
FULL­
TII'fS.
.
.
.
.
01
EMPLOTED
PART­
TIME.
.
.
02
I
(
GO
TO
C­
5)

RETIRED
.
.
.
.
.
.
.
.
..
O3
NOT
EMPLOTSD
.
.
.
.
.
.
..
O4
AHOUSEWIFS
.
.
.
.
.
.
..
O5
ASTUDSNT
.
.
.
.
.
.
.
..
O6
OTHER
(
SPECIFf)
.
.
.
.
07
(
GO
TO
C­
4)
`
C7zd
7
:­
22
q.

`
24­
25)

(
26­:
7)

(
28­
29)

(
30­
31)

(
32­
33)

(
34­
351
(
36­
37)

D­
14
c­
4
c­
5
c­
6
Have
a.

b.

c.

d.
you
ever
been
employed?
CIRCLE
YES­
.
.
.
.
.
.
.
.
.
.
.01
NO.
.
.
.
.
.
.
.
.
.
.
.
.02
What
kind
of
work
(
do/
did)
you
called?
NumER.

(
GO
TO
C­
5)

(
Go
TO
C­
6)

do;
that
is,
what
(
s/
was)
your
j
o
b
What
(
do/
did)
you
actually
do
in
that
job?
What
(
are/
were)
some
of
your
main
duties
and
responsibilities?

tit
kind
of
an
organization
(
do/
did)
you
work
for?
(
PROBE
:
$/
hat
do
they
make,
what
do
they
do?)
BE
SURE
TO
NOTE
IF
RESPONDENT
IS
AN
StlPLOYEE
OF
GOVERMENT
AT
ANY
LSVEL
,
INCLUDING
TSE
SCH~
L
SYSTEM.

How
many
hours
(
do/
did)
you
work
at
your
job
in
a
usual
week?

NUMBER
OF
HOURS
WilRXED
IN
A
WESK
What
was
the
last
grade
of
regular
school
that
you
completed
­­
not
counting
specialized
schools
like
secretarial,
art,
or
trade
schools?
CIRCLE
NUMBER.

NO
SCHOOL
.
.
.
.
.
.

GRADE
SCNOOL
(
l­
8).
.

SOME
HIGE
SCHOOL
(
9­
1
HIGH
SCHOOL
GRADUATE
Som
COLLEGE
(
13­
15).

COLLEGE
GRADUATE
(
16)

POST
GMDUATE
(
17+)
.

NO
SESPONSEIRE.
FUSED
.
.
.
01
.
.
02
)
.
.
03
12)
.
04
.
.
.
05
.
.
.
06
.
.
07
.
.
.
08
D­
15
~
8­
39)

10­'
42)

46­
48)

49­
50)

51­
52)
C­
7
ASK
0NL%
IF
NOT
OBVIOUS.
How
would
you
describe
your
racial
or
ethnic
background?
READ
CHOICES
ANTI
CIRCLE
NUHBE.
R.

k?
IITE
OR
CAUCASIAW.
.
.
.
.
01
BLACK
OR
NEGRO.
.
.
.
.
.
.
02
oTHER
(
SPECFIY)
.
.
.
.
.
.
03
C­
8
Here
ia
a
list
of
income
categories.
HAND
RESPONDENT
CARD
8.
Would
you
call
off
the
code
number
of
the
categoqr
that
best
describes
the
combined
income
that
you
(
and
all
other
members
of
your
family)
received=
1980.
Please
be
sure
to
include
wages
and
salaries,
or
net
income
from
your
business,
and
pensions,
dividends,
interest,
and
l
 
ny
other
income.
CIRCLS
NUMRER.

UNDER
$
5,000
.
.
.
.
.
.
.
.
01
$
5,000
­
$
9,999
.
.
.
.
.
.
02
$
10,000
­
$
14,999
.
.
.
.
.
03
$
15,000
­
$
19,999
.
.
.
.
.
04
$
20,000
­
$
26,999
.
.
.
.
.
05
$
25,000
­
$
29,999
.
.
.
.
.
06
$
30,000
­
$
36,999
.
.
.
.
.
07
$
3
S,
000
­
$
39,999
.
.
.
.
.
08
$
60,000
­$
44,999
.
.
...
09
$
45,000
­
$
49,999
.
.
.
.
.
10
$
50,000AHDOVER.
.
.
.
.
.
11
NOT
SURE/
REFUSED.
.
.
.
.
.
12
C­
9
There
is
l
 
possibility
that
my
supervisor
would
like
to
call
you
to
verify
your
participation
in
this
study.
What
is
the
telephone
number
where
you
can
be
reached?

TELEPHONE
NmmER:
(
 
)
.

Thank
you
for
participating
in
this
study.

INTERVIEW
STOP
TIME:
A
n
/
m
(
53­
54)

(
5S­
S6)

Coz.
9097
OHS
#
2000­
0381
~
ppcovsl
Exp~
res:
9120102
ESTIMATING
BENZFITS
OF
UATSR
QUALITY
QUESTIONNAIRE
I,

1
Form
No.
02
(
i)

I.
IDENTIFICATION
INFORMATION
A.
Study
NO.~­
n
B
PfJJ/
Se­
ent
Non­~
(
2­
6)
(
8­
13)

c.
Housing
unit
N'J
m
D.
Intemiewer
ID
No.
(
1s­
17)
(
skip)

E.
Sample
Individual
Roster
Line
No.
m
`"
Questionnaire
Version
;?
9­
20)
B
II.
INTRODUCTION
(
22)

IF
TEE
ENUMERATION
RESPONDENT
IS
ALSO
THS
SELECTED
SAMPLE
INDIVIDUAL,
CONTINUS
YOUR
INTRODUCTION
TO
TNS
STUDY
BY
READING
THE
SECONTI
PARAGRAPH
BELOW.
IF
THE
SAMPLE
INDIVIDUAL
IS
SOHSONE
OTKER
THAN
THE
ENUJTION
RESPONDENT,
READ
THE
ENTIRE
INTRODUCTION
BSLOW
.

Hello,
I`
m
(
NAME~
from
the
Research
Triangle
Institute
in
North
Carolina.
We
are
doing
a
study
for
a
government
agency
to
study
levels
of
water
quality
and
some
outdoor
recreational
activities
people
take
part
in
both
near
and
on
ponda,
lakes,
streams
and
rivers
in
the
lfonongahela
River
Baain.
You
have
been
randomly
selected
to
participate
in
the
study.

Your
participation
is
entirely
voluntary
and
you
may
refuse
to
answer
any
questims.
Becauae
only
a
small
number
of
people
are
being
selected
for
the
study,
the
participation
of
each
person
selected
is
extremely
important.
Moat
of
the
questions
have
to
do
with
your
attitudes
and
opinions
and
there
are
no
right
or
wrong
answers.
The
information
which
you
provide
will
be
kept
strictly
confidential
and
will
be
used
only
for
overall
statistical
results.
If
you
would
like,
we
will
send
you
a
summary
of
the
results
of
the
study.

CHECX
APPROPRIATE
BOX
BELOW
AND
IF
"
YES"
PRINT
RESPONDENT'S
KAILING
ArNXUZss
.
RSSULTS
REQUESTED:
YESO
NOO
Hailing
\
Number/
Street/
RFD
Apt.
No.

Address
I
I
I
City/
State
ZIP
INTERVIEW
START
TIM:
ANf
Pn
D­
17
B­
3
a.
What
is
the
q
 
ost
it
is
worth
to
you
(
and
your
family)
on
a
yearly
basis
to
keep
the
water
quality
in
the
Ploaongahela
River
from
slipping
back
from'
level
"
D"
to
level
"
E",
where
it
is
not
even
clean
enough
for
boating?

$
DOLLARS
(
IF
ANY
AMOUNT,
GO
TO
B­
3.
b.;
IF
2ER0
OOLMRS
,
.
ASK
.
)

(

L
b.

c.
Would
it
be
worth
something
to
you
(
and
your
family)
to
raise
the
water
quality
level
from
level
"
D
g'
to
a
higher
level?
CIRCLE
NUMBRR.

Es
.
.
.
.
.
.
.
.
.
01
(
GO
TO
B­
3.
b.)

No
.
.
.
.
.
.
.
.
.
02
(
GO
TO
B­
3.
e.)

(
In
addition
to
the
amount
YOU
just
told
me,
)
What
is
the
most
tha
you
would
be
willing
to
pay
each
year
in
higher
taxes
and
prices
fo
uroducts
that
comanies
`
sell
to
raise
the
~
ater
quality
from
leve
%"
to
level
"
C",­
where
game
fish
for
other
activities?
can
live
in
it
and
it
is
improve
ANT
AMOUNT,
GO
TO
B­
3.
c.;
ZERO
DOLLARS,
GO
TO
B­
3.
d.
)

How
much
more
than
JREAD
AMOUNT
FROM
b.~
would
you
be
willing
to
pa
each
year
in
higher
taxes
and
prices
for
products
that
companie
sell
to
raise
the
water
quality
from
level
"
C"
to
level
"
B",
wher
it
is
clean
enough
for
swhing
and
it
is
improved
for
other
activi
ties?

s
DOLLARS
(
CO
TO
B­
4)
(
34­
36)

(
3?­
39)

(
40­
42)

.
D­
18
d.
b%
at
is
the
most
that
you
would
be
willing
to
pay
each
year
in
higher
taxes
and
prices
for
products
that
companies
sell
to
raise
the
water
quality
from
level
"
D"
to
level
"
B",
where
it
is
clean
enough
for
swimming
and
it
is
improved
for
ocher
activities?

(
IF
ANT
AMOUNT
IN
a.,
GO
TO
B­
4;
IF
ZERO
DOLLARS
IN
a.
AND:
)
$
DOLLARS
.
AI@
AMOUNT
IN
d.,
00
TO
B­
4.
d.;
.
ZERO
DOLLARS
IN
d.,
CO
TO
B­
3.
e.
/

e.
We
have
found
in
studies
of
this
type
that
people
have
a
lot
of
different
reasons
for
answering
as
they
do.
Some
people
felt
they
did
not
have
enough
information
to
give
a
dollar
amount,
some
did
not
want
to
put
dollar
values
on
l
 
nvironmental
quality,
and
some
objected
to
the
way
the
question
was
presented.
Others
gave
a
zero
dollar
amount
because
that
was
what
it
was
worth
to
them.

Which
of
these
reasons
best
describes
why
you
answered
the
way
you
did?
REPEAT
REASONS
IF
NECSSSARY
AND
CIRCZE
NUMBER.

NOT
SNOUOR
INFORMATION
.
.
01­

DID
NOT
WANT
TO
PL4CE
DOLLAR
VA.
LUE
.
.
.
.
..
O2
OBJECTED
TO
WAY
QUESTION
wABPREssN­
rEu
.
.
.
.
..
o3
TRAT
IS
WHAT
IT
IS
WRTR
.
0~

oTHER
(
SPECIFf)
.
.
.
.
.
05
D­
19
(
00
TO
B­
6)
(
43­
45)

(
46­
47)
:,.,
,:

~
#
2000­
0381
Approval
~
iecs:
9/
20f82
A.

c.

E.
ESTIMATING
BENEFITS
OF
WATER
QUALITY
QWSTIONNAIRE
Form
No.
02
(
1)

IDENTIFICATION
INFORMATION
Study
No.+­
a
B
PSU/
Segment
No.
cl­~
(
2­
6)
(
8­
13)

Housing
Unit
No.
[
I
l
l
D.
~:
fie"
er
UInn
(
15­
1?)
(
skip)

Sample
Individual
Roster
Line
No.
m
F.
Questionnaire
Version
(
19­
20)
c
II.
INTRODUCTION
(
22)

IF
THE
ENUMUL4TION
RESPONDENT
IS
ALSO
THE
SELECTED
SAMPLE
INDIVIDUAL,
CONTINUE
YOUR
INTRODUCTION
TO
TIE
STUDY
BY
READING
~
SECOND
PARAGRAPH
BELOW.
IF
TRE
SAMPLE
INDIVIDUAL
IS
SOt!
EONE
OTRER
THAN
THE
ENUMERATION
RESPONDENT,
READ
THE
ENTIRE
INTRODUCTION
BELOW.

Hello,
I`
q
 
=
from
the
Research
Triangle
Institute
in
North
Carolina.
We
are
doing
a
study
for
l
 
government
agency
to
study
levels
of
water
quality
and
some
outdoor
recreational
acclivities
people
take
part
in
both
near
and
on
ponds,
lakes,
streams
and
rivers
in
the
!
ionongahela
River
Basin.
You
have
been
randomly
selected
to
participate
in
the
study.

