Technology International Trade and Pollution from U S Manufacturing


NCEE#

NATIONAL CENTER FOR

ENVIRONMENTAL ECONOMICS

Technology, International Trade, and Pollution from U.S.

Manufacturing

Arik Levinson

Working Paper Series

Working Paper # 08-02
February, 2008

£

M'i

1 *

U.S. Environmental Protection Agency
National Center for Environmental Economics

5	"	1200 Pennsylvania Avenue, NW(MC 1809)

Washington, DC 20460

¦h

pro^^	http://www.epa.gov/economics

-------
Technology, International Trade, and Pollution from U.S.

Manufacturing

Arik Levinson

NCEE Working Paper Series

Working Paper # 08-02
February, 2008

DISCLAIMER

The views expressed in this paper are those of the author(s) and do not necessarily represent
those of the U.S. Environmental Protection Agency. In addition, although the research described
in this paper may have been funded entirely or in part by the U.S. Evironmental Protection
Agency, it has not been subjected to the Agency's required peer and policy review. No official
Agency endorsement should be inferred.

-------
Technology, International Trade, and Pollution from U.S.

Manufacturing

November 9, 2007

Arik Levinson
Georgetown University Economics Department
aml6@georgetown. edu

Abstract

Total pollution emitted by U.S. manufacturers declined over the past 30 years by about
60 percent, even though real manufacturing output increased 70 percent. This
improvement must result from a combination of two trends: (1) changes in production or
abatement processes ("technology"); or (2) changes in the mix of goods manufactured in
the United States, which itself may result from increased net imports of pollution-
intensive goods ("international trade"). I first show that most of the decline in pollution
from U.S. manufacturing has been the result of changing technology, rather than changes
in the mix of goods produced, although the pace of that technology change has slowed
over time. Second, I present evidence that increases in net imports of pollution-intensive
goods are too small to explain more than about half of the pollution reductions from the
changing mix of goods produced in the United States. Together, these two findings
demonstrate that shifting polluting industries overseas has played at most a minor role in
the cleanup of U.S. manufacturing.

Key Words: pollution havens, trade and environment, pollution intensity.
Subject Matter Classification: International Trade, Industrial Sources

Acknowledgments

I am grateful to the National Center for Environmental Economics (NCEE) and to
Resources for the Future for hosting me during parts of this research, to Mun Ho, Carl
Pasurka, Jared Creason and Wayne Gray for helpful conversations and suggestions, to
Jennifer Blessing for research assistance, and to Randy Becker and Wayne Gray for
providing an advance copy of the updated NBER-CES Manufacturing Productivity
Database. Roy Huntley and Rhonda Thompson at the EPA provided invaluable help
interpreting the National Emissions Inventory.

-------
Technology, International Trade, and Pollution from U.S. Manufacturing
Introduction

Total pollution emitted by U.S. manufacturers has declined over the past 30 years,
by amounts ranging from 30 percent for nitrogen oxides (NOx) to 66 percent for sulfur
dioxide (SO2). At the same time, the real value of manufacturing output has increased by
more than 70 percent. This cleanup can be divided into two components: (1) advances in
production or abatement processes ("technology"), and (2) changes in the composition of
goods manufactured in the United States. The change in industry composition can further
be divided into (a) decreases in pollution-intensive goods consumed, and (b) increases in
pollution-intensive goods imported ("international trade"). How much of the overall
pollution reduction stems from technology, and how much from international trade?

In Part 1 of this analysis, I show that technology accounts for well over half of the
overall reductions in pollution from manufacturing. In Part 2,1 show that increases in net
imports of polluting goods can account for - at most - half of the pollution reductions
resulting from the changing composition of U.S. manufacturing (which is itself a small
part of the total per Part 1). Together, these two results demonstrate that shifting polluting
industries overseas has played at most a minor role in the overall cleanup of the U.S.
manufacturing sector.

