1
Although
the
U.
S.
Supreme
Court
recently
reaffirmed
that
the
Clean
Air
Act's
National
Ambient
Air
Quality
Standards
(
NAAQS)
should
be
set
without
regard
to
costs,
the
Court
also
recognized
that
the
Clean
Air
Act
calls
on
states
and
EPA
to
consider
costs
in
implementing
those
standards.
In
addition,
several
Executive
Orders
require
cost
analysis
of
major
regulatory
actions.

2
Winston
Harrington,
Robert
D.
Morgenstern,
and
Peter
Nelson,
"
On
the
Accuracy
of
Regulatory
Cost
Estimates,"
Resources
for
the
Future
Discussion
Paper
99­
18,
January
1999.

1
Assessing
the
Impact
of
Progress
and
Learning
Curves
on
Clean
Air
Act
Compliance
Costs
Cynthia
J.
Manson,
Matthew
B.
Nelson,
and
James
E.
Neumann
Industrial
Economics,
Incorporated
Cambridge,
MA
Revised
July
12,
2002
INTRODUCTION
The
accuracy
of
EPA
estimates
of
regulatory
compliance
costs
has
long
been
a
primary
focus
of
discussion
among
academics,
regulators,
regulated
entities,
and
other
stakeholders.
1
The
EPA
Office
of
Air
and
Radiation
relies
on
cost
estimates
to
design
economically
efficient
strategies
to
implement
all
aspects
of
the
Clean
Air
Act
and
its
1990
Amendments,
including
federally
mandated
stationary
and
mobile
source
controls
that
support
NAAQS
compliance;
controls
on
hazardous
air
pollutants;
measures
to
reduce
emissions
of
stratospheric
ozone
depleting
substances;
and
requirements
to
limit
acid
rain
precursor
emissions.
Because
of
the
central
role
of
costs
in
the
regulatory
process,
EPA
is
continually
assessing
methods
for
refining
approaches
to
cost
estimates.

One
way
to
improve
the
accuracy
of
regulatory
cost
estimates
is
identify
where
systematic
estimation
biases
may
occur
and
attempt
to
correct
for
them.
A
study
by
Harrington,
Morgenstern,
and
Nelson
(
1999)
evaluated
the
accuracy
of
EPA
and
OSHA
estimates
of
25
ex
ante
regulatory
cost
estimates
relative
to
ex
post
studies
of
actual
costs,
and
concluded
that
initial
cost
estimates
by
EPA
tend
to
overstate
costs.
2
Based
on
the
Harrington
et
al.
analysis
and
on
other
studies
of
regulatory
costs,
in
our
own
prior
work
we
identified
three
key
factors
inherent
in
the
production
and
adoption
of
compliance
measures
that
suggest
that
direct
compliance
cost
projections
may
more
often
be
overestimated
than
underestimated.
These
factors
include:
"
learning
by
doing,"
or
learning
curve
impacts,
which
refer
to
efficiency
gains
related
to
cumulative
production;
innovation
and
3
Cynthia
Manson
and
Jim
Neumann,
"
Assessing
Prospective
Estimates
of
the
Costs
of
Clean
Air
Act
Compliance:
Are
Costs
Overestimated?"
memorandum
prepared
by
Industrial
Economics,
Incorporated
for
U.
S.
Environmental
Protection
Agency,
Office
of
Policy
Analysis
and
Review,
June
29,
2001.
Note
that
Harrington
et
al.
(
1999)
identify
several
reasons
for
cost
overestimation
errors,
including
a
built­
in
conservative
bias
and
inaccuracies
in
estimating
the
size
of
the
affected
universe.
These
factors
also
affect
the
specification
of
benefits.
Without
further
information
on
case­
specific
marginal
benefit
and
cost
curves,
it
is
difficult
to
assess
overall
effects
of
such
mis­
estimations
on
benefit/
cost
ratios.
IEc
therefore
focused
on
factors
that
affect
only
costs.

4
In
reality,
"
learning"
and
"
innovation"
lie
on
a
continuum.
However,
as
we
discuss
below,
incremental
innovations
which
do
not
require
new
capital
investments
(
i.
e.,
new
processes)
are
generally
reflected
in
progress
curves.
In
contrast,
more
dramatic
changes
in
processes,
or
significant
shifts
in
energy
or
raw
materials
use,
are
less
predictable
and
are
not
generally
addressed.

5
Energy
Information
Administration,
Assumptions
for
the
Annual
Energy
Outlook
1997,
U.
S.
Department
of
Energy,
Washington,
DC,
January
1997,
pages
58
and
63­
64.

6
Other
efforts
include
rules
for
Tier
2
highway
vehicles
(
65
FR
6698,
February
10,
2000,
marine
diesel
engines
(
64
FR
73300,
December
29,
1999)
nonroad
diesel
engines
(
63
FR
56968,
October
23,
1998)
and
highway
diesel
engines
(
62
FR
54694,
October
21,
1997).

2
technological
change;
and
cost­
reducing
features
of
regulatory
design.
3
Most
existing
Agency
cost
estimates
do
not
make
adjustments
to
account
for
these
factors.

This
report
suggests
an
approach
to
incorporate
learning
curve
impacts
into
relevant
regulatory
cost
estimates.
We
focus
specifically
on
the
treatment
of
learning
curves
(
and
the
more
comprehensive
progress
curves),
rather
than
the
potentially
more
dramatic
but
less
predictable
"
innovations"
which
involve
significant
changes
in
process.
Learning
and
progress
curves,
by
contrast,
measure
incremental
increases
in
efficiency
that
occur
as
cumulative
production
of
a
product
increases.
4
These
improvements
result
from
a
number
of
factors,
including
increased
experience
among
employees
and
refinements
to
the
manufacturing
process
and
products.
Learning
and
progress
curves
have
been
quantified
in
empirical
data
for
over
50
years,
are
widely
used
in
engineering
and
operations
research
literature,
and
have
recently
been
incorporated
into
forecasts
for
emerging
technologies
at
the
Energy
Information
Administration.
5
In
addition,
a
since
1997
several
EPA
regulatory
efforts
have
incorporated
progress
curve
impacts
into
future
cost
estimates.
These
efforts
include
analyses
supporting
the
Tier
2
tailpipe
regulations
(
USEPA
1999),
the
Heavy
Duty
Diesel
rule
(
USEPA
2000a),
and
the
Phase
2
Final
Rule
on
Handheld
Spark­
Ignition
Engines
(
USEPA
2000b).
6
This
report
first
discusses
the
theoretical
basis
for
identifying
and
estimating
learning
and
progress
curve
impacts
in
the
regulatory
context,
and
then
documents
several
cases
where
progress
curves
have
affected
estimates
of
compliance
costs
associated
with
Clean
Air
Act
regulations.
Finally,
we
summarize
the
potential
implications
and
practical
considerations
involved
in
incorporating
progress
curve
impacts
into
EPA
compliance
cost
estimates.
7
Belkouai,
(
1986),
p.
6.
See
also
Dutton
and
Thomas
(
1986);
Greening,
et.
al.
(
2001);
McDonald
and
Schrattenholzer
(
2001);
Montgomery
and
Day
(
1985).

3
THEORETICAL
BASIS
FOR
ESTIMATING
LEARNING
AND
PROGRESS
CURVES
Managers
have
long
observed
that
per­
unit
production
costs
decrease
as
cumulative
production
increases.
Economists
have
suggested
that
the
decrease
results
in
part
from
"
learning"
within
the
producing
organization
as
workers
become
more
experienced.
In
1936,
T.
P.
Wright
published
the
first
study
to
quantify
this
trend.
He
observed
that,
on
average,
as
cumulative
output
doubled
in
the
aircraft
production
industry,
labor
and
input
requirements
decreased
to
approximately
eighty
percent
of
their
former
cost.

According
to
the
equation
first
presented
by
Wright,
costs
will
decrease
by
a
constant
percentage
(
the
P
value)
with
every
doubling
of
cumulative
production,
reflecting
a
log­
linear
function:

A
1
=
A
0

XN
Where:

A
=
Cost
of
Production
X
=
Index
of
cumulative
production
N
=
log
P
/
log
2
=
learning
index,
where
P
is
the
rate
of
cost
reduction
as
production
increases
In
other
words,
Wright's
"
80
percent
rule"
(
i.
e.,
a
P
value
of
0.8,
or
80
percent)
is
associated
with
a
20
percent
(
i.
e.,
100
­
80)
reduction
in
costs.
The
80
percent
rule
predicts
that
a
product
which
costs
$
50
to
produce
when
the
100th
unit
is
in
production
will
cost
$
40
to
produce
when
the
200th
unit
is
produced.

Since
the
1930s,
researchers
have
documented
empirical
cost
trends
that
fit
this
log­
linear
function
in
a
wide
range
of
industries,
including
the
pollution
abatement
technology
industry.
While
other
functions
have
been
developed
to
estimate
learning
patterns
in
specific
circumstances,
practitioners
in
industry
generally
agree
that,
in
the
words
of
Ahmed
Belkaoui,
"
the
log­
linear
model
has
been
and
still
is,
by
far,
the
most
useful
model."
7
Empirical
studies
performed
over
the
past
half­
century
have
also
documented
that
a
P
value
of
approximately
80
percent
appears
to
occur
across
a
wide
range
of
industries.
As
a
result,
the
"
80
percent
rule"
is
commonly
used
in
the
development
of
future
cost
estimates.
EPA
has
recently
undertaken
a
limited
application
of
the
80
percent
rule
as
part
of
its
compliance
cost
estimates
for
activities
associated
with
the
Tier
2
regulations
(
1999),
the
Heavy
Duty
Engine/
Diesel
Fuel
rule
(
2000),
the
Phase
2
Final
Rule
on
Handheld
Spark­
Ignition
Engines
(
2000),
and
other
recent
efforts.
8
As
a
result,
analysts
argued
that
the
increasing
mechanization
of
production
would
lead
to
a
slower
learning
rate
since
"
machines
cannot
`
learn'
to
run
any
faster"
[
Belkaoui,
p.
5(
1986)].

9
The
study
found
a
median
P
value
of
77.5
percent,
with
over
three
quarters
of
the
estimated
P
values
between
70
and
90
percent
[
Montgomery
and
Day,
p.
4
(
1985)].

10
The
authors
distinguish
between
progress
curves
and
experience
curves
on
the
basis
of
the
data
used
to
establish
the
curve.
Progress
curves
rely
on
cost
information,
while
experience
curves
use
price
data.
Cost
data
(
and
progress
curves)
have
the
advantage
of
clarity;
prices
often
reflect
by
market
strategy
and
may
not
follow
reductions
in
material
or
labor
costs.

4
The
80
percent
rule
is
well­
documented
and
widely
accepted,
but
several
issues
surrounding
its
definition
and
application
should
be
considered
by
EPA
in
developing
an
approach
that
will
support
a
broader
application
of
learning
curve
theory.
Three
key
theoretical
issues
are:
identifying
an
appropriate
definition
of
"
learning;"
addressing
variability
among
empirical
estimates
of
learning;
and
considering
other
forces
(
such
as
institutional
"
forgetting")
that
may
affect
the
curve.
Below
we
discuss
the
importance
of
these
issues
in
a
regulatory
compliance
setting.
In
addition,
we
outline
several
technical
issues
that
affect
the
accurate
assessment
of
learning
and
progress;
these
include
the
stage
of
product
development,
the
costs
affected
(
e.
g.,
fixed
or
variable),
the
existence
of
multiple
progress
curves,
and
the
extent
to
which
innovations
and
improved
efficiencies
such
as
energy
efficiency
will
affect
compliance
behaviors.