Your
participation
is
entirely
voluntary
and
you
may
refuse
to
snswer
any
questions.
Becsuse
only
a
small
number
of
people
are
being
selected
for
the
study,
the
participation
of
each
person
selected
is
extremely
important.
Most
of
the
questions
have
to
do
with
your
attitudes
and
opinions
and
there
are
no
right
or
wrong
answers.
The
information
which
you
provide
till
be
kept
strictly
confidential
and
will
be
used
only
for
overall
statistical
results.
If
you
would
like,
we
will
send
you
l
 
sunnary
of
the
results
of
the
study.

CNSCYi
APPROPRIATE
BOX
BELOW
ANO
IF
"
YES"
PRINT
RESPONDENT
`
S
WILING
ADrlRsss
.
RSSULTS
REQUESTED:
YESD
NOO
!
lailing
I
Number/
Street/
RED
Address
Apt.
No.

H
]

City/
State
ZIP
INTKRVIEW
START
TItil?:
AH/
Pm
.

D­
20
B­
3
a.
To
you
(
and
your
family),
would
it
be
worth
$
25
each
year
in
higher
taxes
and
prices
for
products
that
companies
sell
to
keep
the
water
quality
in
the
Monongahela
River
from
slipping
back
from
level
"
D"
to
level
"
E"?
CIRCLE
NUXBER.

~
YEs
.
.
.
.
.
.
.
..
ol
NO
.
.
.
.
.
.
.
..
O2
1
+
IF
YES,
INCRSASE
TH31
DOLIAR
MOUNT
IN
$
5
INCREMENTS
UNTIL
A
"
NO"
ANSWER
IS
GIVEN.
E.
G.,
"
Would
it
be
worth
$
30
each
year
to
keep
water
auslitv
from
slipping
from
level
­
~
~
S,'
D~
S~
eIv:
lG~~
r'
slipping
from
level
`
E'?"
ETC.
WNEN
A
ETC.
!,
!
GIVEN,
RSCORD
IX3LLAR
RECORO
DOLLAR
Amoom.
"
YES"
ANSWER.
`
D'­
to
`
level
"
NO"
ANSWER
IS
AUOUNT
OF
LAST
IF
NO,
DECREASE
TIDI
DOLLAR
AMOUNT
IN
$
5
INCREMENTS
UNTIL
A
"
YES"
ANSWER
IS
GIVEN.
E.
G.,
"
Would
it
be
worth
$
20
each
year
to
keep
water
quality.
from
(
IF
ANY
AMOUNT,
GO
TO
B­
3.
b.;
$
DOLLARS
IF
ZJIRO
DOLLARS
IS
FINAL
MOUNT,
ASK
7
)

I
I
k
Would
it
be
worth
something
to
you
(
and
your
family)
to
raise
the
water
quality
level
from
level
"
D"
to
a
higher
level?
CIRCLE
N1.
MSER
.

I
YES
.
.
.
.
.
.
.
.
.
01
(
GO
TO
B­
3.
b.)

I
NO
.
.
.
.
.
.
.
.
.
02
(
GO
TO
B­
3.
e.)

D­
21
L(
34­
36)
b.
Would
you
(
and
your
family)
be
willing
to
pay
(
an
additional)
$
2S
each
year
in
higher
taxes
and
prices
for
products
that
companies
sell
to
raise
the
water
quality
from
level
"
D"
to
level
"
C",
where
game
fish
can
live
in
it
and
it
is
improved
for
other
activities?
CIRCLE
NUMBER
.

r"'­

E?
YES,
INCRXASE
TNE
.
.
.
.
.
.
.
01
.
.
.
.
.
.
.
.
02
1
DOLLAR
AMOUNT
IN
IF
NO,
DECREASE
THE
DOLLAR
AMOUNT
IN
$
5
IN&
E?!
lNTS
UNTIL
A
"
NO"
ANSWER
IS
$
5
INiREKENTS
UNTIL
A
"
YES"
ANSWER
IS
GIVEN.
E.
G.,
"
Would
you
be
willing
GIVEN.
E.
G.
,
"
Would
you
be
willing
to
to
pay
$
30
(
more)
esch
year
to
raise
pay
$
20
(
more)
each
year
to
raise
the
the
water
quality
from
level
`
D'
to
water
quality
from
level
`
D'
to
level
level
`
C'
?"
ETC.
WHEN
A
"
NO"
ANSWER
`
C'?"
ETC.
IS
GIVEN,
WHEN
A
"
YES"
ANSWER
IS
RECORD
DOLLAR
AMOUNT
OF
GIVEN
,
RECORD
DOLLAR
AMOUNT.
LAST
"
YES"
ANSWER
.

(
IF
ANY
AMOUNT,
GO
TO
B­
3.
c.;
$
DOLLARS
)
IF
ZERO
DOLLARS
IS
FINAL
AMOUNT,
GO
TO
B­
3.
d.

c.
Would
you
(
and
your
family)
be
willing
to
pay
an
additional
$
2S
l
 
ach
year
in
higher
taxes
and
prices
for
products
that
companies
sell
to
raise
the
water
quality
from
level
"
C"
to
level
"
B",
where
you
can
swim
in
it
and
it
is
improved
for
other
activities?
CIRCLE
NUMBER.

I­­­
m
s"""
""""
o"
o'
NO
.
.
.
.
.
.
.
..
O2
1
+
*
IF
YES,
INCREASE
TIE
DOLLAR
AMOUNT
IN
IF
NO,
DECRSASE
THE
DOLLAR
AMOUNT
IN
$
5
INCREMENTS
UNTIL
A
"
NO"
ANSWER
IS
$
5
INCREMENTS
UNTIL
A
"
YES"
ANSWER
IS
GIVEN.
E.
G.,
"
Would
you
be
willing
GIVEN.
E.
G.,
"
Would
you
be
wiIling
to
to
pay
$
30
more
each
year
to
raise
pay
$
20
more
each
year
to
raise
the
the
water
aualitv
from
level
`
C'
to
water
quality
from
level
`
C'
to
level
level
`
B'?""
ETC.
­

WNEN
A
`*
NO"
ANSWER
`
B+?*
ETC.
"
IS
GIVIN
,
RECORD
DOLLAR
MOUNT
OF
~
A
"~
srt
AN'W'R
IS
GIVEN
,
RECORD
DOLLAR
AMOUNT.
ANSWER.

s
DOLIARS
(
GO
TO
B­
4)

.­

D­
22
(
37­.
39J
(
4&
4~)
d.
Would
you
(
and
your
family)
be
willing
to
pay
$
2S
each
year
in
higher
taxes
and
prices
for
products
that
companies
sell
to
raise
the
water
quality
from
level
"
D"
to
level
"
B",
where
you
can
swim
in
it
and
it
is
improved
for
other
activities?
CIRCLS
NUMBER.

rms'""""'"""
o'

I
NO
.
.
.
.
.
.
.
..
O2
1
+

IF
YES,
INCRSASE
TEE
DOLLAR
AMOUNT
IN
$
5
INCREMENTS
UNTIL
A
"
NO"
ANSWER
IS
GIVEN.
E.
G.,
"
Would
you
be
willing
to
pay
$
30
each
year
to
raise
the
water
quality
from
level
`
D'
to
level
`
B'?"
ETC.
WTIXN
A
"
NO"
ANSWSR
IS
GIVEN,
RECORD
00UAR
AMOUNT
OF
LAST
"
YES"
ANSWSR.
+

IF
NO,
DECREASE
THE
DOLIAR
AMOUNT
IN
$
5
INCREMENTS
UNTIL
A
"
YES"
ANSWSR
IS
GIVEN.
E.
G.,
"
Would
you
be
willing
to
pay
$
20
each
year
to
raiae
the
water
quality
from
level
`
D'
to
Level
`
B'?"
ETC.
kliEN
A
"
YES"
ANSWER
IS
GNEN,
RSCORD
DOLLAR
AMOUNT.

(
IF
ANY
MOUNT
IN
a.,
GO
TO
B­
4;

s
DOLLARS
IF
ZERO
DOLLARS
IN
a.
AND:
.
ANY
AMOUNT
IN
d.,
GO
TO
B­&.
d.;
)
.
ZERO
DOLLARS
IN
d.,
GO
TO
B­
3.
e.

l
 
.
We
have
found
in
studies
of
this
type
that
people
have
a
lot
of
different
reaaons
for
answering
as
they
do.
Some
people
felt
they
did
not
have
enough
information
to
give
a
dollar
amount,
some
did
not
want
to
put
dollar
values
on
environmental
quality,
and
some
objected
to
the
way
the
question
was
presented.
Others
gave
a
zero
dollar
amount
because
that
was
what
it
was
worth
to
them.

k?
iich
of
these
reasons
best
describes
why
you
answered
the
way
you
did?
REPEAT
REASONS
IF
NECESSASY
AND
CIRCLE
NUKBER.

NOT
ENOUGR
INFORMATION
.
.
01
DID
NOT
WANT
TO
PLACE
rmLARvALuR...
.
..
o2
OBJECTED
TO
WAY
QUESTION
msPREsENmD.
.
.
.
..
o3
THAT
IS
WNAT
IT
IS
WORTN
06
OTXER
(
SPECIFY)
.
.
05
D­
23
(
GO
TO
B­
6)
(
43­
45)

(
46­
47)
ESTIMATING
BENEFITS
OF
WATER
QUALITY
QUESTIONNAIRE
Form
Xo.
02
(
1)

1.
IDENTIFICATION
INFORMATION
4.
Study
!?
0.~
­

D
B
Psu/
sement
NO
 
q
 
­
~
(
2­
6)
(
8­
13)

c.
Housing
Unit
No.
[
Ill
D.
::
t~=
ewer
~
(
15­
17)
(.
Sip)

E.
Sample
Individual
Roster
Line
No.
m
F.
Questionnaire
Version
D
(
19­
20)

II.
INTRODUCTION
(
22)

IF
THE
ENUI!
EMTION
RESPONDENT
IS
ALSO
17E
SELECTED
SAMPLE
INDIVIDUAL,
CONTINUE
YOUR
INTRODUCTION
TO
THE
STUDY
BY
READING
TEE
SECOND
PARAGRAPH
BELOW.
IF
THE
SAMPLE
INDIVIDUAL
IS
SOHEONE
OTHER
TRAN
TIE
ENUMERATION
RESPONDENT,
READ
THB
ENTIRE
INTRODUCTION
BILLOW.

Hello,
I'IS
(
NAME)
from
the
Research
Triangle
Institute
in
North
Carolina.
We
are
doing
a
study
for
a
government
agency
to
study
levels
of
water
quality
and
some
outdoor
recreational
activities
people
take
part
in
both
near
and
on
ponds,
lakes,
streams
and
rivers
in
the
Plonongahela
River
Basin.
You
have
been
randomly
selected
to
participate
in
the
study.

Your
participation
is
l
 
mtirely
voluntary
and
you
may
refuse
to
answer
any
questiona.
Becauae
only
a
small
number
of
people
are
being
selected
for
the
study,
the
participation
of
each
person
selected
ia
extremely
important.
!
lost
of
the
questions
have
to
do
with
your
attitudes
and
opinions
and
there
are
no
right
or
wrong
answers.
The
information
which
you
provide
will
be
kept
strictly
confidential
and
will
be
used
only
for
overall
statistical
results.
If
you
would
like,
we
will
send
you
a
swmary
of
the
results
of
the
study.

CNECX
APPROPRIATE
BOX
BELOW
AND
SF
"
YES"
PRINT
DESPONDENT'S
MAILING
ADDRESS
.
RESULTS
REQUESTED:
ma
Non
Mailing
I
Number/
Street/
RFD
Apt.
No.
Address
I
I
1
City/
State
ZIP
INTERVIEW
START
TIME:
An/
Pu
D­
24
.
.
B­
3
a.
To
you
(
and
your
family),
would
it
be
worth
$
125
each
year
in
higher
taxes
and
prices
for
products
that
companies
sell
to
keep
the
water
quality
in
the
Monongahela
River
from
slipping
back
from
level
"
D"
to
level
"
E"?
CIRCLE
NUMBER.

r
Tits.
.
.
.
.
.
.
.01
No
.
.
.
.
.
.
.
.
.02
1
f
IF
YES,
INCREASE
TRE
DOLLAR
AMOUNT
IN
$
10
INCRSHEN'TS
UNTIL
A
"
NO"
ANSWER
IS
GIVEN.
E.
G.,
"
Would
it
be
worth
$
135
l
 
ach
year
to
keep
water
quality
from
slipping
from
level
`
E'?"
ETC.
WKEN
A
GIVEN,
RECORD
DOLLAR
"
YES"
ANSWER.
`
D;
to
`
level
"
NO"
ANSWER
IS
AMOUNT
OF
LAST
IF
NO,
DECNEASE
TRE
DOLLAR
AMOUNT
IN
$
10
INCREMENTS
UNTIL
A
"
YES"
ANSWER
IS
GIVEN.
E.
G.,
"
Would
it
be
worth
$
115
each
year
to
keep
water
quality
from
slipping
from
level
`
D'
to­
level­
`
E'?"
ETC.
WREN
A
"
YES"
ANSWER
IS
GIVEN,
RECORD
DOLLAR
AMOUNT.