Allocating credit for the cleanup of manufacturing among these trends in
technology and international trade is important for several reasons. Most U.S.
environmental regulations have been designed explicitly to alter the technology of
production, not the mix of goods produced or consumed. This paper asks whether U.S.
manufacturing pollution has declined as a consequence of following the stated intent of
the laws (changing technology) or as a consequence of avoiding those laws by producing
overseas (international trade). More importantly, if the regulations have simply changed
what gets manufactured domestically and have increased imports of polluting products
from developing countries, the improvements experienced by the United States would not
be replicable indefinitely on a global scale, because the poorest countries will never have
even poorer countries from which to import their polluting products. However, if the
cleanup has been the result of technology, that may well be replicable indefinitely, and
may even be replicable more easily if there are economies of scale or learning-by-doing
in abating pollution.

-------
To be clear, manufacturing is not the only source of pollution in the United States.
It accounts for less than 25 percent of the most common air pollutants (U.S. EPA 2000).
Nevertheless, manufacturing does account for a large share of the rhetoric in the debate
about the economic consequences of environmental regulations, the changes over time in
the structure of the U.S. economy, and the effect of international trade on U.S. production
workers and other countries' environments. Other major polluting sectors, such as electric
utilities and transportation, are not subject to concerns about pollution havens or
industrial flight, which is why I focus on manufacturing here.

In Part 1,1 use data from the U.S. Environmental Protection Agency (EPA) to
determine how much of the cleanup of U.S. manufacturing comes from changes in
technology versus changes in the mix of industries. Because the EPA data report
emissions back to 1970 for only four common air pollutants, this first part focuses on air
pollution. I show that for the typical air pollutant, the cleanup from technology is at least
as large as the cleanup from compositional change, and for some it is more than two
times as large. The pace of that technological progress, however, has been slowing over
time. Even though this type of decomposition analysis has deep roots in economics,
going back at least to Leontief (1970), to my knowledge this is the first paper to divide
pollution changes into its components in this way, or to demonstrate the technological
slowdown.1

In Part 2,1 address whether the decline in U.S. pollution that results from
increased imports is sufficiently large to explain the decline in pollution arising from the
change in the composition of U.S. industries. Because this part does not require emissions
data back to the 1970s, I expand the analysis to include 9 different air, water, and toxic
pollutants. The basic approach was outlined by Koo (1974) and implemented in part by
Khan (2003), Cole (2004), Ederington et al. (2004), and Gamper-Rabindran (2006). All
this previous research finds that increases in imports of polluting goods cannot explain

1 See Ang (1999), Rose (1999) and Metcalf (2007) for decomposition analyses of energy use and pollution.
Most such analyses fall into one of two categories: (1) index decomposition analysis (IDA), which uses
industry-level data; and (2) structural decomposition analysis (SDA), which combines industry-level data
with input-output tables. In this paper, I use elements from both approaches, relying on IDA for Part 1,
where input-output tables are unnecessary, and then adding the input-output tables in Part 2 to study
changes caused by imports, but without allowing the input-output requirements coefficients to change
annually. See also Pasurka (2003) for an update and an application using two-digit Standard Industrial
Classification (SIC) codes and S02 pollution.

-------
the composition shift of U.S. manufacturing toward cleaner goods. By ignoring the
pollution caused by production of the intermediate inputs to imports, however, these
papers all understate the degree to which those imports have displaced pollution in the
United States.

To account for intermediate inputs to imports, in Part 2 I use EPA data along with
U.S. Bureau of Economic Analysis (BEA) input-output tables to construct total emissions
intensities for each U.S. manufacturing industry in 1997, including the pollution caused
by each industry's intermediate inputs. I show that even after accounting for intermediate
inputs, the composition of imports to the United States has shifted towards cleaner goods,
and that increases in international trade can account for at most only half of the pollution
reductions from the changing composition of U.S. manufacturing. Foreign pollution
havens have had a relatively small role in the composition shift of U.S. manufacturing
away from polluting goods, and that composition shift in turn has had a small role in the
overall cleanup of U.S. manufacturing. Technology has played by far the most important
role in cleaning up U.S. manufacturing pollution.

Part 1: Technology - An Indirect Assessment of the Technique Effect

Between 1972 and 2002, total air pollution from U.S. manufacturing declined by
approximately 60 percent. In this part of the paper, I show that changing production
techniques account for the largest share of this cleanup, but that this share has been
declining over time.

