63
ATTACHMENT
1:
Informational
Slide
Show
125
ATTACHMENT
2:
Slide
Show
Script
132
ATTACHMENT
3:
Full
Text
of
Stated
Preference
Survey
Component
145
ATTACHMENT
4:
Federal
Register
Notice
154
ATTACHMENT
5:
Description
of
Statistical
Survey
Design
155
Details
of
Statistical
Design
for
316(
b)
Stated
Preference
Survey
This
attachment
discusses
the
statistical
design
for
the
proposed
316(
b)
choice
experiment
survey.

General
Approach
to
Experimental
Design
Experimental
design
for
the
proposed
choice
experiment
survey
follows
established
practices
published
in
the
literature.
Fractional
factorial
design
is
used
to
construct
choice
questions
with
an
orthogonal
array
of
attribute
levels,
with
questions
randomly
assigned
to
individual
respondents
(
Louviere
et
al.
2000;
Louviere
1988).
Unlike
paper
(
e.
g.,
mail)
surveys,
the
proposed
internet
survey
approach
allows
for
random
assignment
of
questions
from
the
experimental
design
to
individual
respondents,

thereby
eliminating
the
need
to
assign
(
or
"
block")
the
questions
to
predetermined
survey
versions.
This
has
numerous
advantages,
among
them
the
elimination
of
any
unanticipated
anchoring
or
sequence
effects
that
might
result
from
a
pre­
established
ordering
of
questions
in
particular
survey
versions.
Based
on
standard
choice
experiment
experimental
design
procedures
(
Louviere
et
al.
2000),
the
number
of
questions
per
respondent
(
3)
is
determined
by:
a]
the
number
of
attributes
in
the
final
experimental
design
and
complexity
of
questions,
b]
particular
interactions
and
higher­
level
effects
for
which
estimates
are
desired,
and
c]
pretests
revealing
the
number
of
choice
experiment
questions
that
respondents
are
willing/
able
to
answer
in
a
single
survey
session,
and
d]
the
number
of
attributes
that
may
be
varied
within
each
question
while
maintaining
respondents'
ability
to
make
appropriate
neoclassical
tradeoffs.

Based
on
the
general
approach
outlined
above
and
recommendations
in
the
literature,
EPA
has
created
an
experimental
design
that
allows
for
an
ability
to
estimate
main
effects,
quadratic
effects,
and
two­
way
interactions
between
policy
attributes
(
Louviere
et
al.
2000).
Choice
sets
(
Bennett
and
Blamey
2001),
including
variable
level
selection,
have
been
designed
based
on
the
goal
of
illustrating
policy
scenarios
that
"
span
the
range
over
which
we
expect
respondents
to
have
preferences,
and/
or
are
practically
achievable"
(
Bateman
et
al.
2002,
p.
259),
and
that
are
viewed
as
realistic
by
respondents.

Details
of
the
Proposed
Experimental
Design
The
experimental
design
was
conducted
by
a
Donald
Anderson,
President
of
StatDesign,
Inc.,
a
statistician
with
significant
experience
in
experimental
designs
for
choice
experiments
(
e.
g.,
Johnston
et
al.
2003;
Newell
and
Swallow
2002)
As
noted
above,
the
experimental
design
allows
for
both
main
effects,
quadratic
effects,
and
selected
two­
way
interactions
to
be
efficiently
estimated,
based
on
a
choice
experiment
framework.

Each
treatment
(
survey
question)
includes
two
choice
Options
(
indicated
by
subscripts
A
and
B),

characterized
by
four
attributes
that
vary
across
the
two
choice
options
(
Fish
Saved,
Population
Change,

Catch
Change,
and
Household
Cost).
Hence,
there
are
a
total
of
eight
attributes
for
each
treatment.
156
Based
on
focus
groups
and
pretests,
and
guided
by
realistic
ranges
of
attribute
outcomes,
EPA
allows
for
four
different
potential
levels
for
Fish
Saved,
Population
Change,
and
Catch
Change,
and
allows
for
five
different
levels
of
monthly
Household
Cost.
Levels
for
attributes
were
chosen
to:
a]
span
a
range
of
sufficient
width
to
capture
all
realistic
policy
combinations,
b]
account
for
the
statistical
implications
of
choice
set
design
(
Hanemann
and
Kanninen
1999),
and
c]
provide
policy
options
viewed
as
feasible
by
respondents,
to
minimize
potential
protest
responses.
To
this
end,
results
of
focus
groups
were
indispensable
in
developing
appropriate
choice
sets
(
Bennett
and
Blamey
2001).