/
IF
ANY
AMOUNT.
GO
TO
B­
3.
b.:
\
$
Dom
(
)
IF
ZERO
DOtiS
IS
FINAi
AHOUNT
,
ASK
7
")

Would
it
be
worth
something
to
you
(
and
your
family)
to
raise
the
water
quality
level
from
level
"
D"
to
a
higher
level?
NUKBBR
.
CIRCLE
YES
.
.
.
.
.
.
.
.
.
01
(
GO
TO
B­
3.
b.)

No
.
.
.
.
.
.
.
.
.
02
(
GO
TO
B­
3.
e.)
E:
34­.
S6)

I
D­
25
­
 
.
..
 
 
 
.
.
­.
"
.
.
_
.­
 
 
.
 
.
..
 
 
 
 
 
 
­
.
 
.
..
 
.
..
­
 
 
 
.
­
­.
 
b.
Would
you
(
and
your
family)
be
willing
to
pay
(
an
additional)
$
125
each
year
in
higher
taxes
and
prices
for
ptoducts
that
companies
sell
to
raise
the
water
quality
from
level
"
D"
to
level
"
C",
where
game
fish
can
live
in
it
and
it
is
improved
fyr
other
activities?
CIRCLE
NIJHBER.

­"
s"

IF
YES,
INCREASE
THE
$
10
INCREMENTS
UNTIL
.
.
.
.
.
.
.
.
01
.
.
.
.
.
.
.
.
02
1
DOLLAR
MOUNT
IN
IF
HO,
DECREASE
THE
DOLLAR
AMOUNT
IN
A
"
NO"
ANSWER
IS
S1O
INCREMENTS
UNTIL
A
"
YES"
ANSWER
IS
GIVEN.
E.
G.,
"
Would
you
be
willing
GIVEN.
E.
G.,
"
Would
you
be
willing
to
to
pay
$
135
(
more)
each
year
to
raise
pay
$
115
(
more)
l
 
ach
year
to
raise
the
the
water
quality
from
level
`
D'
to
water
quality
from
level
`
D'
to
level
level
!
c,?,~
ETC.
WHEN
A
"
NO"
ANSWER
`
C'?"
ETC.
~
A
!!~'$~
ANSWER
IS
IS
GIVSN,
RECORD
DOLLAR
AMOUNT
OF
GIVEN,
RECORD
DOLLAR
ANOUNT.
LAST
"
T'S"
ANSWER.

(
IF
s
00LURS
IF
Go
ANY
AMOUNT,
GO
TO
B­
3.
c.;

)
ZERO
DOLLARS
IS
FINAL
MOUNT,
TO
B­
3.
d.

c.
Would
you
(
and
your
family)
be
willing
to
pay
an
additional
$
125
each
year
in
higher
taxea
and
prices
for
products
that
companies
sell
to
raise
the
water
quality
from
level
"
C"
to
level
"
B",
where
you
can
swim
in
it
and
it
is
improved
for
other
activities?
CIRCLE
NuNEER.

rms"""""""""
o'
I
NO
.
.
.
.
.
.
.
..
O2
1
+

IF
YES,
INCREASE
TIE
DOLLAR
AMOUNT
IN
IF
NO,
DECRSASE
`
HIS
DOLLAR
AMOUNT
IN
$
10
INCREMENTS
UNTIL
A
"
NO"
ANSWER
IS
$
10
INCREMENTS
UNTIL
A
"
YES"
NWER
IS
GIVEN.
E.
G.
,
"
WOuld
you
be
willing
GIVEN.
E.
G.
,
"
Would
you
be
willing
to
to
pay
$
135
more
each
year
to
raise
pay
$
115
q
 
ore
each
year
to
raise
the
the
water
quality
from
level
`
C'
to
water
quality
from
level
`
C`
to
level
level
`
B'
?"
ETC.
WHEN
A
"
XO'*
ANSWER
`
B'?"
ETC.
~
A
"~
s"
ANSWER
IS
RECORD
DOLL4R
AMOUNT
OF
GIVEN
,
RECORD
DOLLAR
AMOUNT.
ANSWER
.

$
DOLLARS
(
GO
TO
B­
4)

.

D­
26
(
37­
39)

(
40­
42)
d.
Would
you
(
and
your
family)
be
willing
to
pay
$
12!
5
each
year
in
higher
taxes
and
prices
for
products
that
companies
sell
to
raise
the
water
quality
from
level
"
D"
to
Level
"
B",
where
you
can
swim
in
it
and
it
is
improved
for
other
activities?
CIRCLE
NUMBER
.

r
YEs
NO
.

+

IF
YES,
INCREASE
TNE
S
10
INCREMENTS
UNTIL
.
.
.
.
.
.
.
.
01
.
.
.
.
.
.
.
.
02
1
00LMR
AMOUNT
IN
IF
NO,
DECRBASE
THE
DOLMR
MOUNT
IN
A
"
NO"
ANSWER
IS
$
10
INCREMENTS
UNTIL
A
"
YES"
ANSWBR
IS
GIVEN.
E.
G.
,
"
Would
you
be
willing
GIVEN.
E.
G.,
"
Would
you
be
willin~
to
to
pay
$
135
l
 
ach
year
to
caise
th;
pay
$
115
ea&
year
to
raise
the
w~
ter
water
quality
from
level
`
D'
to
level
qUalitY
frOR
level
`
D'
to
level
`
B
e?'~
`
B'?"
`
ETC.
WNEN
A
"
NO"
ANSWBR
IS
ETC.
`~
A
~
f~
st
ANSWER
IS
GIVXX,
GIVEN
,
RECOSD
DOLLAR
AMOUNT
OF
LAST
RSCORD
COLLAR
AHOUNT
.
"
YES"
ANSWSR.

(
IF
ANY
AMOUNT
IN
a.,
GO
TO
B­
4;

$
DOLLARS
IF
ZERO
DOfJARS
IN
a.
AND;
"
ANY
AMOUNT
IN
d.,
GO
TO
B­
4.
d.;
.
ZERO
DOLLARS
I.
N
d.,
GO
TO
B­
3.
e
e.
We
have
found
in
studies
of
this
type
that
people
have
a
lot
o
different
reasons
for
answering
aa
they
do.
Some
people
felt
the
did
not
have
enough
information
to
give
a
dollar
amount,
some
di
not
want
to
put
dollar
values
on
environmental
quality,
and
som
objected
to
the
way
the
question
was
presented.
Others
gave
a
zer
dollar
amount
because
that
was
what
it
was
worth
to
them.

Which
of
these
reasons
best
describes
why
you
answered
the
way
yo
did?
REPEAT
SIMSONS
IF
NECESSARY
AND
CIRCLE
NUlfSER.

N~
ENOUGE
INFORMATION
.
.
01
DID
NOT
WANT
TO
PLACE
DOLLAR
VALUS.
.
.
.
..
O2
OBJECTED
TO
WAY
QUESTION
\
(
GO
TO
B­
6)
UASPRESENTBD
.
.
.
.
..
O3
TIMT
IS
WNAT
IT
IS
WORTN
.
04
OTHER
(
SPECIFY)
­
.
.
.
­
05
J
(
43­
4s)

(
46­
471
D­
27
PART
2
SUGGESTIONS
FOR
IMPROVING
THE
QUESTIONNAIRE
FOR
FUTURE
USE
Any
survey
questionnaire
can
be
improved
based
on
the
additional
information
learned
in
the
execution
of
the
survey.
This
questionnaire
is
not
an
exception.
One
of
the
most
significant
changes
would
amend
the
word
"
additional
to
the
introduction
of
Question
B­
7
to
clarify
that
the
bid
amount
is
in
addition
to
the
amounts
previously
bid.
It
is
also
unclear
whether
the
supply
uncertainty
dimension
added
in
this
question
is
effectively
expressed.
This
could
be
improved
with
a
couple
of
clarifying
sentences.

The
introduction
to
Question
B­
5
could
be
improved
by
better
explaining
how
water
quality
might
be
worsened
only
for
some
weekends.
For
example,
a
sentence
describing
"
the
effect
of
higher
water
temperatures
in
the
summer
months
could
reduce
water
quality
only
in
that
part
of
the
year"
might
clarify
the
supply
uncertainty
that
is
intended
in
this
question.

The
explanation
and
introduction
to
the
contingent
ranking
format
is
too
brief.
While
this
may
be
minimized
by
the
respondent's
familiarity
with
water
quality
from
the
other
contingent
valuation
questions,
it
would
require
expansion
for
an
application
as
an
independent
format.
This
introduction
could
also
explain
in
more
detail
the
relation
between
water
quality
levels
and
the
amounts
paid.

There
is
a
slight
difference
in
wording
between
Versions
A
and
B
and
C
and
D
as
a
result
of
a
word
processing
error.
The
phrase
"
where
it
is
not
even
clean
enough
for
boating"
was
inadvertently
omitted
from
Question
B­
3a
in
Versions
C
and
D.
The
water
quality
ladder
would
have
shown
that
E
was
not
suitable
for
boating,
so
the
potential
bias
here
is
likely
small
but,
nonetheless
could
be
avoided
in
future
use.

Finally,
some
changes
might
be
useful
in
the
visual
aids.
The
cards
in
the
contingent
ranking
should
be
same
size
as
the
other
aids
to
make
them
easier
to
handle.
For
consistency
with
the
other
aids,
the
value
card
could
have
been
done
in
bolder
print
to
make
it
stand
out.
There
is
some
debate
that
a
visual
aid
describing
the
payment
vehicle
might
have
made
it
clearer
to
people
how
they
currently
pay
for
water
quality.
On
the
other
side
of
this
argument
is
the
thinking
that
this
may
actually
increase
the
respondents'
confusion

In
summary,
the
questionnaire
performed
well
for
most
of
the
key
questions
but
some
relatively
minor
changes
might
have
made
it
even
better.
The
question
responses
most
affected
by
the
change
are
the
existence
value
responses
in
Question
B­
7.

.
D­
28
.
APPENDIX
E
TECHNICAL
WATER
QUALITY
MEASURES:
AN
ECONOMIST'S
PERSPECTIVE
E.
1
INTRODUCTION
A
discussion
of
water
quality
measurement
should
define
the
term
water
guality,
including
descriptions
of
the
various
attributes
that
determine
quality.
Although
seldom
together,
several
disciplines
have
repeatedly
explored
this
issue,
and
a
significant
amount
of
literature
is
relevant
to
the
questions
that
arise
in
benefit
estimation.
This
appendix
discusses
several
of
these
questions

E.
2
AN
OVERVIEW
OF
TECHNICAL
WATER
QUALITY
MEASURES"

E.
2.
1
Introduction
The
following
sections
briefly
describe
technical
measures
of
water
quali
t
y
.
Sections
E.
2.2
and
E.
2.3
discuss
freshwater
systems,
focusing
on
their
characteristics
and
their
ability
to
assimilate
effluents.
Section
E.
2.4
discusses
commonly
used
parameters,
noting
their
importance
in
an
ecosystem,
their
measurement,
and
the
ability
of
individuals
to
perceive
their
changes.

E,
2.2
Water
Quality
in
Freshwater
Systems
Freshwater
areas
are
intricate
systems
differing
in
attributes
and
causal
relationships.
Freshwater
system
descriptions
are
complicated
by
climate,
geography,
land
use,
water
management,
and
existing
plants
and
animals.
Because
these
particular
characteristics
are
usually
unknown,
actual
physical
relationships
cannot
be
determined.
Descriptions
are
further
complicated
when
scientific
analysis
cannot
measure
deleterious
long­
term
or
synergistic
effects
in
a
natural
setting.

Freshwater
systems
are
either
Ientic
systems,
which
contain
standing
water
such
as
lakes,
or
lotic
systems,
which
contain
running
water
such
as
streams
and
rivers.
However,
classifying
a
system
as
Ientic
or
Iotic
can
be
difficult
when
natural
impoundments,
dams,
and
reservoirs
occur
in
either.
In
addition,
while
the
basic
nutrient
cycles
are
the
same
for
both
systems,
life
cycles
and
pollution
effects
differ
considerably.