Scale, Composition, and Technique in Theory

Environmental economists now have a convention for thinking about changes in
total pollution as coming from three sources: the overall size of the economy ("scale"),
the mix of sectors comprising the economy ("composition"), and the technologies
employed in production and abatement ("technique").2 Mathematically, the total amount
of pollution from manufacturing, (P), can be written as the sum of pollution from each of
its component industries, (pi). This in turn can be written as the total value shipped from
manufacturing, (J7), multiplied by the sum of each industry's share of total output, (0, =

2 See, for example, Grossman and Krueger (1993) or Copeland and Taylor (2005).

-------
Vj/V), times an emissions coefficient that reflects the amount of pollution per dollar of
value shipped in that industry, (zi=pjvl).

P = HPr =HVrZr =VH0rZr

(1)

Or, in vector notation

P = VB'z

(2)

where Pisa scalar representing the total pollution from manufacturing, and 0 and z are
nx 1 vectors containing the market shares of each of the n industries and their pollution
intensities, respectively. Totally differentiating equation (2) yields3

The first term in equation (3) is the scale effect, which explains what happens to
total pollution as the overall size of the manufacturing sector increases, holding the
composition of industries and their pollution intensities fixed. The second term is the
composition effect, which accounts for the changing mix of industries, holding their scale
and pollution intensities constant. And the third term is the technique effect, which
captures changes in pollution intensities, holding the scale and composition of
manufacturing fixed.4

Data on total pollution, (P), are taken from the National Emissions Inventory
(NEI). The NEI is the U.S. EPA's clearinghouse for the wide variety of pollution data
compiled by states and industries. It includes point, mobile, and area sources of pollution,
including individual facility-level data for large point sources. The NEI reports emissions
of four common air pollutants (known as "criteria" pollutants) back to 1970: SO2, NOx,
carbon monoxide (CO), and volatile organic compounds (VOCs; a precursor to ozone, or

3	Equation (3) assumes no interaaction terms - that changing the scale of manufacturing does not affect the
pollution intensities for example. Or, equivalently, we can think of equation (3) as assuming that all of
those interaction terms are combined into what I am calling technology, the third term which will just be a
remainder after accounting for the first and second terms.

4	Note that this technique effect could result from movement along a production possibility frontier (using a
different mix of inputs within known production technologies), or a shift outward in a production
possibility frontier (inventing new technologies). See Fare et al. (1989 and 2001).

dP = Q'zdV + Vz'dQ + VQ'dz

(3)

4

-------
"smog"). To be consistent with the international trade data used in Part 2 below, I limit
the analysis here to the period from 1972 to 2001.

Unfortunately, the NEI does not simply report the amount emitted by
manufacturing alone. Instead, its emissions data are organized by "activity" (such as fuel
combustion and transportation, among others). Table 1 describes the 14 categories of
activities in the NEI. Those marked with an asterisk in the table are included here as an
approximation of manufacturing emissions.

For data on total manufacturing output and the output of each industry (V and i',),
I use the manufacturing productivity database (NBER-CES; Bartelsman and Gray 1996).
These are derived from the Annual Survey of Manufactures conducted by the U.S.
Census Bureau.

Data on pollution intensities, (z), are the final element. For these, I rely on the
U.S. EPA's Trade and Environmental Assessment Model (TEAM), which has as its core
a list of emissions intensities by six-digit North American Industry Classification System
(NAICS) codes. These data were assembled by the U.S. EPA and Abt Associates (2004)
to assess the environmental effect of economic changes, such as those that might arise
from international trade agreements. TEAM uses the raw inputs to the 1997 NEI, first
matching emissions to individual facilities and then aggregating across the six-digit
NAICS codes to which those facilities belong. TEAM can be used to generate emissions
factors (environmental consequences per dollar of output) for 1,099 six-digit NAICS
industry codes, and for more than 1,000 different environmental outcomes, including air
pollutants, individual toxic chemicals, hazardous waste, and land use.5 Here I focus on
the 473 six-digit NAICS codes that comprise the manufacturing sector and on the four
criteria air pollutants reported by the NEI consistently since 1970.

Estimating how much of the pollution reduction over the past 30 years can be
attributed to each of the three effects in equation (3) requires annual data on total
pollution, (P), total output, (V), and each industry's contribution to output, ((),). The one
element of equation (3) that is not available by year is z, each industry's emissions

5 TEAM also disaggregates these emissions factors geographically (by U.S. county), although I am relying
on national averages for this study.