Attributes
and
levels
embedded
in
the
experimental
design
are
summarized
as
follows:


Fish
SavedA,
Fish
SavedB
(
4
possible
levels
each:
25,
50,
75,
and
95%)


Population
ChangeA,
Population
ChangeB
(
4
possible
levels
each:
0,
5,
8,
and
10%)


Catch
ChangeA,
Catch
ChangeB
(
4
possible
levels
each:
0,
2,
5,
and
10%)


CostA,
CostB
(
5
possible
levels
each;
$
1,
$
2,
$
3,
$
4,
$
6)

The
subscripts
(
A,
B)
denote
to
the
attributes
in
Option
A
and
B,
respectively.
The
available
or
potential
levels
for
each
attribute
are
balanced
across
A
and
B.
In
addition,
there
is
a
context
variable
which
indicates
the
starting
point
of
fish
population,
denoted
Baseline.
This
context
attribute
is
the
same
across
Options
A
and
B
for
each
question,
but
may
influence
the
marginal
utility
of
the
other
attributes
through
interactions
in
the
utility
function.


BaselineAB
(
4
possible
levels;
40%,
50%,
60%,
70%)

Beyond
the
levels
specified
above,
each
question
includes
a
"
neither
plan"
or
status
quo
option,

characterized
by
zero
values
for
all
attributes
except
for
the
Baseline.
This
represents
the
status
quo
option
necessary
for
appropriate
welfare
estimation
in
a
choice
experiment
context
(
Adamowicz
et
al.

1998).

Following
standard
practice,
EPA
constrained
the
design
somewhat
in
response
to
findings
in
focus
groups
and
the
prior
literature.
For
example,
focus
groups
showed
that
respondents
react
negatively
and
often
protest
when
offered
choices
in
which
one
option
dominates
the
other
in
all
attributes.
Given
that
such
choices
provide
negligible
statistical
information
compared
to
choices
involving
nondominant
dominated
pairs,
they
are
often
avoided
in
choice
experiment
statistical
designs
(
Louviere
et
al.

2000).
For
example,
Hensher
and
Barnard
(
1990)
recommend
eliminating
profiles
including
dominating
or
dominated
profiles,
because
such
profiles
generally
provide
no
useful
information.
Following
this
guidance,
EPA
constrained
the
design
to
eliminate
such
dominant/
dominating
pairs.
EPA
also
constrained
the
design
to
eliminate
the
possibility
of
pairs
in
which,
when
looking
across
two
options,
one
of
the
options
offers
both
a
greater
reduction
in
fish
losses
and
a
smaller
increase
in
the
population.
The
elimination
of
such
apparently
nonsensical
(
or
non­
credible)
pairs
is
common
practice,
and
is
done
to
avoid
protest
bids
and
confusion
among
respondents
(
Bateman
et
al.
2002).
As
shown
by
the
statistical
157
evaluation
of
the
design
presented
below,
these
constraints
do
not
negatively
influence
the
performance
and
properties
of
the
design
in
a
significant
manner.

The
resulting
experimental
design
is
characterized
by
64
unique
A
vs.
B
option
choice
sets,
where
attribute
levels
for
alternatives
A
and
B
differ
across
each
of
the
pairs.
Each
choice
set
represents
a
unique
choice
modeling
question 
with
a
unique
set
of
attribute
levels
distinguishing
options
A
and
B.

As
noted
above,
each
alternative
within
a
choice
set
is
specified
by
attributes
Fish
Saved
(
4
levels),

Population
Change
(
4
levels),
Catch
Change
(
4
levels),
and
Cost
(
5
levels).
In
addition,
there
is
a
context
variable
Baseline
(
4
levels)
which
indicates
the
starting
point
of
the
population
and
is
common
to
both
alternatives
in
a
set.
Also
as
noted
above,
the
design
requirements
include
the
ability
to
estimate
linear
by
linear
interactions
of
Baseline
with
all
attributes
and
all
combinations
of
Fish
Saved,
Population
Change,
and
Catch
Change.