The
scope
of
this
project
limits
discussion
only
to
Iotic
systems.
impoundments
are
considered
due
to
their
general
dynamic
nature.
However,
because
the
unique
Ientic
system
characteristics
sometimes
appear
in
natural
and
manmade
impoundments,
problems
common
to
both
system
types
are
also
discussed.

E­
1
E.
2.3
Assimilative
Capacity
The
ability
of
a
Iotic
system
to
Iution
levels.
Assimilative
capacity
sence
of
deleterious
effects
with
a
assimilate
is
usually
effluents
determines
actual
pol
­
defined
with
respect
to
the
abgiven
level
of
discharge
into
a
receiving
water.
However,
any
materials
discharged
into
the
water
have
an
effect.
The
major
problem
is
one
of
identifying
and
measuring
these
changes
and
of
determining
when
they
become
deleterious.
An
effluent's
effect
on
the
environment
is
influenced
by
time
period,
amount
of
available
oxygen,
plant
nutrients,
and
locational
characteristics.

Daily
and
seasonal
variation
in
the
speed
of
nutrient
cycling
are
major
determinants
of
an
effluent's
effect
on
water
quality.
Lotic
systems
derive
most
of
their
nutrients
from
soil
runoff,
causing
primary
productivity
to
vary
seasonally.
As
land
nutrient
and
groundwater
levels
vary,
so
does
the
Iotic
environment's
assimilative
capacity.
Available
sunlight
is
the
primary
source
of
daily
variation,
with
the
peak
rate
of
photosynthesis
in
the
afternoon
hours
causing
peak
levels
of
dissolved
oxygen.

Assimilative
capacity
is
commonly
measured
by
the
availability
of
dissolved
oxygen.
Because
ail
aquatic
animal
life
depends
on
dissolved
oxygen,
low
dissolved
oxygen
levels
may
cause
a
reduction
in
species
diversity
and
number.
Some
effluents
reduce
dissolved
oxygen
because
they
change
the
rate
of
photosynthesis
the
volubility
of
oxygen,
and
the
diffusion
of
atmospheric
oxygen
or
they
increase
aerobic
bacteria
activity.

Existing
plant
nutrients
also
determine
the
effect
of
effluents.
Each
ecosystem
has
a
defined
nitrogen­
phosphorus
ratio,
and
all
organisms
within
the
system
can
use
nutrients
only
in
this
ratio.
When
an
effluent
increases
nutrient
levels,
a
natural
growth
limit
is
eliminated,
resulting
in
excessive
plant
growth,
which
eventually
decomposes
and
decreases
dissolved
oxygen.

Long­
term
changes
in
assimilative
capacity
occur
due
to
an
aging
process.
As
erosion
takes
place,
headwaters
tend
to
migrate
upstream,
as
will
plant
and
animal
communities.
Erosion
is
also
responsible
for
increases
in
suspended
solids,
which
deteriorate
and
affect
the
composition
of
the
river
bottom
over
time.

E.
2.4
Water
Quality
Parameters
The
capacity
of
a
water
system
to
accommodate
uses
may
be
defined
by
a
series
of
hydrological,
physical,
chemical,
and
biological
parameters.
Theso
parameters
are
relevant
in
explaining
the
effects
of
an
effluent
on
the
equilibrium
and
existing
conditions.
Both
relative
and
absolute
measurements
aro
important
in
evaluating
parameters.
No
single
parameter
can
be
used
as
an
adequate
measure
of
water
quality,
yet
in
many
cases
focusing
on
one
parameter
is
dictated
by
data
limitations.
Several
types
of
parameters
descri
b.
water
quality,
and
a
brief
discussion
of
each
follows.

E­
2
Hydrological
Parameters
Hydrological
parameters
determine
the
level
of
physical,
chemical,
and
biological
parameters.
These
parameters
characterize
the
atmosphere
and
catchment
area,
and
care
is
required
in
placing
the
analysis
in
a
particular
hydrological
process.
Consideration
should
therefore
be
given
to
climate,
properties
of
air,
precipitation,
erosion,
and
vegetation.

Most
studies
that
attempt
to
measure
water
quality
do
not
explicitly
consider
hydrological
parameters.
Care
is
taken
only
to
place
measurements
in
a
particular
season.
For
example,
flow
is
often
described
as
important
but
not
considered
directly.
This
treatment
can
be
explained
by
a
lack
of
data
on
how
often
hydrological
parameter
changes
occur
and
their
synergistic
effect
on
the
level
of
other
parameters.
A
possible
methodology
to
include
these
parameters
would
be
to
use
water
quality
modeling.
This
technique,
however,
requires
large
amounts
of
information
and
time.

Physical
Parameters
Physical
parameters
are
commonly
used
water
quality
measures.
However,
their
values
vary
significantly
due
to
seasonal
and
diurnal
patterns
and
sitespecific
characteristics.
Readings
may
not
be
applicable
to
wide
areas
due
to
these
variations.
These
parameters
include
the
following:

T
u
r
b
i
d
i
t
y
is
caused
by
the
presence
of
suspended
solids.
These
solids
are
usually
a
variety
of
substances
influenced
by
man­
made
and
natural
occurrences.
Increases
in
suspended
solids
will
affect
the
level
of
photosynthesis
as
transparency
is
decreased.
Also,
as
settling
occurs,
eggs
and
larva
may
be
suffocated,
affecting
fish
reproduction
and
species
diversity.
Water
turbidity
is
usually
measured
by
a
Secchi
disk.
This
disk
is
lowered
into
the
water
until
it
disappears,
and
the
resulting
depth
is
recorded.
Alternatively
the
Jackson
Turbidity
Unit
can
be
used.
Regardless
of
the
measurement
technique,
i
n
d
i
v
i
d
u
a
l
p
e
r
c
e
p
t
i
o
n
s
o
f
t
u
r
b
i
d
i
t
y
a
r
e
thought
to
be
generally
correlated
with
measured
levels,
explaining
its
common
use
in
water
quality
studies.
Unfortunately,
little
is
known
of
how
sensitive
individuals
are
to
small
turbidity
changes
and
what
importance
this
has
in
their
decisionmaking.

Color
is
important
in
determining
both
transparency
and
aesthetics
of
water.
Water
may
contain
a
variety
of
compounds
that
change
the
amount
of
sunlight
allowed
in
a
water
column,
resulting
in
a
change
in
the
photosynthesis
rate.
Color
is
usually
determined
by
visual
comparison
to
a
group
of
standard
colors.
The
use
of
this
parameter
in
water
quality
studies
is
rare
due
to
the
lack
of
consistent
measurement
over
time
and
among
sites.
The
link
between
color
and
individual
perceptions
is
also
not
well
known.

Temperature
is
a
major
determinant
of
the
level
of
biological
and
chemical
activity
because
temperature
changes
also
cause
a
change
in
the
equilibrium
of
a
water
system.
Lotic
systems
are
greatly
affected
by
E­
3
.!
atmospheric
temperature
and
usually
do
not
contain
any
thermal
stratification.
For
these
reasons
organisms
are
usually
tolerant
of
large
temperature
changes.
When
impoundments
occur
in
the
Iotic
environment,
temperature
stratifications
do
occur,
inhibiting
the
availability
of
dissolved
oxygen
at
certain
levels.
Temperature
readings
are
taken
at
various
depths
with
a
reversing
thermometer
or
bathythermograph.
Simple
temperature
readings
are
not
a
good
indicator
of
water
quality.
A
more
appropriate
measure
would
be
deviation
from
the
norm
caused
by
man­
made
and
natural
infractions.
A
change
in
temperature
is
usually
perceived
through
indirect
changes
such
as
algae
growth,
changes
in
fish
population,
and
physiological
disturbances
in
swimmers.

Odor
and
taste
measure
the
presence
of
industrial
discharges,
microscopic
organisms,
and
vegetation.
These
factors
are
usually
the
result
of
industrial
discharge
or
aquatic
decomposition.
The
measurement
of
odor
is
determined
by
concentration
levels
of
various
compounds
in
a
sample.
Effects
of
odor
are
difficult
to
measure
because
perceptions
vary
depending
on
the
individual
and
distance
to
the
water.

Chemical
Parameters
Chemical
parameters
characterize
natural
and
man­
made
components
of
a
particular
water
sample.
Reported
results
are
often
misleading
because
the
parameters
may
not
be
measured
from
a
desired
area.
The
choice
of
parameters
and
sample
sites
usually
is
based
on
pollutants
expected
due
to
regional
and
man­
made
characteristics.
Also,
cause
and
effect
relationships
are
not
precisely
known
in
the
scientific
community
nor
are
changes
well
perceived
by
individuals.
Thus,
we
cannot
determine
exact
relationships
between
parameters
and
water
quality.
Usually
only
the
direction
of
change
in
water
quality
is
known.
Common
chemical
parameters
are
as
follows:

Dissolved
oxy
qen
measures
the
intensity
of
organic
decomposition
and
the
ability
of
self
purification.
Dissolved
oxygen
is
necessary
for
respiration
of
plants
and
animals
and
aerobic
decomposition.
Concentrations
of
dissolved
oxygen
are
increased
with
photosynthesis
and
atmospheric
reaeration.
Decreases
are
caused
by
vitrification,
biological
oxygen
demand,
and
benthal
oxygen
demand.
Many
species
are
not
tolerant
of
low
levels
of
dissolved
oxygen,
and
offensive
odor
may
also
occur
as
decomposition
occurs
without
the
presence
of
oxygen.
Dissolved
oxygen
is
expressed
in
terms
of
mg/
liter
or
percent
saturation.
Extensive
work
has
been
completed
on
fish
populations
and
levels
of
dissolved
oxygen.
These
controlled
experiments
relate
fish
reproduction
rates
to
minimum
dissolved
oxygen
requirements
for
various
species.

Total
dissolved
solids
represent
the
concentration
of
nondegradable
wastes
in
a
water
sample.
These
solids
may
be
toxic
to
the
surrounding
food
chain,
but
little
is
known
about
this
relationship.
Concentrations
are
usually
in
terms
of
mg/
liter.

E­
4
~
is
an
index
of
the
acidic­
basic
relationship
of
various
mineral
and
basic
substances.
Under
natural
conditions,
pH
ranges
from
5.0
to
8.6
on
a
scale
of
1
to
14.
Heavily
polluted
water
may
cause
a
low
pH
(
i.
e.
,
an
increased
concentration
of
acid).
Existing
plant
and
animal
life
may
not
be
tolerant
of
severe
pH
changes.
A
pH
change
generally
results
in
a
smaller
variety
of
organisms.
Recreation
use
of
water
usually
requires
a
pH
in
the
range
occurring
in
natural
conditions.
However,
swimming
may
require
a
narrow
range
of
6.5
t
o
8
.
3
.
Individual
perceptions
of
pH
are
sensitive
only
to
large
changes,
though
a
change
may
be
perceived
through
eye
irritation
or
touch.

Nitrates
are
formed
by
the
biochemical
oxidation
of
ammonia.
Some
stratification
occurs
naturally,
resulting
in
surface
waters
having
higher
concentrations.
Increased
concentration
may
indicate
fecal
pollution
in
the
preceding
period.
The
concentration
of
nitrates
may
also
indicate
the
rate
of
self
purification
of
a
water
system.
Nitrates
are
usually
reported
as
mg/
liter.

Metals
present
in
a
Iotic
environment
can
be
caused
by
soil
drainage.
Therefore,
seasonal
changes
will
affect
the
concentration
of
metals
present.
Industrial
sources
of
metals
include
mine
pit
discharge,
ore
enriching
factories,
and
iron
and
steel
factories.
The
effects
of
several
metals
such
as
copper,
lead,
and
mercury
are
commonly
studied
and
well
known.
The
effects
of
other
metals
such
as
chromium
cadmium,
cobalt,
and
nickel
are
not
as
well
known.
Concentrations
are
usually
reported
as
mg/
liter.
Severe
concentrations
may
inhibit
development
if
they
are
passed
to
higher
members
of
the
food
chain.

Surface
active
agents
represent
a
variety
of
man­
made
compounds.
These
agents
or
surfactants
are
usually
found
in
detergents.
Concentrations
result
in
the
normal
breakdown
of
organic
material.
More
noticeable
effects
are
a
bitter
taste,
a
soapy
and
kerosene
odor,
and
the
presence
of
foam.
Concentrations
usually
are
measured
in
terms
of
mg/
liter.