-------
intensity. I have data on z for only one year, 1997, and so I calculate the first and second
terms in equation (3) and calculate the third term as the remainder.

At this point, it is worth raising a few conceptual data issues - the NEI
approximation of manufacturing pollution, outsourcing electricity generation, changing
industry definitions, and changing relative product prices.

The NEI Manufacturing Approximation

The NEI categorizes pollution by "activity code", as listed in Table 1, and no
combination of those codes exactly matches the manufacturing sector. The best I can do
is to approximate total manufacturing pollution using the five activity codes marked with
asterisks in Table 1. This means that the definition of "manufacturing" in the TEAM data
(used to construct z) will differ from the definition in the NEI data (used to construct P).
Fortunately, all I am really concerned about is explaining pollution changes in percentage
terms (with units being irrelevant to the decomposition). For this reason, any mismatch
between the TEAM and NEI data will not be a problem as long as the ratio of the NEI
approximation to the true pollution from manufacturing remains constant over time. If
true manufacturing pollution grows relative to this NEI approximation, I will be
increasingly understating pollution from manufacturing and exaggerating the role of
technology in abating pollution. More likely, if true manufacturing pollution shrinks
relative to the NEI approximation, I will be increasingly overstating manufacturing
pollution and understating the role of technology.

For some recent years (1990 and the period from 1996 to 2001), the U.S. EPA
documented pollution by two-digit Standard Industrial Classification (SIC) codes,
making it possible to see whether the ratio of true manufacturing pollution to the NEI
approximation changed over time, if only for a short portion of the entire three-decade
time span. Table 2 presents this ratio, the sum of pollution from all manufacturing SIC
codes (20 through 39), divided by the total manufacturing pollution from the NEI
approximated by the five starred activity codes in Table 1. The ratios remain remarkably
stable. None display marked upward trends. If anything, the overall trend is downward,
suggesting that the NEI approximation may increasingly overstate pollution from
manufacturing, If so, my decomposition analysis will understate the role of technology in
abating that pollution.

-------
Outsourcing Electricity Generation

A second potential concern about this analysis is that much of the pollution being
studied is emitted as a byproduct of energy generation. If over time manufacturers
increasingly purchase electricity from utilities, rather than generating power themselves,
pollution may simply be moving from the manufacturing sector to the utility sector. Any
cleanup we perceive in manufacturing may be the spurious result of outsourcing
electricity generation.

To see whether manufacturers are increasingly purchasing electricity from offsite
generators, I examined data from the Manufacturing Energy Consumption Survey
(MECS), which has been conducted every third or fourth year since 1985. Net purchases
of electricity by manufacturers has remained constant at between 22 and 24 percent over
the past two decades (see Table 3). While I cannot say for sure that increased outsourcing
of electricity generation does not explain some of the declines in manufacturing pollution
prior to 1985, outsourcing has not been responsible for the declines since 1985.

Changing Industry Definitions

The U.S. manufacturing data (and the international trade data used in Part 2) are
organized by four-digit SIC codes that were defined in 1987. The TEAM data use the
NAICS codes as defined in 1997. Each is a hierarchical numerical classification of
industries, with similar industries grouped into separate classifications.

To match the SIC and NAICS industry classifications, I rely on a Census Bureau
publication of the 1997 industry-level aggregates using both the SIC and NAICS
classifications (U.S. Census Bureau 2000).6 From these 1997 data I constructed a
concordance, or "crosswalk," between the 1987 SIC codes and the 1997 NAICS codes.
The concordance reports the fraction of the output of each four-digit SIC code that is
attributable to each six-digit NAICS code and vice versa.7

6 Electronic versions of this publication can be found at www.census.gov/epcd/ec97brdg.

7 For some industries, the Census Bureau withholds the value shipped to avoid disclosing confidential
business information. In those cases the share of establishments serves as a proxy for value shipped. I
assume that the share of establishments equals the share of value shipped for industries where value
shipped is undisclosed. Within industry groups (the NAICS four-digit codes), I then subtract the sum of
value shipped for the reporting industries to obtain the residual undisclosed amount. Next, I apply the
proportions calculated from establishment numbers to the residual undisclosed amount to estimate the value
shipped for each undisclosed industry.