Learning,
Experience,
and
Progress
Curves
Early
studies
of
learning
curves
focused
mainly
on
"
learning
by
doing"
or
"
labor
learning,"
or
on
the
concept
that
employees
learn
to
do
their
jobs
more
effectively
and
at
lower
cost
with
increasing
experience.
8
During
the
1960s,
however
The
Boston
Consulting
Group
(
BCG)
firmly
established
the
presence
of
an
"
experience
curve"
which
expanded
the
definition
of
"
learning"
to
include
changes
in
"
all
costs
of
every
kind
required
to
deliver
the
product
to
the
ultimate
user....
There
is
no
question
that
R&
D,
sales
expenses,
advertising,
overhead,
and
everything
else
is
included"
(
IEA,
p.
26).
BCG's
definition
of
learning
was
broader
than
previous
analyses,
but
an
18­
product
experience
curve
cost
analysis
using
Wright's
log­
linear
formula
revealed
that
experience
curves
were
almost
identical
to
the
more
narrowly
defined
learning
curve,
with
a
median
P
value
of
77.5
percent
[
Wooley
(
1972)].
9
Furthermore,
in
1984
Dutton
and
Thomas
argued
that
cost
trends
reflect
more
than
just
labor
learning:
their
"
progress
curve"
combines
"
changes
in
materials
inputs,
process
or
product
technologies,
or
managerial
technologies"
with
labor
learning
(
p.
235).
Again,
despite
the
shift
to
more
general
causes
for
cost
reductions,
the
Dutton
and
Thomas
survey
of
108
samples
provided
a
"
basis
for
the
widely
publicized
80
percent
progress
curve"
(
p.
237).
10
While
there
is
no
clear
explanation
for
the
unexpected
consistency
of
results
across
different
definitions
of
learning,
researchers
have
postulated
that
Wright's
learning
curves
may
have
captured
more
than
labor
learning.

Recent
analyses
tend
to
define
costs
comprehensively.
In
addition
to
labor
learning,
recent
studies
identify
management
techniques,
technological
innovation,
economies
of
scope,
and
11
Economies
of
scope
present
a
different
challenge
than
economies
of
scale
because
they
cannot
be
attributed
to
a
single
technology,
and
are
therefore
very
difficult
to
isolate.
However,
in
a
regulatory
analysis
it
is
reasonable
to
assume
that
these
impacts
will
occur
for
any
technology
in
a
proportion
equal
to
that
reflected
in
industry
P
values.

12
For
example,
a
firm
with
consistent
annual
production
of
100
scrubbers
will
experience
learning
each
year,
but
will
only
see
cost
changes
related
to
economies
of
scale
in
the
year
that
production
is
increased.

5
economies
of
scale
as
components
of
the
progress
curve
(
Joskow
and
Rose,
pp.
7­
8;
Montgomery
and
Day,
pp.
6­
7;
McDonald
and
Schrattenholzer,
p.
260).
In
a
regulatory
setting,
the
progress
curve
has
the
advantage
of
providing
a
single
estimate
for
a
number
of
forces
that
are
known
to
affect
costs
but
are
difficult
to
isolate.
However,
it
is
important
to
consider
the
relationship
between
"
progress"
and
two
other
forces
which
contribute
to
cost
decreases:
economies
of
scale
and
innovation.

Economies
of
Scale
Like
learning,
economies
of
scale
(
and
scope)
contribute
to
cost
reductions,
but
predicting
and
quantifying
their
impacts
is
difficult,
and
isolating
them
from
other
components
of
"
learning"
is
daunting.
11
As
a
result,
many
progress
curves
simply
include
economies
of
scale
and
scope
as
part
of
a
cumulative
effect
on
per
unit
cost
production.
McDonald
and
Schrattenholzer
explain,
"
Given
the
data
that
are
available,
model
inputs
in
which
learning
and
scale
economies
are
lumped
into
a
single
estimated
progress
rate
may
be
simpler,
as
reliable,
and
therefore
more
useful
than
efforts
to
extract
the
two
separate
effects
from
the
empirical
data"
(
2001).
The
advantage
to
policy
makers
of
a
single,
reliable
method
for
assessing
several
forces
is
obvious:
because
regulatory
costs
focus
on
total
compliance­
related
costs,
it
is
seldom
necessary
to
estimate
factors
such
as
economies
of
scale
in
isolation.

However,
authors
such
as
Argote
and
Epple
(
1990)
caution
that
studies
that
do
not
control
for
economies
of
scale
can
over­
estimate
the
effects
of
"
learning"
by
incorporating
one­
time
gains
due
to
scale
and
scope.
12
While
McDonald
and
Schrattenholzer
may
be
correct
that
controlling
for
economies
of
scale
is
difficult
with
available
data,
analysts
should
use
caution
in
projecting
aggressive
estimates
of
future
progress
for
technologies
that
are
not
likely
to
experience
changes
in
scale
or
reflect
changes
in
scope.

Innovation
"
Innovation"
represents
a
broad
range
of
process
and
input
changes,
from
minor
efficiency
improvements
to
dramatic
changes
in
product.
Progress
curves
tend
to
reflect
some
technological
innovation,
including
minor
process
adjustments
that
speed
production,
improve
fuel
efficiency,
or
reduce
necessary
inputs.
These
changes
tend
to
be
incremental
and
regular,
and
lead
to
cost
reductions
consistent
with
the
progress
curve
[
Jacobsen,
p.
150
(
2001)].
Major
innovations,
13
According
to
IEA,
technology
structural
changes
redraw
the
log­
linear
graph
in
the
shape
of
a
double
knee,
reflecting
a
"
transition
period"
where
the
slope
of
the
progress
curve
(
the
P
value)
is
significantly
higher.

14
In
one
area
 
energy
technology
 
researchers
are
attempting
to
predict
significant
technological
change,
particularly
within
the
long
time
frames
relevant
to
climate
change.
Modeling
and
research
efforts
have
focused
on
predicting
increases
in
energy
efficiency
and
emergence
of
new
technologies,
based
in
part
on
known
energy
price
elasticities.
Future
research
in
this
area
may
provide
information
that
could
be
used
to
further
refine
regulatory
cost
analysis.

6
however,
can
significantly
alter
the
cost
of
products
or
services
and
effectively
"
shift
the
curve."
For
instance,
Henry
Ford's
production
line
completely
changed
the
cost
of
producing
automobiles,
and
Dow
Chemical's
development
of
the
HCFC
caused
a
dramatic
drop
in
the
cost
of
producing
CFC­
free
refrigerators.
The
International
Energy
Agency
refers
to
these
changes
as
technology
structural
changes.
13
These
changes
are
difficult
to
foresee
and
do
not
conform
to
any
pattern
of
gradual
cost
reductions.
14
While
these
shifts
are
unpredictable
and
are
not
factored
into
future
progress
curve
projections,
the
resulting
cost
reductions
provide
what
IEA
labels
"
a
beneficial
surprise
for
the
policy­
maker"
(
IEA,
92).
The
record
of
Clean
Air
Act
implementation
over
the
past
30
years
contains
a
number
of
examples
of
technological
innovations
(
e.
g.,
HCFCs)
that
have
increased
the
cost­
effectiveness
of
air
pollution
control.

Progress
Curves
in
Regulatory
Analysis
While
the
literature
reflects
many
definitions
of
"
learning"
and
"
progress,"
the
broader
progress
curves
identified
by
BCG,
Dutton
and
Thomas
(
1984)
and
more
recent
studies
are
likely
to
be
the
most
relevant
for
regulatory
analysis,
because
they
capture
a
range
of
cost
impacts
that
are
relevant
to
the
projection
of
accurate
regulatory
costs.
In
addition
to
the
economies
of
scale
and
scope
and
the
incremental
innovations
discussed
above,
progress
curves
are
likely
to
reflect
concurrent
adaptations
such
as:


Market
Adaptations:
Continuing
changes
in
the
structure
of
markets
and
practices
related
to
financing
and
trade
frequently
contribute
to
changes
in
costs.
Some
of
these
may
be
adverse
(
i.
e.,
if
raw
materials
become
scarce)
but
controlled
adaptations
have
tended
to
focus
on
increased
efficiency.


Reduced
Transaction
Costs:
Similar
to
market
shifts,
transaction
cost
changes
may
contribute
to
progress
as
costs
associated
with
adopting
technologies
are
reduced.
Regulations
and
voluntary
programs
(
e.
g.,
EPA's
Energy
Star)
can
be
constructed
to
reduce
such
costs
and
thereby
enhance
progress.
15
Note
because
empirically
based
progress
curves
represent
"
net"
changes
in
cost,
use
of
these
curves
also
accounts
for
the
"
costs"
that
are
often
associated
with
learning,
such
as
the
investment
in
management
expertise,
to
the
extent
that
these
costs
are
reflected
in
the
cost
data
used
to
develop
the
progress
curves.

16
The
authors
write:
"
In
any
given
industry,
firms'progress
functions,
as
well
as
progress
rates,
vary
widely."
Furthermore,
"
steepness
of
cost
decline
(
the
P
value)
frequently
is
controllable
via
creative
managerial
efforts."

7

Changes
in
Energy
Efficiency:
Energy
efficiency
improvements
(
either
by
utilities,
manufacturers,
or
regulated
entities)
can
have
an
important
impact
on
costs,
often
by
both
"
saving
costs"
and
encouraging
productivity
increases.

While
these
forces
are
not
specifically
"
learning"
(
i.
e.,
labor
learning),
they
are
inevitably
reflected
to
some
degree
in
the
more
broadly
defined
and
measured
progress
curves
and
are,
like
economies
of
scale,
difficult
to
isolate
from
labor
learning.
Furthermore,
the
empirical
data
supporting
the
existence
of
progress
curves
are
likely
to
reflect
all
of
these
forces
to
some
degree.
The
stability
of
results
from
quantitatively
estimating
progress
curves
from
studies
performed
over
the
last
60
years
suggests
that
over
time,
these
changes
and
adaptations
are
relatively
consistent
in
scope.
Therefore,
the
careful
application
of
a
comprehensive
"
progress
factor"
in
estimating
the
impacts
of
learning
on
regulatory
compliance
costs
appears
to
be
a
more
efficient
initial
approach
than
attempting
to
develop
separate
estimates
for
the
impacts
of
multiple
factors,
except
in
cases
where
specific
data
suggest
that
exceptional
events
or
forces
are
affecting
costs.
15
Variance
Among
Progress
Rates
Although
multi­
sector
studies
tend
to
reveal
progress
rates
with
P
values
in
the
80
percent
range,
most
progress
studies
also
reveal
significant
variance
in
P­
values
both
across
industries
and
for
firms
within
the
same
industry.
For
instance,
McDonald
and
Schrattenholzer's
survey
of
learning
in
the
energy
technology
industry
identified
P
values
ranging
from
55
percent
to
99
percent
(
i.
e.,
doubling
of
cumulative
reduction
could
result
in
cost
reduction
that
ranged
from
45
percent
to
only
1
percent,
respectively),
with
a
median
value
of
83
to
84
percent
[
McDonald
and
Schrattenholzer
(
2001)].
This
within­
sector
variance
is
notably
higher
than
the
variance
across
sectors.