Pesticides
are
any
substance
designed
to
destroy
plant
or
animal
organisms
These
compounds
enter
the
water
indirectly
from
runoff
and
drainage
or
by
direct
application.
Agriculture
is
the
dominant
source
of
pesticide
contamination.
Many
pesticides
have
a
cumulative
effect,
causing
increased
concentrations
at
higher
levels
of
the
food
chain.
As
concentrations
increase,
the
natural
development
of
organisms
will
be
altered.
Pesticides
include
a
wide
variety
of
compounds
and
are
usually
described
in
mg/
liter.
Even
though
their
diversity
usually
precludes
their
use
as
a
measure
of
water
quality,
pesticides
are
considered
an
important
indicator
of
water
quality.

E­
5
Biological
Parameters
Biological
parameters
reveal
the
quality,
size,
and
type
of
animal
and
plant
populations
within
a
water
system.
Data
readings
vary
significantly
with
the
season
and
flow
velocity,
but
these
parameters
may
give
a
reliable
picture
of
the
average
situation
since
organisms
cannot
rapidly
adapt
to
change.
Individuals
do
not
directly
perceive
changes
in
these
parameters
but
notice
them
through
such
effects
as
odor,
algae,
and
resulting
illness.
These
factors
are
most
important
to
direct
contact
uses
but
also
apply
to
secondary
recreation
The
two
important
biological
parameters
are
as
follows:

Biological
oxyg
en
demand
measures
the
rate
of
oxygen
consumption
in
a
system
due
to
organic
decomposition.
High
levels
of
organic
waste
cause
an
increase
in
the
biological
oxygen
demand
and
a
resulting
decrease
in
available
dissolved
oxygen.
T
h
e
s
e
r
a
t
e
s
will
differ
depending
on
the
state
of
the
matter
being
decomposed.
Since
temperature
controls
the
rate
of
organic
activity,
it
also
greatly
influences
oxygen
demand.
Biological
oxygen
demand
is
generally
measured
as
the
amount
of
oxygen
removed
from
a
sample
in
a
5­
day
period
and
is
an
important
part
of
most
water
quality
determinations.
However,
sample
readings
may
not
be
comparable
due
to
changes
in
assimilative
capacity.
For
example,
a
reading
may
have
a
large
value
and
yet
have
little
effect
on
water
quality
due
to
characteristics
such
as
large
available
dissolved
oxygen
and
strong
flow.

Microbiological
parameters
determine
the
presence
of
waterborne
disease.
The
parameters
would
include
bacteria,
viruses,
and
algae.
Both
bacteria
and
viruses
may
be
excreted
in
the
feces
of
infected
animals
The
most
common
parameter
of
fecal
contamination
is
the
test
f
o
r
coliform
bacteria
expressed
as
number
of
bacteria
per
liter.
Limits
are
currently
set
on
fecal
coliform
depending
on
the
use
of
the
river.
The
presence
of
bacteria
and
viruses
does
not
affect
the
appearance
of
the
water.
Except
at
high
levels,
algae
is
not
toxic
but
may
indicate
overfertilization
of
the
system
by
man
or
other
mammals.
Algae
may
be
considered
a
pollutant
since
it
is
readily
noticed
in
water.

E.
3
ISSUES
IN
DETERMINING
WATER
QUALITY
E.
3.1
Introduction
Several
issues
arise
in
attempts
to
define
water
quality,
the
most
important
of
which
involve
the
uses
of
a
water
system
as
they
affect
quality
and
the
selection
of
an
appropriate
site.
A
discussion
of
these
two
issues
follows,
including
a
brief
description
of
how
they
relate
to
this
study.

E.
3.2
Water
Quality
and
Use
Water
quality
is
directly
dependent
on
current
and
future
uses
of
Common
use
categories
are
drinking,
swimming,
fishing,
boating,
and
t
r
i
a
l
.
This
list
is
an
obvious
simplification
as
it
does
not
recognize
the
E­
6
a
sit..
indus
­
attrib
­
t
I
I
1
I
I
utes
desirable
for
each
use.
The
use
of
water
for
drinking,
for
example,
may
occur
within
a
wide
range
of
attributes
given
various
levels
of
water
treatment
The
inability
to
define
these
attribute
ranges
causes
oversimplification
when
water
quality
is
measured
over
various
uses.

Uses
of
a
water
system
are
related
to
each
other
in
a
spatial
and
temporal
sense.
As
the
level
of
one
use
changes,
the
benefits
derived
from
competing
uses
will
also
change.
This
relationship
is
not
well
defined
because
it
depends
on
several
variables,
including
the
particular
uses
considered,
characteristics
of
the
area,
and
the
time
frame
considered.
In
some
instances,
the
relationship
may
depend
on
differential
preferences
of
the
potential
users
(
e.
g.
,
teenagers
and
young
families
may
desire
a
crowded
beach
while
honeymooners
and
older
people
may
prefer
an
uncrowded
beach),
and,
in
extreme
cases,
uses
may
be
completely
independent
or
mutually
exclusive.

To
ensure
the
same
uses
at
each
site
in
the
travel
cost
approach,
this
study
used
only
U
.
S.
Army
Corps
of
Engineer
areas.
Using
only
these
areas
eliminates
part
of
the
problem
of
defining
uses,
but
it
does
not
account
for
competing
uses.
Ideally,
more
consideration
should
be
given
to
variation
in
uses
between
sites
and
their
relationship
to
each
other.

E.
3.3
Water
Quality
Within
an
Area
Water
quality
is
related
to
the
physical
boundaries
of
the
study
area
in
two
ways:
boundaries
determine
both
the
physical
attributes
and
the
scientific
parameters
to
consider.
In
turn,
physical
attributes
determine
the
u
s
e
s
allowed
and
the
interrelationship
between
uses.
For
example,
the
presence
of
a
dam
increases
the
damage
caused
by
an
industrial
effluent
on
fish
populations

The
determination
of
the
appropriate
scientific
parameters
is
subject
to
the
continuous
nature
of
water
quality.
As
these
measurements
vary
between
measuring
sites,
the
problem
becomes
more
complex.
Quality
of
water
to
a
user
is
determined
by
the
immediate
and
surrounding
area.
How
to
incorporate
these
readings
is
not
clear.
Consideration
should
be
given
to
uses
involved
as
well
as
the
physical
relationship
between
areas.
This
issue
is
clouded
by
incomplete
data
when
water
quality
is
actually
measured.

Data
availability
ultimately
constrains
the
determination
of
the
study
area.
The
locations
of
existing
monitoring
sites
are
based
on
a
variety
of
concerns
such
as
location
of
fisheries
,
effluents
present,
and
convenience.
Quite
often
the
measurements
obtained
do
not
conform
to
the
desirable
study
requirements.
Hence,
the
use
of
these
data
may
bias
results
depending
on
site
proximity
to
the
study
area
and
the
use
being
considered.

E.
4
MEASUREMENT
OF
WATER
QUALITY
E.
4.1
1
ntroduction
A
useful
measure
of
water
quality
would
be
a
universal
number
or
index
that
can
compare
uses
and
scientific
parameters.
Both
individual
perceptions
E­
7
ot
parameters
and
scientlTlc
measures
OT
parameters
coula
be
usecl
Inalvlaually
or
to
compare
to
an
index.
However,
assigning
the
appropriate
weights
to
each
measure
is
a
difficult
task.
A
brief
discussion
of
advantages
to
various
methods
to
describe
water
quality
follows.

E.
4.2
Human
Perceptions
and
Water
Quality
Measurement
Individual
perceptions
play
an
important
role
in
water
quality
determination
but
consistent
measurement
of
perceptions
is
a
major
problem.
Studies
have
shown
that
perceptions
usually
vary
with
questionnaire
design
information
provided
and
sample
population.
Binkley
and
Hanemann
[
1978]
found
that
respondents
base
evaluations
of
water
quality
on
incorrect
information.
Ditton
and
Goodale
[
1973]
found
that
respondents
tended
to
describe
areas
closest
to
their
residence,
which
causes
large
variations
in
the
water
quality
rating
over
the
entire
study
area.
Moreover,
changes
in
other
site
attributes
limit
the
ability
to
draw
general
conclusions
as
to
the
effects
of
changes
in
water
quality
alone.
On
the
other
hand,
Bouwes
and
Schneider
(
1979)
found
reasonably
good
correlation
between
perceptions
and
the
scientifically
based
lake
condition
index.

Some
differences
in
perceptions
have
been
attributed
to
characteristics
of
the
respondents.
Barker
[
1971
]
found
that
users
of
an
area
tend
to
rate
water
quality
more
favorably
than
nonusers.
Ditton
and
Goodale
[
1973]
determined
that
swimmers'
perceptions
of
water
quality
differed
from
fishermen's,
both
in
terms
of
their
ratings
of
water
quality
and
the
relative
importance
of
individual
features.

E.
4.3
Technical
Water
Quality
Measurement
Scientifically
measured
parameters
are
usually
good
indicators
of
water
quality
changes.
Unlike
individual
perceptions,
the
technical
water
quality
tests
are
usually
comparable
over
time
and
between
sites.
Determination
of
important
parameters
is
difficult,
however,
since
most
scientific
information
is
obtained
only
through
controlled
experiments.
Changes
in
water
quality
caused
by
parameters
are
difficult
to
determine
because
particular
site
characteristics
must
be
known
to
determine
an
expected
change,
even
in
the
short
run.
I
n
addition,
long­
term
and
synergistic
effects
also
usually
cannot
be
determined
because
of
poor
information.

E.
4.4
Water
Quality
Indexes
An
ideal
water
quality
measure
would
combine
scientifically
measured
parameters,
individual
perceptions,
and
alternative
uses
of
an
area.
Unfortunately
these
measures
require
considerable
information,
and
their
components
may
vary
between
sites.
In
lieu
of
complete
information,
many
studios
have
used
approaches
that
rely
on
individual
parameters
or
indexes
to
dotormine
water
quality.
While
most
studies
have
used
one
or
more
individual
parameters
without
determining
their
relative
importance,
other
studies
havo
used
the
index
approach
to
solve
several
of
the
problems
noted
above.
Thus,
while
far
from
perfect,
the
index
approach
does
represent
a
tractable
method
of
relating
water
quality
to
users,
perceptions,
and
scientific
judgment.

E­
8
E
.4.4.1
The
National
Sanitation
Foundation
Index
The
ideal
measure
of
water
quality
would
incorporate
scientific
parameters
public
perception
of
the
water,
and
potential
uses
of
the
water.
As
an
attempt
to
incorporate
these
considerations,
the
National
Sanitation
Foundation
(
NSF)
index
is
a
constructive
approach
to
several
problems
in
water
quality
measurement.
A
composite
of
nine
parameters,
the
NSF
index
was
developed
through
several
questionnaires
given
to
individuals
with
water
quality
experience
Respondents
first
selected
parameters
they
felt
were
important
to
water
quality.
Followup
contacts
were
then
made
to
give
the
previous
group
responses
to
the
respondents
and
to
allow
them
to
change
their
initial
responses.
A
rating
of
these
parameters
in
terms
of
water
quality
and
synergistic
effects
was
then
developed
based
on
these
responses.
The
final
parameters
chosen
included
dissolved
oxygen,
fecal
coliform
density,
pH,
5­
day
biological
oxygen
demand,
nitrates,
phosphates,
temperature,
turbidity,
and
total
solids.

The
next
step
in
developing
the
NSF
index
required
the
development
of
water
quality
curves
for
each
parameter.
These
curves
represent
the
expected
result
of
parameter
concentrations
on
water
quality
and
must
be
combined
with
the
relative
weights
derived
from
the
respondents'
rankings
of
each
parameter.
These
quality
curves
and
weights
constitute
the
final
components
of
the
index.
More
details
on
this
index
can
be
found
in
EPA
[
1982].

Researchers
have
applied
the
NSF
index
in
a
number
of
studies.
The
U
s
.
Environmental
Protection
Agency
(
EPA)
applied
the
NSF
index
to
the
Kansas
River
basin
to
determine
its
effectiveness,
including
an
appraisal
of
sampling
and
computing
difficulties.
The
Kansas
River
basin,
a
wide,
shallow
river
of
moderate
velocity,
has
light
industry
and
receives
treated
municipal
wastes
from
over
40
cities
and
towns.
EPA
calculated
two
forms
of
the
NSF
index
with
almost
600
water
samples
from
over
26
sites.
Calculated
index
values
were
consistent
with
researchers'
attitudes
toward
the
various
reaches
of
the
river.

The
index
calculations
were
also
used
to
examine
several
other
concerns.
For
example,
the
correlations
between
several
variables
were
measured
to
test
the
validity
of
substituting
parameters
when
certain
data
do
not
exist.
The
study
determined
that
suspended
solids
can
be
substituted
for
turbidity
and
total
coliform
for
fecal
coliform.