The
base
design
used
was
the
4x4x4
full
factorial
and
all
of
associated
interactions.
This
base
design
was
used
to
generate
64
profiles
to
populate
alternative
A
(
with
only
four
Cost
levels).
A
second
randomization
of
the
4x4x4
full
factorial
was
used
to
generate
64
profiles
for
alternative
B
(
again
with
only
four
Cost
levels).
To
accommodate
the
five
level
Cost
attribute
the
levels
1,
2,
3,
5
were
used
for
the
alternative
A
profiles,
and
levels
1,
3,
4,
5
for
the
alternative
B.
Over
the
whole
design,
cost
levels
1,
3,
5
appear
more
frequently
than
levels
2
and
3.
This
increases
the
efficiency
of
the
design
slightly
for
estimating
linear
and
quadratic
effects
over
uniform
distribution
of
the
five
levels.
The
combined
128
profiles
allow
orthogonal
estimation
of
all
linear
and
quadratic
main
effects
and
all
required
linear
by
linear
interactions.

The
constraints
on
the
A
v.
s.
B
pairings
did
not
allow
an
orthogonal
construct
between
pairs
nor
any
cyclic
generation
of
pairs.
A
computer
program
was
written
to
generate
pairings,
test
for
all
constraints,
and
to
the
extent
possible
balance
attributes
across
pairs.
In
a
few
cases
(
fewer
than
5%
of
the
profiles)
the
Cost
attribute
was
changed
to
allow
the
pairings
without
violating
constraints
and
maintaining
efficiency.
The
resultant
128
profiles
arranged
into
64
choice
set
pairs
allow
near
orthogonal
estimation
of
all
linear
and
quadratic
main
effects
and
all
required
interactions.

The
64
choice
sets
have
been
ordered
by
the
baseline
attribute.
As
the
current
plan
is
to
randomly
select
three
choice
sets
for
each
respondent
(
the
order
of
the
profiles
within
choice
set
could
also
be
randomized
since
these
are
generic
alternatives),
this
arbitrary
ordering
does
not
affect
the
efficiency
or
performance
of
the
design.
However,
certain
constraints
may
be
placed
on
the
random
assignment
of
questions
to
respondents.
In
particular,
assignment
may
be
constrained
such
that
each
respondent
views
choice
sets
with
only
a
single
baseline,
to
avoid
potential
confusion
that
would
result
from
considering
resource
changes
under
changing
baselines.
158
Statistical
Attributes
of
Proposed
Design
To
test
the
statistical
performance
and
properties
of
the
proposed
experimental
design,
a
variety
of
standard
evaluations
were
conducted.
These
include
an
assessment
of
the
Pearson
correlation
coefficients
and
Variance
Inflation
Factors
associated
with
the
main
effects,
quadratic
effects,
and
interactions
for
which
EPA
desires
parameter
estimates.
Results
of
these
assessments
are
found
in
Tables
1
and
2.

Table
1:
Correlations
(
Pearson)
BASEL
FISHSL
POPCHL
CATCHL
COSTL
FISHSQ
POPCHQ
CATCHQ
FISHSL
0.000
POPCHL
0.000
0.000
CATCHL
0.000
0.000
0.000
COSTL
­
0.052
0.057
0.033
0.052
FISHSQ
0.000
0.000
0.000
0.000
0.011
POPCHQ
0.000
0.000
0.000
0.000
­
0.011
0.000
CATCHQ
0.000
0.000
0.000
0.000
­
0.053
0.000
0.000
COSTQ
0.028
0.000
0.020
0.028
­
0.007
0.009
0.027
­
0.027
BxFISH
0.000
0.000
0.000
0.000
­
0.042
0.000
0.000
0.000
BxPOP
0.000
0.000
0.000
0.000
0.002
0.000
0.000
0.000
BxCATCH
0.000
0.000
0.000
0.000
0.006
0.000
0.000
0.000
FxP
0.000
0.000
0.000
0.000
­
0.004
0.000
0.000
0.000
FxC
0.000
0.000
0.000
0.000
­
0.008
0.000
0.000
0.000
PxC
0.000
0.000
0.000
0.000
­
0.019
0.000
0.000
0.000
BxCOST
0.036
­
0.043
0.002
0.006
0.023
0.053
­
0.005
0.005
(
continued)
COSTQ
BxFISH
BxPOP
BxCATCH
FxP
FxC
PxC
BxFISH
0.021
BxPOP
0.012
0.000
BxCATCH
0.027
0.000
0.000
FxP
­
0.011
0.000
0.000
0.040
FxC
­
0.032
0.000
0.040
0.000
0.000
PxC
­
0.037
0.040
0.000
0.000
0.000
0.000
BxCOST
­
0.030
0.042
0.045
0.049
0.019
0.084
­
0.049
(
Note:
Within
interaction
estimates,
B
=
Baseline;
F
=
FishSL;
P
=
PopCHL;
C
=
Catch)
159
Table
2:
Variance
Inflation
Factors
Predictor
VIF
BASEL
1.0
FISHSL
1.0
POPCHL
1.0
CATCHL
1.0
COSTL
1.0
FISHSQ
1.0
POPCHQ
1.0
CATCHQ
1.0
COSTQ
1.0
BxFISH
1.0
BxPOP
1.0
BxCATCH
1.0
FxP
1.0
FxC
1.0
PxC
1.0
BxCOST
1.0
Note:
VIF
=
0
implies
orthogonal
estimation
Results
from
Tables
1
and
2
clearly
indicate
the
orthogonality
of
the
proposed
experimental
design
and
the
appropriateness
for
estimation
of
the
designed
main,
quadratic,
and
interaction
effects.
This,
paired
with
focus
group
results
indicating
the
general
appropriateness
of
the
selected
attribute
levels,
indicate
that
the
proposed
experimental
design
will
allow
appropriate
implementation
and
estimation
of
the
desired
choice
experiment
model.