Researchers
have
interpreted
this
wide
range
of
progress
rates
at
the
firm
level
to
indicate
that
production
experience
creates
the
opportunity
for
cost
reduction,
but
not
all
organizations
in
an
industry
exploit
this
potential
equally.
It
is
generally
believed
that
a
company's
progress
rate,
or
P­
value,
is
a
dependent
variable
which
changes
relative
to
the
skill,
creativity,
and
intelligence
of
management
[
Dutton
and
Thomas
(
1984)].
16
In
contrast,
a
competitive
industry
should
ultimately
reflect
the
success
of
its
competitive
firms
and
will
reveal
broad
progress
trends,
as
"
slower"
players
exit
and
more
efficient
firms
continue
to
compete.
17
The
general
application
of
an
industry­
level
progress
curve
does
not
address
the
need
for
a
separate
analysis
of
small
entity
impacts;
if
dramatic
progress
takes
place
only
at
large
companies
in
a
specific
industry,
analysis
required
by
the
Regulatory
Flexibility
Act
should
address
this.

18
The
authors
examined
existing
studies
of
manufacturing
processes
in
industries
including
electronics,
machine
tools,
and
manufacture
of
automobiles,
aircraft,
steel,
paper,
and
apparel.

19
In
the
absence
of
specific
industry
data,
use
of
the
80
percent
value
assumes
(
1)
that
firms
with
rapid
progress
rates
will
not
dominate
the
market
over
time,
leading
to
a
lower
industry­
wide
P
values
and
(
2)
that
the
market
remains
competitive
and
barriers
to
entry
are
not
unreasonable,
since
monopolies
with
low
progress
rates
would
lead
to
less
cost
reduction
across
an
entire
industry.

8
Variance
among
firm­
level
progress
curves
is
not
generally
problematic
in
a
regulatory
setting,
because
EPA
cost
estimates
address
industry­
wide
impacts.
17
Therefore,
defensible
estimates
of
industry­
level
progress
are
more
important
in
predicting
compliance
costs.
While
studies
reveal
that
progress
rates
range
widely
among
firms
and
industries,
these
surveys
also
consistently
show
a
bell­
shaped
distribution
of
P­
values
with
both
the
median
and
mode
P
value
in
the
75
to
85
percent
range;
see
Exhibit
1
for
one
such
distribution,
from
Dutton
and
Thomas's
work.
Dutton
and
Thomas's
survey
of
108
manufactured
products
in
multiple
industries
yielded
both
a
median
and
mode
P
value
of
81
percent
(
i.
e.,
a
19
percent
cost
reduction
for
each
doubling
of
production).
18
McDonald
and
Schrattenholzer
calculated
a
median
P
value
of
83
to
84
percent,
and
the
75
to
85
percent
range
was
most
common.
Montgomery
and
Day
report
that
Wooley's
survey
found
a
median
P
value
of
77.5,
with
over
three
quarters
of
the
estimated
P
values
between
70
and
90
percent.

In
some
cases
more
specific
industry
information
may
suggest
a
different
conclusion.
In
Greening
et
al.'
s
survey
of
twelve
pollution
abatement
technologies,
six
of
the
12
preliminary
P
values
calculated
fell
between
80
and
90
percent;
the
median
for
the
group
of
12
was
90
percent
(
i.
e.,
a
10
percent
reduction
in
costs
as
cumulative
production
doubled).
These
distributions
suggest
that
future
industry­
wide
production
cost
forecasts
should
generally
reflect
progress
rates
of
between
75
and
85
percent,
unless
specific
industry
studies
provide
more
exact
estimates.
19
Note
on
Experience
Depreciation
and
Organizational
Forgetting
While
learning
is
nearly
ubiquitous,
some
recent
studies
have
focused
on
apparent
exceptions
where
production
costs
increase
(
or
fail
to
decrease)
over
time.
Several
authors
have
established
the
possibility
of
both
"
experience
depreciation"
and
"
organizational
forgetting"
[
McDonald
and
Schrattenholzer
(
2000);
Argote
and
Epple
(
1990);
Benkard
(
2000)].
These
authors
argue
that
cumulative
production
may
be
affected
by
the
departure
of
trained
staff
and
imperfect
transfer
of
knowledge
from
old
employees
to
new
employees,
particularly
when
large
layoffs
are
coupled
with
slowed
production.
McDonald
and
Schrattenholzer
(
2000)
assert
that
"
experience
gained
from
units
built
last
year
results
in
greater
current
cost
reductions
than
experience
from
10
years
ago."
One
example
of
a
technology
that
could
potentially
have
suffered
from
"
forgetting"
is
photovoltaic
solar
technology,
which
languished
after
the
era
of
high
oil
prices
in
the
1970s.
However,
because
regulations
generally
tend
to
require
increases
in
production
and
more
rapid
adoption
of
technologies,
"
forgetting"
is
unlikely
to
be
a
significant
issue
in
regulatory
analysis.
9
Exhibit
1
20
Note
that
"
learning
by
using"
is
in
fact
a
portion
of
the
general
process
of
learning
at
the
regulated
entity,
and
could
conceivably
be
reflected
in
the
products
of
this
company.
However,
only
the
portion
of
production
costs
associated
with
compliance
are
relevant
to
the
regulation.

10
Issues
Related
to
Implementation
of
Progress
Curves
in
a
Policy
Setting
The
theoretical
and
empirical
basis
is
strong
for
incorporating
progress
curve
impacts
into
estimates
of
future
compliance
costs.
However,
a
number
of
issues
related
to
progress
curve
"
behavior"
present
key
challenges
to
the
effective
implementation
of
progress
curves.
EPA
has
addressed
several
of
these
issues
during
implementation
of
progress
curve
estimates;
in
particular
we
examined
the
Agency's
approach
in
developing
the
Tier
II
and
Heavy
Duty
Engine/
Diesel
Fuel
rules.
Below
we
provide
a
summary
of
the
issues;
Exhibit
2
presents
a
summary
of
EPA's
approach
in
the
Tier
II
regulatory
impact
analysis.


Product
and
Industry
Stage
of
Development:
Because
the
progress
curve
is
dependent
on
cumulative
production,
an
accurate
projection
of
progress
relies
on
an
accurate
estimate
of
the
"
life
stage"
of
the
technology
in
question.
That
is,
if
a
regulation
requires
widespread
application
of
a
new
or
emerging
technology
(
as
in
the
case
of
catalytic
converters)
to
meet
compliance
requirements,
then
progress
curve
impacts
on
cost
may
be
dramatic
as
cumulative
production
doubles
rapidly.
In
contrast,
if
a
regulation
requires
broader
adoption
of
an
existing
technology
with
a
considerable
cumulative
production
history
(
e.
g.,
a
MACT
standard
that
requires
smaller
facilities
to
control
emissions
from
conventional
boilers),
then
changes
in
cost
due
to
progress
may
be
subtle.
For
this
reason,
it
is
important
to
identify
the
current
cumulative
installed
base
of
a
pollution
control
technology,
as
well
as
predicting
accurately
the
number
of
regulated
entities
that
will
adopt
the
technology
(
and
therefore
contribute
to
cumulative
production)
and
the
"
doubling
time."


Multiple
Progress
Curves
in
the
Regulatory
Compliance
Context:
In
the
context
of
regulatory
compliance,
more
than
one
progress
curve
may
be
reflected
in
costs.
For
example,
manufacturing
costs
for
flue
gas
desulfurization
units
have
decreased
over
time,
reflecting
progress
and
reducing
capital
costs
to
the
regulated
community
(
assuming
that
market
prices
also
decrease).
In
addition,
regulated
entities
will
likely
improve
the
efficiency
of
their
operations
and
maintenance
related
to
the
units
and
affected
processes.
These
increases
in
efficiency
represent
a
separate
"
learning
by
using"
impact
that
would
result
in
further
cost
reductions
over
time.
20
Because
regulators
are
generally
concerned
with
the
end
user
of
a
new
technology,
consideration
of
the
additive
impacts
of
multiple
progress
opportunities
may
be
warranted.
11
The
extent
to
which
the
Agency
should
consider
multiple
progress
curves
will
depend
on
the
construct
of
the
regulation,
and
the
"
age"
of
the
affected
technologies.
If
regulations
will
likely
result
in
significant
capital
investments
and
process
retooling
to
incorporate
innovative
technologies,
then
it
may
be
reasonable
to
examine
learning
in
several
areas.
If,
however,
a
program
emphasizes
readily
available
or
alternatively,
"
low
maintenance"
technology,
then
progress
will
likely
be
limited
to
only
one
process.


Identifying
Costs
Affected
by
Learning:
Judicious
application
of
progress
curves
is
critical
in
developing
defensible
cost
estimates;
even
the
most
expansive
definitions
of
progress
curves
do
not
include
all
the
costs
borne
by
industry.
For
example,
one­
time
installation
costs
may
not
be
sensitive
to
learning
unless
they
are
performed
by
a
third
party.
Similarly,
some
monitoring­
related
costs
and
other
requirements
may
employ
a
very
consistent
approach
and
represent
fixed
costs
that
are
relatively
impervious
to
learning.
Although
the
literature
reveals
several
approaches
to
estimating
progress,
including
the
use
of
price,
total
costs,
and
marginal
costs,
an
approach
that
limits
application
of
the
curve
to
marginal
costs
is
the
most
defensible.


The
Role
of
Energy
Efficiency
in
Progress
and
Compliance:
Economists
have
recently
focused
considerable
effort
on
the
accurate
prediction
of
energy
prices
and
the
cost
implications
of
efficiency
as
innovative
technologies
emerge.
This
issue
is
particularly
crucial
in
the
analysis
of
regulations
to
counter
climate
change,
which
involves
predictions
over
long
time
horizons.
However,
energy
prices
and
efforts
to
improve
efficiency
and
conservation
also
have
short
term
implications
for
costs,
and
it
is
important
to
consider
the
interplay
between
these
activities
and
progress
curves.