The
NSF
index
provides
a
scientifically
based
method
of
linking
changes
in
water
quality
to
the
effects
of
those
changes.
The
index,
however,
does
not
provide
a
linkage
to
individual
perceptions
of
water
quality
changes
and
cannot
differentiate
threshold
values
for
specific
uses
like
fishing
or
swimming.

E.
4.4.2
Resources
for
the
Future
Water
Quality
Ladder
A
significant
problem
with
the
NSF
index
is
that
it
does
not
take
into
account
potential
uses
for
a
particular
body
of
water.
At
Resources
for
the
Future
(
RF
F),
Vaughan
in
Mitchell
and
Carson
[
1981]
used
a
variation
of
the
NSF
index
to
determine
minimum
levels
of
water
quality
for
various
uses.
Specifically,
Vaughan's
index
used
five
NSF
index
parameters
chosen
on
the
E­
9
Table
E­
1.
Water
Quality
Classes
by
Parameter
and
Index
Values
Measurable
water
quality
characteristics
Fecal
Disso!
ved
5­
Day
Water
quality
coliform
oxygen
a
BOD
Turbidity
Ladder
use
designation
(#/
100
mL)
(
mg/
L)
(
mg/
L)
(
JTu)
pH
value
Acceptable
for
o
7.0
(
90)
o
5
7.25
9
.
5
drinking
without
treatment
Acceptable
for
200
6.5
(
83)
1
.
5
10
7.25
7
.
0
swimming
Acceptable
for
1,000
5.0
(
64)
3
.
0
50
7.25
5.1
game
fishing
Acceptable
for
1,000
4.0
(
51)
3
.
0
50
7.25
4
.
5
rough
fishing
Acceptable
for
2,000
3.5
(
45)
4
.
0
100
4.25
2
.
5
boating
a
Numbers
in
parentheses
are
percent
saturation
at
85
°
F.

basis
of
judgment
and
data
availability:
fecal
coliform,
dissolved
oxygen,
biological
oxygen
demand,
turbidity,
and
pH.
As
shown
in
Table
E­
1,
Vaughan
associated
specific
parameter
levels
with
five
use
designations.
He
then
used
a
truncated
version
of
the
NSF
index
to
place
each
minimum
use
designation
on
an
index
value
range
from
O
to
10
with
the
final
index
values
for
each
use
classification
shown
in
Table
E­
1
.

The
RFF
index
provides
a
valuable
link
between
various
parameters
and
use
designations.
Even
though
the
parameter
choice
may
be
somewhat
arbitrary
the
parameters
neatly
map
into
desirable
attributes
for
a
particular
use.
However,
the
RFF
index
does
not
account
for
differing
individual
perceptions
that
may
be
easily
incorporated
with
further
research.
The
RFF
ladder
is
used
in
this
study,
as
shown
in
Figure
4­
5.

E.
5
SUMMARY
The
questions
involved
in
defining
water
quality
are
complex,
and
there
are
no
clear
answers.
Water
quality
studies
must
jointly
determine
the
parameters
to
be
considered,
the
uses
to
be
considered,
and
the
definition
of
tho
site
to
be
studied.
I
n
addition,
each
of
these
issues
has
many
aspects,
such
as
how
to
define
the
relationship
between
uses,
and
each
is
subject
to
tho
constraint
of
available
data.
To
date,
very
little
has
been
done
to
measur~
water
quality
between
sites
or
over
time.
One
exception
would
be
the
RF
F
and
NSF
indexes,
which
measure
various
aspects
of
water
quality
and
weigh
them
using
informed
judgment.
Further
research
in
this
direction
could
lead
to
an
index
that
incorporates
individual
perceptions
and
unique
characteristics
of
an
area.

E­
10
APPENDIX
F
TRAVEL
COST:
SUPPORTING
TABLES
This
appendix
contains
tables
displaying
data
that
support
the
travel
cost
analysis
presented
in
Chapter
7.
Tables
F­
1
through
F­
4
provide
additional
data
for
the
benefits
calculations.
Table
F­
5
shows
the
tailored
models
that
were
estimated
for
selected
sites.

F­
1
 
­
ma,­
s
 
.
­.=­­
s..
­
­
­
­
­

Table
F­
1.
Distribution
of
Benefit
Estimates
(
Consumer
Surplus
Loss
Avoided
)
for
Loss
of
Use
of
the
Monongahela
River
by
Income
Levels
for
33
Sites
for
Individual
Users
Benefit
estimate
(
1977
dollars)
a
200
Income
O­
10­
20­
30­
40­
50­
60­
100­
150­
and
(
1981
dol
1
ars)
O
10
20
30
40
50
60
100
150
200
above
Total
o
­
5,000
0
1
0
0
0
0
2
0
1
0
5,000
­
10,000
1
1
3
1
1
0
1
0
1
0
10,000
­
15,000
1
0
1
1
0
0
0
0
2
1
15,000
­
20,000
2
2
0
1
3
0
1
0
2
0
20,000
­
25,000
1
1
1
1
1
0
0
1
0
0
25,000
­
30,000
2
1
2
0
0
1
2
0
0
0
30,000
­
35,000
1
0
0
0
0
0
0
0
3
0
35,000
­
40,000
0
0
0
1
0
0
2
1
1
0
40,000
­
45,000
0
0
0
0
0
0
0
1
0
0
45,000
­
50,000
0
0
1
1
0
0
2
0
0
0
50,000
and
above
g
g
()~~)~?
l
!!

Total
8
6
8
7
5
1
10
3
11
1
0
4
0
9
0
6
1
12
0
6
1
9
0
4
0
5
0
1
0
4
1
G;
 
3
a
To
convert
to
1981
dollars,
multiply
the
endpoints
of
the
benefit
scale
by
1.55.

.­

F­
2
Table
F­
2.
Distribution
of
Benefit
Estimates
(
Consumer
Surplus
Loss
Avoided)
Due
to
Loss
of
Use
of
the
Monongahela
River
by
Survey
User
Income
for
33
Sites­­
Includes
Multi~
le
Visits
Benefit
estimate
(
1977
dollars)=

Income
O­
10­
20­
30­
40­
50­
60­
70­
80­
(
1981
dollars)
10
20
30
40
50
60
70
80
90
Total
o
­
5,000
5,000
­
10,000
10,000
­
15,000
15,000
­
20,000
20,000
­
25,000
25,000
­
30,000
30,000
­
35,000
35,000
­
40,000
40,000
­
45,000
45,000
­
50,000
50,000
and
above
Total
o
0
0
0
1
1
0
0
0
0
g
2
0
0
0
0
0
0
0
0
0
0
g)

o
0
0
0
1
1
1
0
0
0
0
g
3
1
0
0
1
1
1
1
0
1
2
0
0
2
0
1
2
0
0
0
4
3
4
0
2
5
4
0
0
5
3
0
2'
7
5
1
0
2
0
4
4
8
2
0
1
21
0
1
1
1
0
1
0
1
0
0
4
0
g
,:
2
J?
 
5
39
17
10
11
8
18
6
22
7
3
3
4
2
 
94
a
To
convert
to
1981
dollars,
multiply
the
endpoints
of
the
benefit
scale
by
1.55.

F­
3
*
.
.
..­
.
.
.
.
.­
 
.
.
.
.
.
 
.
.
 
.
­.
..
 
.
­
 
 
.
.
.
._
 
.
 
 
.
 
_
.
 
­
T
 
.
 
.
..
 
 
.
...­­.
 
.
.­..
 
 
­­­
.
Table
F­
3.
Distribution
of
Benefit
Estimates
(
Consumer
Surplus
Increment)
Due
to
Water
Quality
Improvement:
Boatable
to
Fishable
by
Survey
User
Income
for
33
Sites­­
Includes
Multiple
Visits
Benefit
estimate
(
1977
dollars)
a
Income
(
1981
dollars)
o­
1o
10­
20
20­
30
30­
40
Tota
I
o
­
5,000
5,000
­
10,000
10,000
­
15,000
15,000
­
20,000
20,000
­
25,000
25,000
­
30,000
30,000
­
35,000
35,000
­
40,000
40,000
­
45,000
45,000
­
50,000
50,000
and
above
Total
o
0
0
1
3
5
7
3
3
4
2
 
28
1
0
0
5
3
17
0
0
0
0
0
 
26
7
11
8
12
0
0
0
0
0
0
0
 
38
2
0
0
0
0
0
0
0
0
0
Q
2
10
11
8
18
6
22
7
3
3
4
2
 
94
a
To
convert
to
1981
dollars,
multiply
the
endpoints
of
the
benefit
scale
by
1.55.

F
­
4
.
.

Table
F­
4.
Distribution
of
Benefit
Estimates
(
Consumer
Surplus
Increment)
Due
to
Water
Quality
Improvement:
Boatable
to
Swimmable
by
Survey
User
Income
for
33
Sites­­
Includes
Multiple
Visits
1
ncome
Benefit
estimate
(
1977
dollars)
a
(
1981
dollars)
o­
1o
10­
20
20­
30
30­
40
40­
50
50­
60
Total
o
­
5,000
5,000
­
10,000
10,000
­
15,000
15,000
­
20,000
20,000
­
25,000
25,000
­
30,000
30,000
­
35,000
35,000
­
40,000
40,000
­
45,000
45,000
­
50,000
50,000
and
above
Tota
I
o
0
0
0
1
3
3
3
3
4
2
 
19
0
0
0
2
2
11
4
0
0
0
_
Q
19
1
0
0
3
3
8
0
0
0
0
0
 
15
2
0
5
13
0
0
0
0
0
0
_
Q
20
4
11
3
0
0
0
0
0
0
0
0
 
18
3
0
0
0
0
0
0
0
0
0
g
3
10
11
8
18
6
22
7
3
3
4
2
 
94
a
To
convert
to
1981
dollars,
multiply
the
endpoints
of
the
benefit
scale
b
y
1.55.

.

F­
5
\

Table
F­
5.
Regression
Results
of
Tailored
Models
for
Selected
Sites
a
.
.
 
.
 
 
Site
 
 
.
 
.
 
Income
Site
Fnumber
Intercept
T+
M
cOst
Income
squared
Age
Sex
RECIMP
D
a
y
R
2
DF
ratio
Beaver
Lake,
AR
Lock
and
Dam
No.
2
302
2.63
(
Arkansas
River
(
7.12)
Navigation
System),
AR
2.39
(
8.24)

2.25
(
8.06)

1.94
(
6.02)

2.2s
(
9.41)

303
1.69
(
9.09)

1.70
(
11.98)

(
1::%

1.48
(
12.03)

1.74
(
16.32)

Blakely
Mt.
Dam,
Lake
307
1.58
Ouaqhlta,
AR
(
5.59)

1.53
(
5.87)

1.69
(
9.71)

1.28
(
5.95)

1.88
(
9.22)
­
0.012
(­
2.20)

­
0.012
(­
2.08)

­
0.013
(­
2.29)

­
0.013
(­
2.46)

­
0.010
(­
1.67)

­
0.007
(­
12.46)

­
0.007
(­
12.10)

­
0.007
(­
13.04)

­
0.007
(­
12.85)

­
0.006
(­
11.75)

­
0.008
(­
5.14)

­
O.
OO8
(­
5.18)

­
0.008
(­
5.08)

0.008
(­
5.23)

­
0.007
(­
4.89)
8
.
7
x
1o­
5
(­
1
.37)

­
1.8
X
10
­
5
(­
1.08)

­
1.6
X
10­
5
(­
0.86)

­
1.5
x
10­
5
(­
0.88)

­
8
.
5
x
1o­
6
(­
0.45)

.1.2
x
10­
5
(
0.68)

­
3.8
X
10­
6
(­
0.84)

­
2.3
x
10­
6
(­
0.51)

­
4.()
x
10­
6
(­
0.91)

­
1.9X
10
­
6
(­
0.43)

6.6X
10­
6
(
0.24)

­
6.3
X
10­
6
(­
0.79)

­
7
.
9
x
10
­
6
(­
1.01)

­
9
.
8
x
1o­
6
(­
1.31)

­
7.
OX
10
­
6
(­
0.92)
3
.
1
x
10­
9
(
1.12)

1.9
x
10­
9
(
0.
s1)

­
3.2
X
,()­
10
(­
0.53)
0.17
37
­
0.003
0.15
37
(­
0.50)

0.062
0.14
37
(.
037)

0.378
0.20
37
(
1.62)

­
0.209
0.18
37
(­
1.30)

0.43
222
­
0.003
0.44
222
(­
0.92)

0.212
0.45
222
(
2.32)

0.191
0.44
222
(
1.82)

­
0.310
0.46
222
(­
3.18)

0.24
87
0.005
0.24
87
(
0.84)