References
Adamowicz,
W.,
P.
Boxall,
M.
Williams,
and
J.
Louviere.
"
Stated
Preference
Approaches
for
Measuring
Passive
Use
Values:
Choice
Experiments
and
Contingent
Valuation."
American
Journal
of
Agricultural
Economics
80(
1998):
64­
75.

Bateman,
I.
J.,
R.
T.
Carson,
B.
Day,
M.
Hanemann,
N.
Hanley,
T.
Hett,
M.
Jones­
Leee,
G.
Loomes,
S.

Mourato,
E.
Ozdemiroglu,
D.
W.
Pierce,
R.
Sugden,
and
J.
Swanson.
2002.
Economic
Valuation
with
Stated
Preference
Surveys:
A
Manual.
Northampton,
MA:
Edward
Elgar.

Bennett,
J.
and
R.
Blamey,
eds.
2001.
The
Choice
Modelling
Approach
to
Environmental
Valuation.

Northampton,
MA:
Edward
Elgar.

Dillman,
D.
A.
2000.
Mail
and
Internet
Surveys:
The
Tailored
Design
Method.
New
York:
John
Wiley
and
Sons.

Hanemann,
W.
M.
and
B.
Kanninen.
1999.
"
The
Statistical
Analysis
of
Discrete­
Response
CV
Data."
In
Valuing
Environmental
Preferences:
Theory
and
Practice
of
the
Contingent
Valuation
Method
in
the
US,
EU,
and
Developing
Countries.
Edited
by
I.
J.
Bateman
and
K.
G.
Willis,
Oxford
University
Press,
Oxford,
UK.

Hensher,
D.
A.
and
P.
O.
Barnard.
1990.
"
The
Orthogonality
Issue
in
Stated
Choice
Designs."
in
Fischer,

M.,
P.
Nijkamp
and
Y.
Papageorgiou,
eds.
Spatial
Choices
and
Processes.
Amsterdam:
North
160
Holland,
265­
78.

Johnston,
R.
J.,
S.
K.
Swallow,
T.
J.
Tyrrell
and
D.
M.
Bauer.
2003.
"
Rural
Amenity
Values
and
Length
of
Residency."
American
Journal
of
Agricultural
Economics
85(
4):
1000­
1015.

Louviere,
J.
J.
1988.
Analyzing
Decision
Making:
Metric
Conjoint
Analysis.
Sage
University
Paper
67.

Newbury
Park:
Sage
Publications.

Louviere,
J.
J.,
D.
A.
Hensher,
and
J.
D.
Swait.
2000.
Stated
Preference
Methods:
Analysis
and
Application.
Cambridge,
UK:
Cambridge
University
Press.

Newell,
L.
W.
and
S.
K.
Swallow.
2002.
"
Are
Stated
Preferences
Invariant
to
the
Prospect
of
Real­
Money
Choice?"
Selected
paper,
American
Agricultural
Economics
Association,

Long
Beach,
California,
July
28­
31.