Specifically,
regulations
that
increase
the
production
cost
may
encourage
regulated
entities
to
adapt
process
innovations
or
efficiency
measures
(
e.
g.,
reducing
energy
use
to
reduce
NO
x
outputs
to
targeted
limits).
These
measures
represent
low­
cost
compliance
approaches
that
provide
considerable
benefits
(
and
may
reflect
effective
regulatory
design).
However,
the
impact
on
progress
curve
estimates
may
be
to
slow
cost
decreases
by
reducing
the
demand
for
units
(
and
therefore
increasing
"
doubling
time").
While
this
issue
is
essentially
a
challenge
in
estimating
regulatory
response
and
not
in
developing
accurate
progress
curve
estimates,
it
is
important
to
consider
compliance
options
and
responses
in
developing
progress
curve
doubling
times
and
estimates.
12
Exhibit
2.
EPA
Progress
Curve
Estimation
in
the
Tier
II
Regulatory
Impact
Analysis
EPA
has
undertaken
a
limited
application
of
the
80
percent
rule
in
recent
regulatory
efforts,
including
the
Tier
II
sulfur
regulations
(
1999),
the
Heavy
Duty
Diesel
rule
(
2000),
and
the
Phase
2
Final
Rule
on
Handheld
Spark­
Ignition
Engines
(
2000).
In
these
efforts,
the
Agency
adopted
a
general
approach
that
employed
the
multi­
industry
average
"
80
percent
rule,"
projecting
that
costs
would
decrease
by
20
percent
as
cumulative
production
doubled.
Recognizing
that
the
80
percent
rule
is
a
general
average
that
may
over­
or
under­
state
progress
rates
in
specific
industries,
EPA
also
adopted
a
number
of
conservative
assumptions
in
applying
the
curve,
in
order
to
assure
that
progress
impacts
were
not
overstated
(
i.
e.,
that
costs
would
not
be
underestimated).
Assumptions
included:


Some
processes
were
not
included
in
estimates
of
progress:
EPA
did
not
apply
progress­
related
cost
decreases
to
technologies
(
e.
g.,
evaporative
systems)
which
were
already
widely
available
and
therefore
unlikely
to
be
considerably
affected
by
incremental
decreases
in
learning.
The
80
percent
rule
was
applied
only
to
newer
technologies
that
were
more
likely
to
show
considerable
efficiency
gains
due
to
progress.


"
Learning"
would
occur
after
the
second
year
of
production:
EPA
assumed
that
one
year
of
production
(
in
which
no
learning
would
take
place)
formed
the
base
"
unit"
of
cumulative
production.
In
other
words,
if
100
units
were
produced
in
the
first
year,
then
it
was
assumed
that
100
units
would
be
produced
in
the
second
year
and
the
expected
20
percent
drop
in
costs
would
occur
at
the
beginning
of
the
third
year.
This
assumption
establishes
a
relatively
high
cumulative
production
experience
as
the
basis
for
doubling,
which
in
turn
reduces
the
rate
of
cost
reduction
(
i.
e.,
because
"
doubling"
cumulative
production
requires
more
effort).
In
reality,
rapid
production
during
the
first
(
or
second)
year
could
conceivably
lead
to
considerable
cost
reductions
earlier
in
the
process.


Certain
process
costs
were
not
subject
to
progress:
EPA
did
not
apply
progress­
based
cost
reductions
to
fixed
costs
or
to
certain
variable
costs.
For
example,
in
the
Tier
II
analysis
catalyst
precious
metal
costs
were
omitted
from
consideration
due
to
high
uncertainty
in
metals
price
forecasts.

In
addition,
specifically
for
the
Tier
II
sulfur
regulations,
EPA
limited
its
estimates
of
progress
curve
impacts
to
one
"
learning"
cycle.
The
Agency
did
not
fully
apply
Wrights
log­
linear
function,
which
would
continue
to
reduce
future
production
costs
indefinitely.
Instead
EPA
calculated
only
a
single
reduction
in
costs
of
20
percent
starting
in
the
third
year
of
the
rule,
consistent
with
the
"
first
cycle"
of
the
log­
linear
function.

IMPACTS
OF
PROGRESS
ON
AIR
POLLUTION
CONTROL
TECHNOLOGY
COSTS:
EXAMPLES
The
literature
establishes
that
the
impacts
of
progress
on
costs
are
essentially
ubiquitous,
occurring
across
a
broad
range
of
products
and
industries,
including
a
number
of
pollution
abatement
technologies
(
Greening
et
al.);
progress
curves
have
proven
to
be
a
reasonable
way
to
model
these
impacts.
Below
we
present
three
brief
discussions
of
the
existence
of
progress
curves
in
specific
air
pollution
control
technologies.
The
first
discussion
examines
existing
estimates
of
progess
curves
specific
to
scrubber
technology.
We
then
discuss
the
issues
related
to
developing
progress
functions
for
emerging
NO
x
control
technologies.
Finally,
we
examine
the
broader
support
for
"
progress"
that
may
be
reflected
in
historical
trends
in
the
Vatavuk
cost
indices
for
various
pollution
control
technologies.

Example
1:
Sulfur
Dioxide
Control
Technology
21
See
Burtraw,
Dallas.
"
Cost
Savings
Sans
Allowance
Trades?
Evaluating
the
SO2
Emission
Trading
Program
to
Date",
Resources
for
the
Future,
February
1996,
and
Institute
of
Clean
Air
Companies,
Scrubber
Myths
&
Realities,
White
Paper,
May
1995.

22
Laitner,
S.
"
Analyzing
changes
in
Scrubber
Costs
over
Time"
unpublished
memorandum,
U.
S.
Environmental
Protection
Agency,
Office
of
Atmospheric
Programs.

23
The
large
difference
between
this
progress
rate
and
the
dramatic
progress
rate
of
47
percent
calculated
by
Greening,
et
al.
may
be
a
result
of
different
components
that
are
included
in
13
Flue
gas
desulfurization
units
(
i.
e.,
"
FGD
technology,"
or
"
scrubbers")
present
a
clear
example
of
declining
costs
associated
with
compliance.
Several
research
efforts
have
sought
to
explain
the
dramatic
decreases
in
the
costs
related
to
sulfur
control
since
the
late
1980s.
Authors
such
as
Dallas
Bertraw
(
1996)
and
reports
by
the
Institute
of
Clean
Air
Companies
(
ICAC)
(
1995)
have
documented
the
decreases
in
cost
and
have
attributed
a
portion
of
costs
to
learning
and
incremental
process
innovations.
21
In
addition,
more
recent
work
by
Taylor,
et
al.(
2000)
and
Greening,
et
al.(
2001)
have
identified
specific
progress
ratios
for
scrubber
technologies.
These
articles
identify
P
values
for
scrubbers
of
83
and
88
percent,
respectively.
Finally,
separate
unpublished
estimates
by
Laitner
using
different
assumptions
identify
a
progress
curve
for
capital
costs
of
85
percent.
22
Scrubber
technology
presents
an
ideal
opportunity
for
examining
progress
curves
for
two
reasons:
the
technology
has
a
decade­
long
history
of
development
and
adoption,
and
the
U.
S.
Department
of
Energy
regularly
publishes
data
tracking
the
number
of
installations,
unit
costs,
and
removal
efficiency
of
scrubbers
over
time.
The
Taylor
et
al.,
Greening,
et
al.,
and
Laitner
research
each
employ
EIA
data;
the
different
estimates
of
progress
in
these
articles
arise
principally
from
differing
time
lines
and
differing
definitions
of
affected
costs.

Specifically,
the
two
estimates
developed
by
Laitner
use
data
from
Electric
Power
Annual
to
develop
a
progress
curve
based
on
total
national
cost
trends
in
both
capital
and
O&
M
costs
per
kilowatt
(
kW)
over
the
decade
from
1985
to
1995.
Cumulative
production
is
estimated
as
the
total
installed
capacity
of
megawatts
(
MW)
nationwide.
In
the
Greening
et
al.
study,
the
authors
estimate
a
progress
curve
of
88
percent
(
P
value
of
0.88).
Incorporated
into
this
estimate
are
a
relatively
conservative
progress
ratio
for
capital
costs
of
89
percent
and
a
dramatic
progress
ratio
for
operating
costs
of
47
percent
(
P
values
of
0.89
and
0.47,
respectively
C
the
47
percent
P
value
indicate
a
cost
decrease
of
over
50
percent
as
cumulative
production
doubles).
Laitner's
other
effort
produces
a
capital
cost­
related
progress
curve
of
85
percent
(
P
value
of
0.85),
based
on
a
more
detailed
regression
analysis
of
four
variables:
size
of
FGD
unit,
efficiency
of
removal,
percent
of
SO
2
removal,
and
cumulative
production.

The
research
by
Taylor
et
al.
examines
FGD
system
O&
M
costs
at
88
plants
that
have
reported
at
least
twelve
years
of
continuous
operations
data
to
EIA.
Using
the
standard
log­
linear
progress
function,
the
authors
calculate
an
83
percent
progress
ratio
on
FGD
labor
and
maintenance,
and
supervision
costs.
The
study
does
not
consider
changes
in
capital
costs
(
i.
e.,
retrofits
or
upgrades)
at
the
plants.
23
"
operating
costs"
in
the
EIA
data
(
e.
g.,
energy),
and
also
could
result
in
part
from
the
different
sample
facilities
and
time
frames
examined.

24
The
mean
and
mode
of
Dutton
and
Thomas'
study
is
a
P
value
of
81
percent;
the
standard
deviation
is
7.9
percent.

14
The
88
and
the
85
percent
progress
ratios
reported
in
Greening
et
al.
and
by
Laitner
suggest
that
progress
impacts
related
to
scrubber
capital
costs
are
considerable,
but
may
be
less
significant
than
implied
by
the
80
percent
"
rule
of
thumb."
However,
both
the
88
and
85
percent
values
are
within
one
standard
deviation
of
the
mean
(
P
value
of
81
percent)
of
the
distribution
of
progress
ratios
developed
by
Dutton
and
Thomas
(
1984).
24
An
examination
of
the
three
approaches,
however,
provides
some
useful
insights
for
policy­
makers
wishing
to
apply
progress
curves.


Both
the
sample
and
the
specific
focus
on
operating
costs
are
relatively
narrow
in
the
Taylor
et
al.
study;
the
progress
curve
identified
in
this
effort
is
effectively
limited
to
"
learning
by
using"
by
regulated
entities
and
does
not
consider
changes
in
capital
costs.
In
a
regulatory
analysis,
this
type
of
estimate
could
provide
a
useful
estimate
for
application
to
a
limited
range
of
operating
costs
associated
with
compliance.


The
Greening
et
al.
study
provides
a
"
unified"
estimate
of
progress
that
could
be
applied
more
broadly
to
both
capital
and
O&
M
costs
at
regulated
facilities.
Although
this
estimate
presents
a
more
conservative
progress
rate
than
the
Taylor
study,
its
potential
for
relatively
broad
application
would
increase
its
impact
on
cost
estimations.


Finally,
the
Laitner
research
suggests
that
an
independent
estimate
of
capital
costs
could
be
applied,
possibly
in
conjunction
with
the
Taylor
estimates
of
operations
cost
impacts.
Application
of
these
two
estimates
would
provide
a
more
aggressive
estimate
of
progress
than
the
Greening
estimates;
careful
discussion
of
the
range
of
potential
cost
impacts
and
a
sensitivity
analysis
could
address
these
discrepancies.