0.048
0.24
87
(
0.3­
1)

0.555
0.31
87
(
3.00)

­
0.275
0.26
87
(­
1
.
S6)
2.51
2.12
2.07
3.04
2.67
57.02
57.37
60.05
58.83
62.83
9.13
9
.
3
1
9.05
12.93
10.07
OF
=
Degrees
of
freedom.

at­
values
of
no
association
are
shown
in
parentheses.
RECIMP
is
a
binary
variable
that
is
1
if
the
respondent
ccnsiders
recreation
to
be
important
Day
is
a
binary
variable
that
Is
1
If
the
respondent
stayed
1
or
more
days..
Table
F­
5.
(
continued)
 
Site
Income
FSIte
number
Intercept
T+
fbl
cost
Income
squared
Age
Sex
RECIMP
13av
R
2
DF
ratio
Cordell
Hull
Dam
and
310
1.97
Reservoir,
TX
(
8.96)

1.58
(
7.74)

1.65
(
10.41)

1.87
(
9.14)

1.88
(
14.25)

Dewey
Lake,
KY
312
0.26
(
0.64)

0.16
(
0.55)

0.08
n
(
0.36)

A
0.43
(
2.17)

0.54
(
2.79)

Grapevine
Lake,
TN
314
1.54
(
7.16)

2.16
(
13.76)

1.74
(
13.98)

1.44
(
8.05)

1.80
(
15.13)
­
0.014
(­
5.94)

­
0.015
(­
6.28)

­
0.014
(­
6.33)

­
0.014
(­
5.93)

­
0.013
(­
5.63)

­
0.002
(­
2.74)

­
0.003
(­
3.19)

­
0.003
(­
3.67)

­
0.002
(­
2.91)

­
0.002
(­
1.85)

­
0.007
(­
8.78)

­
0.006
(­
7.89)

­
0.007
(­
8.85)

­
0.007
(­
9.26)

­
0.009
(­
6.62)
­
1.4
x
10
­
5
(­
0.63)
2.8
X
10­
6
(
0.33)
2.4
x
10­
6
(
0.29)

5.6
X
N)­
8
(
0.01)

7.4
x
10­
6
(
0.17)

3.6
X
lo­
s
(
1.01)

1
.
9
X
10­
5
(
1.91)

2.5
x
Io­
5
(
2.74)

2.()
x
10­
5
(
1
.99)

1.9X
10­
5
(
1.96)

3
.
5
X
10­
5
(
1.74)

7.6
X
10­
5
(
1.59)

8
.
0
x
10­
6
(
1.59)

9
.
4
X
10­
6
(
1.92)

9
.
2
x
1
0­
6
(
1.77)
3.6
x
,
0­
1
0
(
0.67)

­
3.,
x
,0­
10
(­
0.47)

­
5.4
x
10­'
0
(­
1.36)
0.007
(
1.74)

0.311
(
2.29)

0.009
(
1.24)

0,498
(
2.89)
0.34
0.36
0.37
­
0.021
0.34
(­
0.11)

­
0.208
0.35
(­
1.35)

0.18
0.21
0.31
­
0.018
0.18
(­
0.10)

­
0.359
0.24
(­
1.78)

0.48
­
0.013
0.52
(­
3.14)

0.109
0.47
(
1.00)

0.392
0.50
(
2.47)

­
0.296
0.44
(­
2.36)
100
I
00
100
100
100
42
42
42
42
42
88
88
88
88
88
17.10
18.40
19.52
16.88
17.79
3.16
3.70
6.46
3.08
4.37
26.94
31.95
26.41
29.60
22.88
DF
=
Degrees
of
freedom.

at­
values
of
no
association
are
shown
In
parentheses.
RECIMP
is
a
binary
variable
that
is
1
if
the
respondent
considers
recreation
to
be
important
Day
Is
a
binary
variable
that
Is
1
if
the
respondent
stayed
1
or
more
days.
Table
F­
5.
(
continued)
.
 
 
 
 
.
.
Site
Income
FSIte
number
Intercept
T+
M
cost
Income
squared
Age
Sex
RECIMP
Day
R*
DF
ratio
Greers
Ferry
Lake,
AR
315
1.49
(
8.04)

1.61
(
10.63)

1.45
(
12.25)

1.15
(
6.69)

1.76
(
15.29)

1.91
(
7.47)

1.81
(
7.07)

2.06
n
(
11.44)
Grenada
Lake,
MS
316
&
1.28
(
4.31)

2.03
(
1
3
.
0
'
)

Lake
Washington
Ship
320
2.69
Canal,
WA
(
3.27)

1.10
(
2.20)

0.81
(
2.11)

(::%
­
0.006
(­
8.
!
31)

­
0.006
(­
9.09)

­
0.006
(­
8.97)

­
0.007
(­
9.34)

­
0.006
(­
8.80)

­
0.010
(­
4.37)

­
0.009
(­
4.31)

­
0.009
(­
4.16)

­
0.010
(­
4.62)

­
0.008
(­
3.57)

­
0.004
(­
4.16)

­
0.003
(­
2.98)

­
0.004
(­
3.81)

­
0.004
(­
3.64)
7,3
x
10­
6
(
0.35)

9,6
X
10­
6
(
1.60)

8.4
X
10­
6
(
1.42)

g.
t)
x
10­
6
(
1.
ss)

1.()
x
10­
5
(
1.92)

2.1
x
10­
5
(
0.41)

­
5.(
J
x
10­
6
(­
0.32)

­
1.()
x
10­
5
(­
0.65)

­
9.6
x
10­
6
(­
0.67)

­
1.8
X
10
­
6
(­
0.12)

­
1
.
6
x
10­
4
(­
2.06)

1.6
X
10­
5
(
0.73)

1
.
9
X
10
­
5
(
0.94)

1.7
x
10­
5
(
0.81)
2.6
X
1
0­
1
1
(
0.05)

­
0.004
(­
1.14)

0.054
(
0.53)

­
1.6
X
10­
9
(­
0.63)

0.005
(
1.13)

­
0.049
(­
0.32)

4.3
x
10
­
9
(
2.36)

­
0.005
(­
0.52)

0.234
(
0.92)
0.372
(
2.39)

0.806
(
2.98)

­
0.079
(­
0.24)
0.28
0.28
0.28
0.29
­
0.494
0.35
(­
4.89)

0.22
0.23
0.22
0.30
­
0.419
0.28
(­
2.51)

0.35
0.26
0.27
0.26
EQUATION
5
IS
NOT
OF
FULL
RANK
BECAUSE
ALL
VISITS
WERE
DAY
VISITS.

Melvern
Lake,
KS
322
1.87
.­
0.008
­
6.7
X
10­
5
1.6x
It3
­
9
(
3.93)
0.11
(­
1.60)
(­
1.36)
(
1.5)

DF
=
Degrees
of
freedom.
213
213
213
213
213
72
72
72
72
72
39
39
39
39
41
27.07
27.66
27.20
29.70
38.06
6.76
7.14
7.36
10.36
9.26
6.95
4.57
4.83
4.48
1.69
at­
values
of
no
association
are
shown
in
parentheses.
RECIMP
is
a
binary
variable
that
is
1
if
the
respondent
considers
recreation
to
be
important
Day
is
a
binary
variable
that
is
1
if
the
respondent
stayed
1
or
more
days.

­+
isw,
Table
F­
5.
(
continued)
 
Site
.
 
Income
Site
number
Intercept
T+
M
cost
F­
1
ncome
squared
Age
Sex
RECIMP
Day
R
2
DF
ratio
Millwmd
Lake,
AR
i
I
Melvern
Lake,
KS
(
con.
)
322
1.10
(
2.39)

1.36
(
4.30)

0.96
(
2.38)

1.47
(
4.39)

323
1.48
(
4.82)

0.83
(
2.35)

0.98
(
4.68)

1.30
(
4.60)

1.
s1
(
8.13)

2.12
(
3.90)

1.01
(
1.89)

1.40
(
4.21)

0.94
(
2.41)

1.32
(
4.05)

Mississippi
River
Pool
325
1.23
No.
6,
MN
(
3.16)

1.24
Mississippi
River
Pool
No.
3,
MN
324
­
0.009
(­
1.72)

­
0.008
(­
1.69)

­
0.007
(­
1.46)

­
0.008
(­
1.62)

­
0.008
(­
3.96)

­
0.009
(­
4.45)

­
0.009
(­
4.59)

­
0.008
(­
3.94)

­
0.007
(­
3.47)

­
0.005
(­
4.22)

­
0.006
(­
4.53)

­
0.006
(­
4.44)

­
0.006
(­
4.97)

­
0.006
(­
4.56)

­
0.007
(­
4.31)

­
0.007
4
.
8
X
10­
6
(
0.36)

5.9
x
lt)­
6
(
0.43)

5.1
x
10­
6
(
0.39)

7.0
x
10­
6
(
0.52)

1
.
1
x
10­
5
(
0.39)

2
.
1
x
10­
5
(
2.57)

1.7
x
10­
5
(
2.29)

1.7
x
10­
5
(
2.03)

1
.
9
x
10­
5
(
2.34)

­
7
.
2
x
1
0
­
5
(­
1
.63)

4
.
8
X
10­
6
(
0.55)

4
.
4
x
10­
6
(
0.50)

3
.
1
x
10­
6
(
0.37)

4
.
5
x
10­
6
(
0.51)

3.1
x
10­
5
(
0.88)

1.4
x
10­
5
0.005
(
0.57)

­
0.135
(­
0.49)

,.
2
x
10­
10
(
0.20)

0.013
(
1,96)

0.691
(
3.41)

1.4
x
10­
9
(
1.78)

0.008
(
0.74)

­
0.143
(­
0.73)

­
3
.
5
x
,1)­
1~
(­
0.51)

0.005
0.07
0.07
0.380
0.09
(
1.21)

­
0.303
0.08
(­
1.02)

0.25
0.30
0.39
0.166
0.25
(
0.59)

­
0.333
0.28
(­
1.50)

0.38
0.34
0.34
0.539
0.38
(
1.69)

0.036
0.34
(
0.18)

0.22
0.23
41
41
41
41
49
49
49
49
49
45
45
45
45
45
66
66
DF
=
Degrees
of
freedom.

at­
values
of
no
association
are
shown
in
parentheses.
RECIMP
tant.
Day
is
a
binary
variabie
that
is
1
if
the
respondent
stayed
1.01
0.98
1.42
1.26
5.41
7.10
10.54
5.55
6.39
9.20
7.89
7.88
9.05
7.63
6.34
6.46
 
 
is
a
binary
variable
that
is
1
if
the
respondent
considers
recreation
to
be
impor­

1
w
more
days.

I
I
I
Table
F­
5.
(
continued)
._

Site
Income
FSlte
number
Intercept
T+
M
cost
Income
squared
Age
Sex
RECIMP
D
a
y
Rz
DF
r
a
t
i
o
Ozark
Lake,
AR
n
I
o
Philpott
Lake,
VA
Mississippi
River
Pool
325
(
4.19)
No.
6,
MN
(
con.
)
1.45
(
6.21)

0.98
(
3.43)

1.42
(
6.74)

331
1.53
(
5.06)

1.64
(
5.25)

1.71
(
7.57)

1.42
(
5.11)

1.80
(
9.04)

333
1.61
(
5.17)

2.26
(
6.85)

2.01
(
9.05)

1.40
(
3.61)

1.92
(
10.03)

Pine
River,
MN
334
0.19
(
0.50)

0.69
(
2.69)
(­
4.31)

­
0.007
(­
4.05)

­
0.007
(­
3.88)

­
0.007
(­
4.15)

­
0.005
(­
4.45)

­
0.005
(­
4.40)

­
0.004
(­
4.19)

­
0.005
(­
4.58)

­
0.003
(­
3.15)

­
0.009
(­
4.56)
­
0.009
(­
4.39)
­
0.008
(­
3.98)
­
0.009
(­
4.64)
­
0.007
(­
3.61)

­
0.001
(­
0.90)

­
0.002
(­
1.36)
(
1.58)

1.3X
lo­
5
(
1.50)

1.
OX
10­
5
(
1.20)

7.4X
70­=
(
1.53)

1
.
3
X
10­
5
­
6
.
2
X
lf)­'
o
(
0.32)
(­
0.58)

­
8
.
6
x
10­
s
(­
0.63)

­
1.
ox
10­
4
(­
0.73)

­
7
.
1
x
10­
6
(­
0.53)

­
2.()
x
10­
5
(­
1.15)

4
.
2
X
10­
5
­
1
.
3
x
10­
9
(
1.10)
(­
1.22)

­
8.6
x
10­
7
(­
0.006)

­
1.5
x
10
­
6
(­
0.12)

5.5
x
10
­
6
(
0.40)

3.4
x
?
0­
6
(
0.27)

5.0
x
10­
5
1.1
x
10
­
9
(
1.64)
(­
1.90)

­
6.6
x
10­
6
(
­
0
.
9
5
)
(
0.78)

­
0.074
0.22
(­
0.36)

0.537
0.27
(
2.08)

­
0.040
0.22
(­
0.21)

0.32
0.001
0.31
(
0.09)

­
0.96
0.32
(­
0.47)

0.285
0.33
(
1.18)

­
0.
s41
0.37
(­
2.15)

0.39
­
0.011
(­
1.41)
0.40
­
0.232
0.39
(­
1.26)

0.449
0.40
(
1.48)

­
0.483
0.46
(­
2.43)

0.08
0.004
0.04
(
0.62)
66
66
66
48
48
48
48
48
34
34
34
34
34
71
71
6.29
8.08
6.25
7.46
7.30
7.40
7.98
9.54
7.28
7.53
7.33
7.64
9.60
2.16
1.04
OF
=
Degrees
of
freedom.

at­
values
of
no
association
are
shown
in
parentheses.
RECIMP
is
a
binary
variable
that
is
1
jf
the
respondent
considers
recreation
to
be
important
Day
Is
a
binary
variable
that
is
1
if
the
respondent
stayed
1
or
more
days,

b.,
Table
F­
5.
(
continued)
 
­
 
 
 
 
.
 
Site
Income
Site
Fnumber
Intercept
T+
M
cost
1
ncome
squared
Age
Sex
RECIMP
Day
R
2
DF
ratio
.
 
 
Proctor
Lake,
TX
Sardls
Lake,
MS
I
Whitney
Lake,
TX
DF
=
Degrees
of
freedom
Pine
River,
MN
(
corl.
)
334
0.82
(
4.51)

0.53
(
2.36)

1.07
(
3.42)

337
2.13
(
8.57)

1.81
(
6.57)

1.99
(
12.86)

?.
94
(
7.54)

2.06
(
11.79)

340
2.07
(
13.95)

1.91
(
13.52)

1.84
(
18.39)

1.12
(
7.57)

1.88
(
20.69)

344
1.50
(
8.68)

1.40
(
8.79)

1.34
(
11.61)

1.23
(
9.22)

1.83
(
14.50)
 
.
_
_
_
~­
 
_
__.
­
0.002
(­
1.06)

­
0.002
(­
1.25)

­
0.002
(­
1.31)

­
0.013
(­
6.48)

­
0.013
(­
7.50)

­
0.014
(­
7.86)

­
0.013
(
7.41)

­
0.013
(­
7.09)

­
0.004
(­
3.95)

­
0.003
(­
3.07)

­
0.003
(­
3.14)

­
0.003
(­
3.93)

­
0.003
(­
3.50)

­
0.003
(­
1.70)

­
0.002
(­
1.75)

­
0.003
(­
1.83)

­
0.003
(­
1.74)

­
0.003
(­
2.09)
­
6.5
X
10­
0
(­
0.92)

­
8
,
2
X
10­
6
(­
1.19)

­
5.3
x
lf)­
6
(­
0.75)

­
6.8
x
10­
6
(­
0.25)
3.7
x
10­
6
(
0.53)

­
3
.
0
x
10­
7
(­
0.04)

1
.
3
x
10­
6
(
0.20)

1
.
2
x
10­
6
(
0.17)

­
2
.
9
x
10­
5
(­
1.78)

3.3
x
10­
6
(
0.57)

4
.
5
x
10­
6
(
0.81)

4
.
2
x
10­
6
(
0.80)

7.5
x
10­
6
(
1.32)

­
7
.
7
x
10­
6
(­
0.45)

3
.
3
x
10­
6
(
0.73)

2
.
9
x
10­
6
(
0.64)

1.6
x
10­
6
(
0.35)

3.4
x
10­
6
(
0.81)
_
 
 
_
=.
 
 
__
..
__=
­
0.092
(­
0.20)

0.363
(
1.91)

­
0.291
(­
0.99)

1.5X
lt)
­
lo
(
0.32)

0.005
(
1.11)

0.273
(
1.81)

0.139
(
0.61)

0.010
(
0.05)

8
.
6
x
10­
10
(
2.18)

­
0.003
(­
0.97)

­
0.057
(­
0.68)

2.3
X
,0­
10
(
0.67)

0.0003
(
0.09)

0.160
(
1.50)

 
.
­­
 
 
.
0.04
0.08
0.05
0.54
0.55
0.57
0.54
0.54
0.07
0.05
0.05
71
71
71
48
48
48
48
48
201
201
201
0.767
0.18
(
5.58)

­
0.208
0.08
(­
2.61)

0.02
0.02
0.03
0.271
0.04
(
2.19)

­
0.601
0.15
(­
5.61)
201
201
198
198
198
198
198
0.93
2.17
1.25
18.61
19.43
20.89
18.61
18,54
5.13
3.79
3.63
14.38
5.84
1.30
1.15
1.91
2.78
11.81
l
t
­
v
a
l
u
e
s
o
f
n
o
associatior~
are
shown
i
n
P
a
r
e
n
t
h
e
s
e
s
.
RECIMP
i
s
a
b
i
n
a
r
y
v
a
r
i
a
b
l
e
that
i
s
I
i
f
t
h
e
r
e
s
p
o
n
d
e
n
t
c
o
n
s
i
d
e
r
s
r.
ecreaLio!!
t
o
b
e
important
Day
1$
a
binary
variable
that
is
1
if
the
respondent
stayed
1
or
more
days,

1
.
.
.
.
.
.
.
.
APPENDIX
G
ALTERNATIVE
REGRESSION
MODELS
This
appendix
provides
a
detailed
listing
of
the
alternative
specifications
of
regression
models.
Listings
are
given
for
both
the
survey
and
travel
cost
models.

G­
1
I
Table
G­
1.
Independent
variable
combinations
used
in
option
price,
user
value,
and
option
value
regressions.
Dependent
variables
are
dollar
bids
given
for"
changes
in
water
quality.
 
 
­
 
 
 
_
_
 
 
 
 
_
_­
 
.
 
 
.
 
Dummy
Bidding
Variables
vs.
Nonto
Denote
Bidding
Survey
Sex
Game
Age
Education
Income
Version
Ounsny
R
Oummy
Dummy
variables
Attitude
Water
variables
to
ength
towards
Quality
to
denote
of
Attitude
cost
User
Ratin
9
denote
Providence
Indexl
Dunmy
Dummy
Dummy
I
n
t
e
r
v
i
e
w
e
r
fession
I
n
d
u
s
t
r
y
Dummy
V
a
r
i
a
b
l
e
s
t
o
denote
SIC
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
;
x
;
:
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
0
x
x
x
x
x
r­
i
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
(
continued)
b
T
a
b
l
e
G­
1
(
continued)

 
 
.
 
Dummy
Bidding
V
a
r
i
a
b
l
e
s
v
s
.
Nonto
Denote
Bidding
Survey
Game
Sex
Age
Education
Income
Version
Dummy
A
t
t
i
t
u
d
e
Water
Length
towards
Q;
ua;
:
y
o
f
A
t
t
i
t
u
d
e
cost
User
Residence
Indexi
Dummy
Dummy
Dummy
i
Dummy
Dummy
variab\
es
Dummy
v
a
r
i
a
b
l
e
s
t
o
Variables
t
o
denote
to
denote
Pro­
denote
SIC
I
n
t
e
r
v
i
e
w
e
r
f
e
s
s
i
o
n
I
n
d
u
s
t
r
y
x
x
x
x
x
x
x
x
­
x
x
x
x
x
x
x
x
x
x
x
x
x
x
 
 
 
 
.

`
This
index
was
constructed
by
adding
responses
to
various
attitudinal
questions.
2
See
question
number
B­
l­
b
in
the
survey
questionnaire.

c1
A
%
x
x
%

x
%
x
x
x
%

%
x
x
%

Yc
x
x
.
.
.
.
.
.
.
&
.
.
.
8X
x
x
x
x
x
%
x
x
x
*
%

<
x
x
<
x
x
%
%
x
.
­
­
­
­
­
x
=

x
x
x
x
x
%
x
x
x
x
­
­
.
x
x
­
0
w~

zc0
%
x
x
x
s
x
x
x
x
x
x
x
x
x
x
x
x
x
x
%
x
­
­
 
x
x
x
x
x
x
x
x
%

x
x
x
x
x
x
x
x
x
x
x
%
x
x
x
%
x
x
x
x
x
x
x
x
x
%
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
G­
4
..­
%
%
x
%
x
x
%
x
x
x
x
x
x
x
x
x
.
­
0
x
x
x
x
x
x
x
x
%
x
x
x
w3c5c0
x
x
x
x
x
:

x
x
x
x
x
%
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
%
x
%
%
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
%
x
x
x
x
x
x
x
x
x
x
x
%
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
%
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
xx
x
x
x
x
x
x
x
x
x
x
x
x
x
%

x
x
x
x
x
x
x
x
x
x
x
x
x
x
%
x
x
G­
5
t
c1
A
Table
G­
2
(
continued)

 
.

Travel
On­
Site
Site
and
1/
3
and
1/
3
Recrea­
Oay
ravel
Trave'
Travel
Travel
Site
tion­
Oay
Hour
Site
and
and
and
and
and
Impor­
Travel
Site
cost
Mile
Hi
le
T
r
a
v
e
l
tance
In­
I
n
c
o
m
e
Day
cost
Hour
cost
Race
Slope
cost
cost
cost
Dumy
come
Squared
Ousmtyl
0umny2
Age
Sex
Dummy3
oummy'
Dummy5
DUMMy6
T
i
m
e
M
i
l
e
S
i
t
e
M
i
l
e
M
i
l
e
c
o
s
t
c
o
s
t
c
o
s
t
cost
cost
.
.
x
,.
x
.
.
x
x
"
"
.

.
.
,.
x
x
,.
x
x
x
x
.
.
x
.
.
.
.

R
x
x
x
x
x
x
x
u
Y
tiplying
day
by
travel
cost.
if
stayed
less
than
one
hour.
Liplying
hour
by
site
cost.
if
white
and
zero
otherwise.
tiplying
day
by
onsite
cost.
­­
lIntercept
dunsny
equal
to
one
if
the
respondent
stays
one
or
more
days
and
zero
otherwise.
`
Slope
duimsy
calculated
by
mu'
~
Intercept
dunmry
equal
to
One
4SloPe
d­
y
calculated
by
mu'
`
Intercept
dussny
equal
to
one
`
Slope
dunsny
calculated
by
mu'
Table
G­
3,
.
Independent
variable
combinations
used
as
tailored
models
for
a
subsample
of
the
43
outdoor
recreation
survey
sites.
Dependent
variable
is
LN
(
visits).
 
 
 
 
 
 
 
 
 
 
 
 
1
/
3
Travel
On
Travel
&
Travel
&
D
a
y
T
r
a
v
e
l
Recreation
Time
Plus
Time
M
i
l
e
S
i
t
e
M
i
l
e
M
i
l
e
Importance
I
ncoine
Day
Mile
Cost
cost
cost
cost
cost
cost
Dummy
Income
Camp
i
ng
Squared
Dummy
Dummy
1
Age
Sex
Dummy
x
x
x
x
)(
)(
x
x
x
x
x
x
.
.
.
.
.
.
.
x
x
x
x
Y
x
x
x
x
x
x
x
x
x
4
n
A
A
h
x
x
x
x
x
x
x
x
x
x
x
x
K
x
x
h
h
A
A
x
x
x
x
x
x
A
A
u
x
x
x
x
x
(
continued)
Table
G­
3
(
continued)
 
 
 
 
.
1
/
3
 
.
 
 
.
 
Day
Travel
Travel
On
Travel
&
Travel
&
Recreation
Time
Plus
Time
M
i
l
e
S
i
t
e
M
i
l
e
M
i
l
e
Importance
Income
Oay
cost
Mile
Cost
Camping
cost
cost
cost
cost
Oummy
Income
Squared
Oummy
Oummy
1
Age
Sex
Dummy
x
x
x
x
x
x
x
x
x
f
x
x
G)
&
,.
,.
m
x
x
x
x
"
"
.

n
h
A
h
x
x
x
x
x
x
x
x
x
x
"
"
.

x
x
x
x
x
x
x
x
M
x
x
x
x
x
x
x
x
)(
u
u
A
x
x
)(
K
x
x
`
intercept
dummy
equal
to
one
is
respondent
engaged
I
n
camping.
 
­.

1