The
estimates
of
progress
curves
applicable
to
scrubber
capital
and
O&
M
costs
are
consistent
with
other
empirical
studies
of
learning
in
that
they
identify
clear
patterns
of
cost
reduction
associated
with
cumulative
production.
In
addition,
the
three
estimates,
though
they
focus
on
different
aspects
of
progress,
fall
between
83
and
89
percent,
well
within
the
range
of
values
that
has
been
established
in
the
literature.
As
a
result,
these
studies
support
the
validity
of
an
approach
that
incorporates
progress
curve
impacts
into
regulatory
cost
estimates.
Moreover,
while
technology­
specific
progress
ratios
clearly
allow
for
a
more
refined
discussion
of
costs,
the
general
consistency
between
"
scrubber
progress"
and
the
80
percent
"
rule
of
thumb"
is
also
notable.
This
consistency
appears
to
support
an
approach
that
adopts
the
80
percent
rule
as
an
initial
estimate
of
impacts
in
cases
where
technologyspecific
ratios
are
difficult
or
impossible
to
develop.
25
SCR
and
SNCR
are
"
fundamentally
similar
in
that
[
they]
both
use
an
ammonia­
containing
reagent
to
convert
the
NO
x
produced
in
the
boiler
to
nitrogen
and
water.
SNCR
accomplishes
this
at
higher
temperatures
(
1700
degrees
Fahrenheit
to
2000
degrees
Fahrenheit)
in
the
upper
furnace
region
of
the
boiler.
In
contrast,
SCR
operates
at
lower
temperatures
(
about
600
degrees
Fahrenheit)
by
using
a
catalyst
[
in
the
stack]
to
produce
the
desired
reaction
between
ammonia
and
NO
x.
In
practice,
these
differences
mean
that
SNCR
has
lower
capital
costs
and
limited
NO
x
reduction
capability
(
typically
30
to
40
percent
but
higher
in
some
cases).
SCR
is
more
capital
intensive
but
is
capable
of
achieving
much
greater
reductions
(
up
to
90
percent
and
higher)"
(
Amar,
2000;
Botsford,
2001).

15
Example
2:
NOx
Control
Technology
Oxides
of
nitrogen
(
NO
x),
like
sulfur
dioxide,
are
a
target
of
pollution
control
legislation
and
regulation
in
the
United
States
today.
Efforts
to
limit
NO
x
emissions
date
back
to
the
early
1970s,
but
prior
to
1990
NO
x
requirements
were
largely
limited
to
new
or
substantially
modified
sources.
Only
in
the
past
decade,
as
a
result
of
the
1990
Clean
Air
Act
Amendments'
particular
emphasis
on
NO
x
emissions
reductions,
have
American
regulations
pushed
industry
to
develop
new
pollution
abatement
technology.
As
a
result,
American
NO
x
control
technological
development
has
a
short
history,
and
it
presents
a
good
example
of
a
new
technology
that
may
reveal
a
measurable
progress
curve
(
Amar,
2000).

Initial
NO
x
Control
Development
NO
x
control
technology
differs
from
most
other
pollution
abatement
technology
in
that
its
initial
development
occurred
in
Germany
and
Japan,
not
the
United
States.
Prior
to
the
passage
of
the
1990
Clean
Air
Act
Amendments
in
the
United
States,
Germany
and
Japan
passed
"
technology
forcing"
NO
x
emissions
regulations
that
obligated
their
utility
industries
to
significantly
reduce
NO
x
emissions.
During
the
1980s
the
Germans
and
Japanese
developed
the
first
models
of
today's
preferred
NO
x
reduction
technologies,
including
selective
catalytic
reduction
(
SCR)
and
to
a
lesser
degree
selective
noncatalytic
reduction
(
SNCR).
25
These
technologies
today
are
considered
the
"
primary
post­
combustion
NO
x
control
options
commercially
available"
for
use
at
utility
plants
(
Amar,
2000).
By
1990,
German
and
Japanese
utilities
had
installed
about
200
SCR
units,
representing
about
40,000
MW
of
capacity.
SNCR
NO
x
control
units,
although
developed
after
SCR,
had
been
installed
at
a
smaller
but
significant
portion
of
European
utilities
(
Eskinazi,
1989;
Amar,
2000).

As
the
Europeans
and
Japanese
advanced
SCR
NO
x
control
technology
through
the
1980s,
they
were
able
to
increase
SCR's
effectiveness,
reduce
negative
externalities,
and
develop
a
number
of
operational
cost­
saving
strategies
(
e.
g.
leaving
space
in
the
reactor
for
future
catalyst
additions)
consistent
with
the
progress
principle.
By
the
end
of
the
decade,
the
SCR
technology
had
gone
from
experimental
to
the
utility
industry
standard
(
Amar,
2000).
Although
no
detailed
record
of
SCR
system
costs
in
Europe
in
the
1980s
has
been
published,
the
decreasing
cost
of
catalyst
is
well
documented.
According
to
Eskinazi
et
al.(
1989),
catalyst
costs
generally
dominate
SCR
system
costs,
so
trends
in
the
cost
of
the
catalyst
offer
a
good
proxy
for
the
cost
of
the
overall
SCR
system.
26
Catalyst
costs
represent
a
major
component
of
SCR
operating
costs,
so
any
change
in
these
costs
will
drive
the
SCR
operating
costs
in
a
similar
direction.
In
the
late
1980s,
increased
competition
in
the
European
catalyst
market
is
credited
with
driving
down
prices.
This
drop
in
price
is
likely
associated,
at
least
in
part,
with
progress
in
the
catalyst
production
industry
as
manufacturers
achieved
economies
of
scale
and
improved
production
processes.

27
Ammonia
`
slip'
refers
to
small,
steady
releases
of
ammonia
to
the
atmosphere
that
may
occur
in
SCR
and
SNCR
systems
during
operation.
Most
jurisdictions
in
the
United
States
have
limited
this
slip
to
5
ppm,
and
the
state
of
Massachusetts
has
limited
it
to
2
ppm.
These
regulations
have
forced
SCR
and
SNCR
manufacturers
in
the
U.
S.
to
redesign
the
technology
(
Botsford,
2001).

16
By
1990,
with
about
120
European
SCR
units
in
operation,
the
cost
of
catalyst
had
dropped
by
two
thirds
(
from
$
900
per
cubic
foot
to
$
300
per
cubic
foot).
26
(
Eskanazi
et
al.,
1989;
Amar,
2000)

U.
S.
NO
x
Control
With
the
passage
of
the
1990
Clean
Air
Act
Amendments,
the
United
States
adopted
substantially
more
stringent
NO
x
regulations,
and
utilities
began
installing
SCR
and
SNCR
systems
on
their
plants.
The
first
full­
scale
SCR
system
for
a
U.
S.
coal­
fired
power
plant
went
on
line
in
1995.
As
of
2000,
10
SCR
and
20
SNCR
systems
were
operational
(
Amar,
2000).
While
Europe
and
Japan
developed
the
SCR
technology,
application
in
the
United
States
has
demanded
further
innovation.
U.
S.
coal's
higher
sulfur
levels
required
design
changes,
and
government
regulation
demanded
much
stricter
control
of
ammonia
slip.
27
While
there
is
no
comprehensive
data
set
on
the
historical
costs
of
SCR
and
SNCR
in
the
United
States,
The
Northeast
States
for
Coordinated
Air
Use
Management
(
NESCAUM)
has
researched
NO
x
control
costs
by
contacting
utilities
and
investigating
cost
trends
on
a
case­
by­
case
basis.
NESCAUM
data
indicate
that
the
first
U.
S.
SCR
and
SNCR
projects
experienced
difficulties
with
ammonia
flow
controls
and
distribution,
ammonia
slip,
pressure
loss,
and
increased
outages
in
their
early
years,
which
is
consistent
with
the
initial
experience
in
Germany
and
Japan.
However,
NESCAUM
also
notes
that,
"
many
of
these
problems
have
been
overcome
as
plant
operators
move
up
the
learning
curve"
(
Amar,
2000).
The
NESCAUM
reports
indicate
both
capital
and
operating
cost
reductions
in
NO
x
control
technologies
have
occurred.

The
NESCAUM
case
studies
also
suggest
that
learning
and
innovation
have
had
a
clear
impact
on
actual
capital
costs.
For
instance,
Southern
California
Edison
retrofitted
its
boilers
with
SCR
systems
for
$
300
million,
a
full
$
650
million
less
than
the
firm's
1991
cost
estimate.
NESCAUM
explains:
"
these
savings
are
believed
to
largely
be
the
result
of
staging
in
controls,
thereby
letting
the
more
expensive
technologies
improve
in
cost
and
performance"
(
Staudt,
1998).
Similarly,
Public
Service
of
New
Hampshire
initially
considered
an
SCR
system
too
expensive
as
a
NO
x
control
option,
but
after
a
short
time
the
SCR
capital
cost
bids
reached
Public
Service's
budget
capability
(
Stradt,
1998).
Finally,
the
capital
costs
of
installing
an
SCR
system
at
the
Plymouth
Co­
generation
plant
in
Plymouth,
NH
were
35
percent
lower
than
expected
based
on
earlier
estimates,
leading
NESCAUM
28
This
power
plant
is
a
internal
combustion,
diesel
fuel
plant,
which
is
not
on
the
same
scale
as
most
public
utilities.
"
SCR
is
the
only
commercially
proven
secondary
NO
x
reduction
method
for
lean­
burn
gas
engines
and
diesel
engines,"
explains
NESCAUM
(
Amar,
2000b).
It
is
unclear
how
the
use
of
SRC
technology
to
control
NO
x
emissions
from
non­
utility
NO
x
sources
impacts
the
progress
rate
of
SRC
use
at
utilities.
Spillover
cost
reductions
may
result
from
the
experience
at
nonutilities
or
it
may
have
no
impact.

17
to
conclude
that
this
result
"
may
also
indicate
that
SCR
technology
is
becoming
less
costly"
(
Amar,
2000b).
28
Total
operating
costs
have
also
dropped
at
plants
with
SCR
and
SNCR
systems.
Examples
include
the
following:


New
England
Power
developed
and
installed
a
new
system
to
control
reagent
utilization
in
its
SNCR
at
Salem
Harbor,
an
innovation
that
has
provided
a
60
percent
reduction
in
reagent
use
and
an
annual
savings
of
$
600,000;


Montaup
Electric
Somerset
Station,
"
plant
operators
have,
over
time,
gained
more
experience
with
operation
of
the
[
SNCR]
system,"
which
has
brought
annual
operational
costs
down
by
a
total
of
over
60
percent
(
Stradt,
1998).

As
NESCAUM
points
out
in
recent
analysis
of
NO
x
control
costs,
progress
effects
have
not
been
factored
into
any
of
the
major
government
and
industry
engineering
cost
estimates
for
SCR
and
SNCR
since
1980.
As
a
result,
NO
x
control
costs
projections
in
1982,
1985,
and
1989
from
have
proven
excessive.
NESCAUM
writes,
"
The
difference
between
early,
pre­
regulatory
estimates
and
current
costs
is
substantial:
costs
declined
by
65
to
90
percent
for
SCR
(
on
a
dollars
per
ton
of
NO
x
removed
basis)
and
by
approximately
65
percent
for
SNCR
(
on
a
levelized
cost
basis).
Although
part
of
this
differential
is
attributable
to
overly
conservative
assumptions
in
the
pre­
regulatory
period,
inuse
experience
with
these
technologies
and
the
innovations
it
inspired
clearly
play
a
major
role"
(
Amar,
2000).
Exhibit
3
provides
summaries
of
EPRI
and
NESCAUM
cost
estimates
for
SCR
and
SNCR.

Exhibit
3
SCR
AND
SNCR
COST
ESTIMATES
SCR
COSTS:
(
1999
Dollars;
Based
on
a
500
MW,
Wall­
Fired
Boiler)

Study
Capital
Costs
($/
kW)
Cumulative
%
Decrease
$/
ton
Cumulative
%
Decrease
EPRI
1985
90­
155
2,800­
11,290
EPRI
1989
125
None
2,500­
5,000
4­
55
NESCAUM
1998
50­
75
40­
60
1,000­
1,100
64­
90
SNCR
COSTS:
(
1999
Dollars)
Exhibit
3
SCR
AND
SNCR
COST
ESTIMATES
29
Note
that
we
assume
that
early
cost
estimates
were
accurate
for
the
year
in
which
they
were
reported;
Amar
suggests
that
initial
costs
may
have
been
overestimated;
if
this
is
true
then
progress
curves
based
on
these
estimates
would
also
likely
be
high.

30
200
units
were
installed
in
Europe
and
Japan
from
1970
to
1990,
and
10
units
were
installed
in
the
United
States
from
1995
to
2000
(
Amar,
2000).
We
know
that
at
least
four
of
these
210
units
were
installed
in
Japan
before
1985
(
Staudt,
1998).
Assuming
that
the
10
SCR
units
installed
in
the
United
States
between
1995
and
2000
were
installed
at
a
rate
of
two
per
year,
we
conclude
that
in
1998
206
units
had
been
fully
installed.
Thus,
this
calculation
assumes
that
no
new
installation
of
SCR
units
occurred
in
Japan
or
Germany
after
1990.
Since
neither
country
passed
new
"
technology
forcing"
regulations
after
1990,
this
assumption
is
probably
accurate.
We
also
assume
18
Study
Capital
Costs
($/
kW)
Cumulative
%
Decrease
$/
ton
Cumulative
%
Decrease
EPRI
1982
29­
35
3.5­
3.75
EPRI
1989
6­
19
45­
80
4­
5
Increase
NESCAUM
1998
15
48­
57
1.25
69­
75
Sources:

SCR
estimates:
Amar
(
2000),
page
V­
5.
EPRI
1985
is
from
Miller,
EPRI
Coal
Combustion
Systems
Division
et
al.
(
1985).
SO2
and
NOx
Retrofit
Control
Technologies
Handbook.
Palo
Alto,
Electric
Power
Research
Institute:
CS­
4277­
SR
;
EPRI
1989
is
from
Eskinazi
et
al.
(
1989);
NESCAUM
1998
is
from
Straudt
(
1998).

SNCR
estimates:
Amar
(
2000),
page
V­
5.
EPRI
1989
is
from
Eskinazi
et
al.
(
1989);
EPRI
1982
is
from
EPA
OAQPS
(
1983)
Control
Technologies
for
Nitrogen
Oxides
Emissions
from
Stationary
Sources.
Research
Triangle
Park,
NC,
U.
S.
EPA
document:
EPA­
450/
3­
83­
002;
NESCAUM
1998
is
from
Straudt
(
1998).

Calculating
a
Progress
Rate
While
the
available
data
on
actual
cost
decreases
are
limited,
it
is
possible
to
use
NESCAUM's
total
reported
changes
in
cost
to
calculate
a
rough
progress
rate
estimate
for
both
SCR
and
SNCR
using
the
log­
linear
function
and
the
evidence
of
decreasing
costs.
This
curve
is
an
approximation
that
reflects
only
two
data
points
and
does
not
attempt
to
correct
for
other
"
nonlearning
forces,
but
it
provides
an
initial
indication
of
what
progress
curves
for
SCR
and
SNCR
could
look
like.
29
To
calculate
an
SCR
progress
rate
based
on
the
actual
percentage
cost
decreases
reported
by
NESCAUM,
we
must
establish
an
estimate
of
cumulative
production
for
the
different
point­
in­
time
cost
estimates.
Based
on
NESCAUM
data,
we
assume
that
there
were
four
coal­
burning
utility
SCR
units
installed
in
the
world
(
including
Europe
and
Japan)
by
1985,
and
a
total
of
206
units
installed
by
1998.30
The
total
actual
cost
decrease
from
1985
to
1998
of
65
to
90
percent
reported
by
that
no
other
units
have
been
installed
in
other
regions
of
the
world.
If
more
units
have
been
installed,
the
progress
rate
we
calculate
would
be
too
aggressive.
For
instance,
if
500
units
had
been
installed
by
1998,
the
progress
rate
would
be
72
to
86
percent,
a
few
percentage
points
higher
than
the
range
we
report
here.

31
To
project
a
conservative
estimate
of
progress
(
eliminating
catalyst
costs)
one
could
use
only
the
capital
cost
estimate
changes
reported
by
NESCAUM.
The
40
to
60
percent
decreases
in
capital
costs
over
time
correspond
to
progress
curves
of
91
and
85
percent,
respectively.
However,
these
estimates
are
likely
to
be
overly
conservative
because
they
ignore
the
extent
to
which
learning
by
using,
learning
by
doing,
and
incremental
innovation
contributed
to
reductions
in
both
the
costs
and
the
required
quantities
of
catalyst.

32
According
to
Amar
(
2000),
300
SNCR
units
were
installed
worldwide
by
2000.
Assuming
no
units
were
installed
until
1988,
and
then
25
were
installed
each
year
after
that,
we
estimate
that
250
units
were
installed
worldwide
by
1998.

33
According
to
Amar
(
2000),
20
SNCR
units
were
installed
in
the
United
States
by
2000,
the
first
of
which
was
installed
in
Salem,
MA
in
1992.
Assuming
2.5
units
were
installed
each
year
from
1992
to
2000,
15
units
would
have
been
installed
by
1998.
Thus,
we
assume
15
domestic
units
in
this
calculation.

19
NESCAUM,
when
considered
as
a
function
of
cumulative
production,
indicates
a
progress
rate
between
83
and
67
percent
(
i.
e.,
cost
reductions
of
between
17
and
33
percent,
respectively,
with
each
doubling
of
production).
This
range
is
likely
to
represent
a
relatively
aggressive
set
of
progress
curve
estimates,
because
it
incorporates
the
significant
changes
in
catalyst
costs
that
may
be
due
in
part
to
factors
other
than
learning
and
progress.
31
To
calculate
a
progress
rate
for
SNCR
development
on
a
worldwide
scale,
we
assume
that
250
SNCR
units
were
installed
in
the
world
from
1982
to
1998.32
Again
relying
on
the
total
cost
decrease
of
69
to
75
percent
reported
by
NESCAUM,
this
technology's
implied
progress
rate
ranges
from
86
to
84
percent,
respectively.
However,
since
SNCR
did
not
fully
develop
in
Japan
and
Germany
before
being
introduced
in
the
U.
S.,
an
alternative
approach
might
include
only
U.
S.
installed
units
when
calculating
SNCR
development
progress
in
the
United
States
(
i.
e.
assuming
that
there
is
no
"
shared"
learning).
Assuming
that
15
SNCR
units
were
installed
between
1982
and
1998
within
the
United
States,
the
implied
progress
rate
for
SNCR
ranges
from
70
to
74
percent.
33
In
reality,
SNCR
in
the
United
States
has
probably
been
influenced
by
the
installation
of
units
around
the
world.
However,
the
quantity
of
domestically
installed
units
is
probably
more
influential
in
bringing
down
domestic
costs,
since
knowledge
and
experience
gained
in
Europe
often
does
not
reach
U.
S.
utilities,
and
the
specific
coal
content
and
local
regulations
create
specific
challenges
to
U.
S.
installation.
Thus,
the
actual
progress
rate
for
SNCR
is
probably
somewhere
in
the
middle
of
the
range
that
we
have
estimated
(
i.
e.,
70
to
86
percent,
indicating
cost
decreases
of
30
to
14
percent,
respectively,
as
production
doubles).
20
Again,
the
progress
ratios
presented
here
are
preliminary,
and
reflect
the
quality
of
the
underlying
estimates.
However,
as
an
initial
exercise
in
estimating
the
potential
impacts
of
learning,
they
do
provide
support
for
the
use
of
an
80
percent
progress
factor
in
preliminary
estimates
of
future
pollution
abatement
technology
costs.

Example
3:
Vatavuk
Air
Pollution
Control
Cost
Index
Since
the
dollar
value
of
goods
and
services
changes
over
time,
government
and
academics
have
developed
cost
and
price
indices
in
order
to
allow
for
the
comparison
of
prices
in
one
time
to
that
of
a
different
era.
General
indices,
such
as
the
Gross
Domestic
Product
(
GDP)
deflator,
the
Consumer
Price
Index
(
CPI),
and
the
Producer
Price
Index
(
PPI)
track
the
costs
and
prices
of
an
extremely
broad
"
basket
of
goods"
in
order
to
measure
the
relative
purchasing
power
of
a
single
dollar
over
time.
These
general
indices
are
usually
seen
as
a
good
measure
of
the
rate
of
"
inflation."

In
some
cases,
analysts
wish
to
track
changes
in
the
cost
of
certain
goods
(
e.
g.
the
Marshall
&
Swift
Equipment
Cost
Index
records
the
price
of
equipment
used
in
"
process
industries,"
such
as
paper
and
paint
production).
These
costs
are
tracked
in
more
specific
indices
that
may
or
may
not
match
the
broader
inflation
indices,
depending
on
the
changes
in
the
price
or
supply
of
specific
raw
materials
and
components.

In
1995,
William
M.
Vatavuk
of
EPA's
Office
of
Air
Quality
Planning
and
Standards
(
OAQPS)
developed
cost
indices
for
particular
pollution
control
technologies.
These
indices,
collectively
known
as
the
Vatavuk
Air
Pollution
Control
Cost
Indices
(
VAPCCI),
are
designed
to
be
used
"
to
escalate
costs
from
the
initial
(
base)
period
(
first
quarter
1994)
forward
to
any
quarter
in
the
future"
(
Vatavuk,
1995).
For
six
volatile
organic
compound
(
VOC)
control
technologies
and
one
particulate
matter
(
PM)
control
technology
(
wet
scrubbers),
the
indices
provide
a
comparative
cost
record
from
1989
to
2001.

As
part
of
our
examination
of
progress,
we
analyze
the
Vatavuk
Indices
to
determine
whether
the
costs
for
VOC
and
PM
control
technologies
are
declining
when
adjusted
for
inflation.
If
the
progress
effect
is
present
in
this
pollution
control
industry,
we
believe
that
correcting
the
Vatavuk
Indices
for
inflation
should
reveal
a
decreasing
cost
trend
for
the
pollution
abatement
technologies
tracked
by
the
Vatavuk
Indices.

This
effort
is
not
designed
to
isolate
the
actual
progress
ratios
of
specific
pollution
control
technologies
for
a
number
of
reasons.
First,
indices
show
changes
in
price
over
time
and
do
not
provide
information
about
cumulative
production.
Second,
the
indices
do
not
reflect
the
total
cost
associated
with
manufacturing
pollution
control
equipment.
Third,
some
cost
changes
recorded
by
the
indices
may
be
the
result
of
non­
progress­
related
shifts
in
the
marketplace
(
i.
e.
changes
in
commodities
markets
or
a
drop
in
oil
prices).
Regardless,
this
index
provides
a
well
established
record
of
pollution
abatement
costs
that
may
provide
strong
qualitative
evidence
that
a
progress
effect
is
active
in
the
pollution
control
industry.
21
The
Index
William
Vatavuk
developed
the
VAPCCI
because
in
his
analysis
he
found
that
more
general
indices
C
such
as
the
(
CPI),
the
(
PPI),
and
even
the
more
specific
Chemical
Engineering
Plant
Cost
Index
C
did
not
accurately
reflect
the
relative
costs
of
pollution
control
equipment
over
time.
The
"
basket
of
goods"
used
in
these
indices
simply
did
not
match
the
inputs
to
pollution
abatement
technology,
and
these
inputs
did
not
necessarily
follow
more
general
cost
trends
(
Vatavuk,
2000).

To
develop
his
index,
Vatavuk
used
the
cost
data
of
individual
basket
items
in
the
PPI
to
compile
a
new
basket
of
goods
which
tracked
the
relative
cost
trends
exclusively
of
each
pollution
abatement
technology's
components.
He
then
created
a
relative
cost
trend
for
each
technology
by
averaging
each
of
the
individual
basket
items,
weighted
each
component
in
proportion
to
its
overall
impact
on
the
technology
cost
(
OAQPS,
1995).

Like
all
indices,
the
VAPCCI
is
actually
a
list
of
relative
values
referenced
to
a
specific
month,
quarter,
or
year.
To
convert
the
price
of
a
good
from
one
period
of
time
to
another,
one
uses
the
following
equation:

Cost
New
=
Cost
Old

(
Index
New
/
Index
Old)

Vatavuk
offers
a
good
example:
"
Suppose
that
in
first
quarter
1994,
a
wet
scrubber
cost
$
100,000.
Now
estimate
how
much
this
scrubber
would
cost
in
first
quarter
1995.
If
the
1994
(
old)
VAPCCI
value
was
100.0,
and
the
1995
(
new)
VAPCCI
value
was
109.9,
then,
the
1995
estimated
equipment
cost
is:

Cost
New
=
$
100,000

(
109.9/
100.0)
=
$
109,900"
(
Vatavuk,
1995)

The
Progress
Effect
Reflected
in
the
Index
While
Vatavuk
assumed
that
pollution
abatement
costs
are
relative
to
the
cost
of
their
inputs,
his
pollution
control
cost
indices
do
not
control
for
the
progress
effect.
As
Montgomery
and
Day
(
1985)
note,
industry­
specific
and
product­
specific
indices
"
may
capture
a
substantial
portion
of
productivity
gains
in
the
industry."
As
a
result,
we
expected
that
the
VAPCCI
would
show
a
lower
growth
rate
than
inflation
as
measured
by
GDP
and
CPI
if
the
progress
effect
was
reducing
costs
in
real
terms
as
inflation
caused
current
dollar
increases.
To
assess
this
possibility,
we
compared
the
Vatavuk
indices
for
technologies
that
have
been
tracked
from
1989
to
2001
to
CPI
and
the
GDP
deflator.

This
analysis
places
the
Vatavuk
indices
in
comparison
to
both
CPI
and
the
GDP
deflator
because,
while
CPI
is
the
most
commonly
used
inflation
index,
Montgomery
and
Day
(
1985)
suggest
that
the
GDP
deflator
probably
offers
a
more
appropriate
measure
of
"
the
changing
value
of
money."
The
GDP
deflator
is
not
bound
to
consumer
prices,
and
it
measures
a
more
comprehensive
basket
of
goods
than
CPI
(
e.
g.,
the
GDP
includes
changes
in
the
cost
of
production
inputs
that
are
not
sold
34
Exhibits
4
and
5
reveal,
especially
around
the
year
1994,
that
input
costs
can
also
rise
relative
to
inflation
due
to
lack
of
supply
or
other
market
forces.
If,
for
instance,
an
earthquake
destroyed
a
major
catalyst
production
facility,
the
cost
of
catalytic
incineration
could
rise
rapidly
due
to
a
shortage
of
catalyst
in
the
marketplace.
The
progress
function
assumes
a
steady
supply
of
inputs
in
a
competitive
marketplace.

35
To
assume
that
the
"
real"
decreases
in
costs
demonstrated
in
Exhibits
1
and
2
represent
an
actual
progress
curve,
one
would
have
to
assume
both
that
all
pollution
equipment
production
began
in
1989
and
that
production
rates
of
these
technologies
have
been
exactly
constant
since
1989.

22
directly
to
consumers,
such
as
manufacturing
equipment.)
As
a
result,
it
is
generally
more
conservative
in
its
estimate
of
inflation
than
CPI
(
for
a
comparison
of
the
two
indices,
see
Exhibit
6).

When
compared
to
the
change
in
cost
as
reflected
in
both
CPI
and
GDP,
the
Vatavuk
indices
for
catalytic
incinerators,
flares,
gas
absorbers,
refrigeration
systems,
regenerative
thermal
oxidizers,
thermal
incinerators,
and
wet
scrubbers
all
showed
slower
growth
between
1989
and
2001,
as
demonstrated
in
Exhibits
4
and
5.
The
decrease
in
real
costs
from
1989
to
2001
ranged
from
1.5
percent
(
thermal
incinerators)
to
20.4
percent
(
refrigeration
systems)
when
compared
to
the
GDP
deflator.
While
this
may
initially
seem
modest
when
compared
to
the
20
percent
reductions
expected
in
the
progress
curves,
it
is
important
to
note
that
the
indices
capture
only
changes
in
the
component
cost
of
each
technology.
A
complete
"
progress
curve"
in
the
regulatory
setting
would
also
have
to
consider
labor
learning
and
assembly
line
improvements,
as
well
as
the
impact
of
"
learning­
by­
using"
on
the
regulated
community's
operating
costs.
Thus,
these
real
cost
decreases
are
consistent
with
the
presence
of
the
progress
curve
in
the
pollution
abatement
industry.

Although
all
seven
technologies
that
we
examine
do
sink
in
cost
when
adjusted
for
inflation,
Vatavuk
emphasizes
the
difficulty
in
determining
which
cost­
influencing
force
(
e.
g.
raw
material
cost
variation,
labor
shifts,
and
market
forces)
drives
each
cost
down;
however,
he
does
identify
two
specific
forces
which
drove
down
prices
consistent
with
the
progress
effect.
He
writes,
"
With
catalytic
incinerators,
a
drop
in
catalyst
prices
led
vendors
to
lower
their
prices
by
roughly
7%
between
fourth
quarter
1991
and
first
quarter
1992...
Similarly,
a
design
innovation
marketed
by
a
leader
in
refrigeration
systems
prompted
the
company
to
lower
prices
in
mid­
1993,
causing
its
competitors
to
follow
suit"
(
Vatavuk,
1995).
These
changes
represent
significant
evidence
of
progress
effects
in
the
pollution
control
industry.
34
Conclusions
Exhibits
4
and
5
indicate
that
the
progress
effect
may
be
present
in
the
Vatavuk
cost
indices.
However,
the
VAPCCI
by
itself
does
not
contain
enough
information
to
allow
this
effect
to
be
isolated
or
quantified.
The
VAPCCI
traces
trends
in
cost
over
time,
not
cumulative
production
(
which
is
the
actual
driver
of
the
log­
linear
progress
effect).
35
To
identify
actual
progress
curves,
it
would
be
necessary
to
determine
cumulative
production
of
each
of
the
VAPCCI
technologies.
36
While
traditional
labor
costs
are
not
quantified
in
the
Vatavuk
indices,
engineering
costs
are
factored
in,
and
labor
does
factor
into
the
cost
of
input
production.
The
potential
unquantified
progress
benefit
specifically
applies
to
improved
assembly
of
the
product
over
time.

23
The
VAPCCI
also
would
likely
reflect
only
a
portion
of
the
progress
curve
associated
with
end­
user
compliance
costs,
because
it
does
not
account
for
changes
in
pollution
technology
production
labor
costs.
36
The
index
also
does
not
record
the
pollution
control
technology
customer
cost
reductions
that
should
develop
as
personnel
learn
to
use
the
products
more
effectively.
Since
the
progress
effect
is
linked
closely
to
the
impacts
of
labor
learning,
the
VAPCCI
data
may
not
fully
record
the
cost
reductions
due
to
cumulative
production.

There
are
clear
limitations
on
the
use
of
the
Vatavuk
indices
to
quantify
the
progress
effect,
but
the
sinking
real
costs
of
pollution
technology
do
indicate
that
learning
and
progress
are
present
on
an
industry­
wide
level.
The
Vatavuk
indices
offer
further
evidence
that
the
progress
curve
applies
to
VOC
control
technologies,
PM
control
technologies,
and
the
pollution
control
industry
as
a
whole.
24
Exhibit
4:
Rate
of
VAPCCI
Growth
Relative
to
CPI
­
30
­
25
­
20
­
15
­
10
­
5
0
5
1988
1990
1992
1994
1996
1998
2000
2002
Year
Growth
of
Index
relative
to
CPI
Catalytic
Incinerators
Flares
Gas
Absorbers
Refrigeration
Systems
Regenerative
Themal
Oxidizers
Thermal
Incinerators
Wet
Scrubbers
Exhibit
5:
Rate
of
VAPCCI
Relative
to
GDP
­
25
­
20
­
15
­
10
­
5
0
5
10
1988
1990
1992
1994
1996
1998
2000
2002
Year
Index
Growth
Relative
to
GDP
Catalytic
Incinerators
Flares
Gas
Absorbers
Refrigeration
Systems
Regenerative
Themal
Oxidizers
Thermal
Incinerators
Wet
Scrubbers
25
Exhibit
6:
CPI
and
GDP
Deflator
Growth
Rate,
1989
to
2001
100
105
110
115
120
125
130
135
140
145
150
1988
1990
1992
1994
1996
1998
2000
2002
Year
Index
value
CPI
GDP
Deflator
IMPLICATIONS
OF
PROGRESS
CURVES
FOR
EPA
COST
ESTIMATION
Our
review
of
the
existing
theoretical
and
empirical
evidence
on
the
impacts
of
learning
and
progress
on
costs
over
time,
as
well
as
our
analysis
of
available
data
on
cost
trends
in
pollution
control
devices,
leads
us
to
two
major
conclusions:


The
learning
curve
concept
has
a
sound
theoretical
and
empirical
basis;
and

There
is
evidence
that
learning
and
progress
do
occur
in
the
production
of
air
pollution
control
devices,
so
that
costs
for
these
devices
can
be
expected
to
decline
over
time,
in
real
terms.

For
these
reasons,
we
find
there
is
a
strong
basis
for
adjusting
regulatory
cost
estimates
to
reflect
the
effects
of
progress.
There
remains
some
uncertainty
about
the
precise
magnitude
of
the
adjustment
that
ought
to
be
applied.
While
the
available
evidence
suggests
that
in
the
vast
majority
of
cases
this
factor
will
have
the
effect
of
reducing
estimates
of
regulatory
compliance
costs,
there
are
some
limited
and
likely
rare
situations
where
an
active
and
consistent
effort
to
invest
in
and
maintain
learning
may
either
be
very
costly
or
may
fail
altogether,
suggesting
that
overall
costs
might
not
decrease
or
could
even
increase
over
time.
Data
reported
by
Dutton
and
Thomas
(
1984),
for
example,
indicate
that
only
one
of
the
108
industries
that
they
studied
experienced
a
cost
increase
over
their
study
period.
37
An
alternative
is
to
use
time
as
a
proxy
for
production
volume,
provided
reasonable
projections
of
production
rates
can
be
made.
This
approach
was
used
in
the
recent
Tier
II
tailpipe
standards
RIA
(
USEPA
1999).

26
The
presence
of
these
uncertainties
in
the
quantitative
effect
of
the
application
of
progress
curves
suggests
that
analysts
should
consider
whether
an
explicit
quantitative
uncertainty
analysis
can
be
applied.
The
use
of
a
single
progress
ratio
may
in
some
cases
suggest
an
unrealistic
sense
of
certainty
in
the
anticipated
pace
of
progress,
so
a
range
or,
better,
a
distribution
of
progress
ratios
could
be
applied
to
characterize
uncertainty.
Some
of
the
data
presented
above
provide
a
basis
for
quantifying
within­
industry
and
cross­
industry
variation
in
the
pace
of
progress
(
see
McDonald
and
Schrattenholzer
(
2001)
for
an
example
of
the
former,
Dutton
and
Thomas
(
1984)
for
the
latter).
However,
this
type
of
quantitative
treatment
of
uncertainty
may
not
always
be
practical
or
appropriate
for
this
particular
parameter.
Available
data
present
a
relatively
tight
distribution
of
estimates
around
the
most
commonly
cited
80
percent
progress
ratio.
Analysts
may
therefore
be
justified
in
applying
a
single
adjustment
factor,
particularly
in
those
cases
where
it
is
clearly
applied
conservatively
to
avoid
overstating
the
effect
of
progress
on
costs.
This
type
of
approach
was
applied
in
the
recent
mobile
source
RIAs,
as
described
above.

Beyond
this
general
recommendation
to
acknowledge
uncertainty,
we
also
suggest
the
following
guidelines
for
application
of
progress
curves
to
regulatory
cost
estimates:


Pay
careful
attention
to
the
production
cycle.
Rapid
cost
decreases
as
a
result
of
progress
curves
and
new
regulatory
requirements
are
likely
to
be
more
appropriate
for
the
production
of
relatively
new
or
small
production
technologies.
A
slower
decrease
in
costs
would
be
expected
for
established
technologies
with
a
long
history
of
production,
because
the
increment
of
increased
production
relative
to
cumulative
production
is
smaller,
implying
a
longer
time
period
for
additional
cost
decreases.
For
this
reason,
proper
application
of
the
80
percent
rule
requires
reasonable
data
on
production
volume
and
projected
production
volume,
as
described
above.
37

Apply
progress
adjustments
only
to
cost
elements
where
a
realistic
possibility
of
learning
exists.
Progress
adjustments
are
generally
most
applicable
to
production
of
specific
devices,
and
are
less
applicable
to
direct
materials
or
energy
costs.
An
argument
can
be
made
that
energy
and
materials
extraction
and
refinement
technologies
are
also
likely
to
grow
more
efficient
over
time,
but
the
markets
for
these
commodities
are
complex
and
clearly
affected
by
forces
other
than
learning.
Sensitivity
analyses
are
likely
to
be
a
more
appropriate
way
of
addressing
potential
cost
estimation
inaccuracies
for
materials
and
energy
cost
elements.


Pay
attention
to
the
timing
of
rule
implementation,
and
whether
it
allows
for
learning.
For
some
rules,
particularly
those
that
prescribe
the
installation
of
a
specific
technology
within
a
short
period
of
time
(
a
year
or
less),
it
may
not
be
realistic
to
assume
that
there
is
enough
time
for
learning
by
doing
or
learning
by
using
to
be
meaningful.
In
those
cases,
learning
by
doing
and/
or
learning
by
using
may
be
more
applicable
for
new
establishments
that
purchase
27
the
technology
after
the
initial
installation
cycle,
assuming
that
the
cumulative
production
volume
is
sufficient
for
significant
learning.


Rely
on
the
80
percent
rule,
and
attempt
to
refine
as
data
permit.
The
case
studies
presented
above
illustrate
the
80
percent
rule
is
a
good
central
estimate
of
a
progress
curve
for
many
types
of
pollution
control
and,
more
generally,
for
manufacturing
industries.
In
some
cases,
however
(
e.
g,
scrubbers),
specific
industry
data
may
be
available
and
analysts
should
consider
applying
an
industry
specific
progress
curve.
Use
of
the
80
percent
rule
as
a
default
value,
however,
is
consistent
with
the
conclusion
that
the
general
literature
on
progress
is
sound
and
has
proven
to
be
broadly
applicable.


Consider
making
use
of
data
from
similar
air
pollution
devices
(
described
above)
to
evaluate
the
validity
of
assumptions
about
the
applicability
of
the
80
percent
rule.
In
general,
the
available
data
sources
to
estimate
progress
curves
for
pollution
control
devices
suffer
from
limitations
that
make
the
data
imperfect
representations
of
a
true
learning.
Even
in
the
case
of
sulfur
scrubbers,
where
there
are
relatively
comprehensive
data
on
costs
over
time,
analysts
make
slightly
different
assumptions
in
calculating
a
progress
curve
that
cause
the
estimates
to
vary,
in
some
cases
over
a
considerable
range.
An
encouraging
finding
of
our
work,
however,
is
that
despite
specific
concerns
over
our
ability
to
isolate
learning
effects
from
other
factors
that
influence
costs
over
time,
the
historic
pollution
control
cost
data
appears
broadly
consistent
with
the
results
of
applying
an
80
percent
progress
ratio.
28
SOURCES
Argote,
L.
and
D.
Epple,
February
23,
1990.
"
Learning
Curves
in
Manufacturing,"
Science,
Vol.
247,
p.
920­
924.

Belkaoui,
A.,
1986.
The
Learning
Curve:
A
Management
Accounting
Tool.
Westport:
Quorum
Books.

Benkard,
C.
L.,
2000.
"
Learning
and
Forgetting:
The
Dynamics
of
Aircraft
Production,"
The
American
Economic
Review,
Vol.
90,
No.
4,
p.
1034­
1054.

Burtraw,
Dallas,
February
1996."
Cost
Savings
Sans
Allowance
Trades?
Evaluating
the
SO2
Emission
Trading
Program
to
Date",
Resources
for
the
Future.

Botsford,
C.,
July,
2001.
"
The
Two
Faces
of
NOx
Control,"
Chemical
Engineering,
p.
66­
71.

Dutton,
J.
and
A.
Thomas,
1984.
"
Treating
Progress
Functions
as
a
Managerial
Opportunity,"
Academy
of
Management
Review,
Vol.
9,
No.
2,
p.
237­
247.

Gritsevskyi,
A.
and
N.
Nakicenovic,
2000.
"
Modeling
Uncertainty
of
Induced
Technological
Change,"
Energy
Policy,
No.
28,
p.
907­
921.

Greening,
L.,
J.
Laitner,
and
S.
Bernow,
September
2000.
"
Capturing
the
Role
of
`
Technology
Learning'
in
Transforming
Energy
Consumption
Patterns
within
the
U.
S.
Manufacturing
Sector,"
Transforming
Energy:
Proceedings
of
the
USAEE/
IAEE
North
American
Conference,
Philadelphia,
Pennsylvania.

Harrington,
W.,
R.
D.
Morgenstern,
and
P.
Nelson,
January
1999.
"
On
the
Accuracy
of
Regulatory
Cost
Estimates,"
Resources
for
the
Future
Discussion
Paper
99­
18.

Institute
for
Clean
Air
Companies,
September
1997.
"
Air
Emissions
Monitoring
for
Safe
and
Efficient
Medical
Waste
Incinerator
Operation."

Institute
of
Clean
Air
Companies,
May
1995.
Scrubber
Myths
&
Realities,
White
Paper.

International
Energy
Agency,
2000.
Experience
Curves
for
Energy
Technology
Policy.
Paris:
IEA
Publications.

Jacobsen,
H.
K.,
2001.
"
Technological
Progress
and
Long­
term
Energy
Demand
 
a
Survey
of
Recent
Approaches
and
a
Danish
Case,"
Energy
Policy,
No.
29,
p.
147­
157.

Jarmin,
R.,
August,
1996.
"
Learning
by
Doing
and
Plant
Characteristics,"
Discussion
Paper
for
the
Center
for
Economic
Studies,
U.
S.
Census
Bureau.
29
Joskow,
P.
and
N.
Rose,
1985.
"
The
Effects
of
Technological
Change,
Experience,
and
Environmental
Regulation
on
the
Construction
Cost
of
Coal­
burning
Generation
Units,"
Rand
Journal
of
Economics,
Vol.
16,
No.
1,
p.
1­
27.

Laitner,
J.,
undated.
"
Analyzing
changes
in
Scrubber
Costs
over
Time"
unpublished
memorandum,
U.
S.
Environmental
Protection
Agency,
Office
of
Atmospheric
Programs.

Laitner,
J.,
undated.
"
Scrubber
Installations
and
Costs."
Internal
Spreadsheet
Analysis
for
the
EPA
Office
of
Atmospheric
Programs.

McDonald,
A.
and
L.
Schrattenholzer,
2001.
"
Learning
Rates
for
Energy
Technologies,"
Energy
Policy,
No.
29,
p.
255­
261.

Montgomery,
D.
and
G.
Day,
February,
1985.
"
Experience
Curves:
Evidence,
Empirical
Issues,
and
Applications,"
Working
Paper
for
the
Market
Science
Institute,
Report
No.
85­
101.

Neji,
L.,
1997.
"
Use
of
Experience
Curves
to
Analyse
the
Prospects
for
Diffusion
and
Adoption
of
Renewable
Energy
Technology,"
Energy
Policy,
Vol.
23,
No.
13,
p.
1099­
1107.

Smith,
R.,
1997.
"
Testimony
of
the
Institute
for
Clean
Air
Companies,
Inc.
Before
the
U.
S.
Senate
Subcommittee
on
Manufacturing
and
Competitiveness
on
EPA's
Revised
NAAQS
for
Particulate
Matter
and
Ozone."

Taylor,
M.
R.,
E.
S.
Rubin,
and
D.
A.
Hounshell,
2001.
"
The
Effect
of
Government
Actions
on
Technological
Innovation
for
SO
2
Control"
white
paper
presented
at
the
August
2001
"
Mega
Symposium"
sponsored
by
U.
S.
Environmental
Protection
Agency,
Electric
Power
Research
Institute,
and
U.
S.
Department
of
Energy.

Wooley,
K.
M.,
1972.
"
Experience
Curves
and
Their
Use
in
Planning,"
unpublished
Ph.
D.
dissertation,
Stanford
University.
As
cited
in
Montgomery,
D.
and
G.
Day,
February,
1985.
"
Experience
Curves:
Evidence,
Empirical
Issues,
and
Applications,"
Working
Paper
for
the
Market
Science
Institute,
Report
No.
85­
101.

Wright,
T.
P.,
1936.
Journal
of
Aeronautical
Science,
3,
122.