-------
The final data set includes only those observations that are defined as
manufacturing in both data sets: SIC codes 2011 through 3999 and NAICS codes 311111
through 339999. This eliminates cases where the industry redefinition changed whether
or not an industry was included in the manufacturing sector. It contains crosswalks
between 453 four-digit SIC codes in 1987 and 469 six-digit NAICS codes in 1997. Of
these, the SIC and NAICS codes matched perfectly in 229 cases, where the
reclassification merely relabeled the industry.

Changing Relative Product Prices

Comparisons across three decades necessitate adjusting for price inflation, which
would be a simple task if prices of all goods changed proportionately. But in some
industries, prices rose faster than the average for the manufacturing sector. Petroleum
prices, for example, grew faster than the producer price index (PPI) from 1972 to 2001
(508 percent for petroleum refineries [SIC 2911] compared to 237 percent for the PPI).
Manufacturing expensive fuel, though, does not pollute more than manufacturing cheap
fuel. If we divide each industry's output by the overall PPI, and then multiply by its
emissions coefficient to get predicted pollution, we will overstate the growth of pollution
from petroleum and exaggerate the role of technology in abating that pollution. This line
of reasoning would suggest using industry-specific price deflators.

On the other hand, many industries have seen spectacular falls in their relative
prices resulting from changes in the natures of their underlying products. The PPI for
computers (SIC 3571) fell 99 percent from 1972 to 2001. If we take that literally, and
adjust each industry's value shipped by an industry-specific measure of inflation, we
would have to multiply the value of computer production by 100. But manufacturing
computers with faster processors does not necessarily pollute more than manufacturing
slow computers. If we divide each industry's output by an industry-specific price index,
then multiply by its emissions coefficient, we will vastly overstate the growth of pollution
from computer manufacturing. Because high-tech industries that have experienced these
types of price deflations tend to be less polluting, using their industry-specific price
indices will overstate the "green" shift in the composition of U.S. manufacturing output

-------
toward clean products and understate the role of technology. This line of reasoning would
suggest using an economy-wide price deflator such as the PPI.8

In the energy industries, prices rose because of factors unrelated to the nature of
the product. Deflating output using the PPI, then, would overstate the growth of energy
industries. Since energy industries are relatively pollution intensive, using the PPI would
exaggerate the predicted amount of pollution based on the scale and composition of
manufacturing, in turn exaggerating the technique effect. Notably, this method would
also exaggerate the growth of energy-intensive industries from 1970 through the mid-
1980s - when energy prices rose fastest - and understate their growth after 1985. This
would lead to overstating the technique effect before the mid-1980s and understating it
thereafter. The end result would overstate the slowdown of technology's contribution to
pollution abatement.

In the computer and electronic industries, the nature of the products changed, and
BEA economists have calculated an implicit price decline. If we adjust nominal computer
sales using the BEA index for computers, we inflate that relatively clean industry's
output, overstate its share in predicted overall manufacturing pollution, and understate the
technique effect.

To be complete, I have calculated results using both the PPI and industry-specific
price deflators, but I focus on the industry-specific analysis, keeping in mind throughout
that this will likely overstate the composition effect relative to the technique effect. Given
that the bottom-line results of this first part of the analysis are that (1) the technique effect
dominates and (2) technique's dominance has declined, using the industry-specific
deflators represents the more conservative of the two approaches.

Scale, Composition, and Technique in Practice

Figure 1 illustrates the analysis for the sulfur dioxide (S02) emissions. (Figures
for the other pollutants look similar.) The top line - line number 1 - simply reports the
total value of manufacturing shipments, scaled so that the 1972 value equals 100. If the
mix of industries making up the manufacturing sector remained constant (d6 = 0), and the

8 Some analysts have even suggested that computer-related industries should use a constant deflator
(Meade 2000).

-------
techniques of production remained constant (dz = 0), this top line would represent how
emissions would have changed over time. The manufacturing sector grew by 71 percent
over this period. This is the scale effect.9

The bottom line - line 2 - plots S02 emissions by the manufacturing sector, as
reported by the NEI and scaled so that 1972 emissions equal 100. Emissions drop
steadily, and 2001 S02 emissions are 66 percent below their 1972 levels. This represents
the combined scale, composition, and technique effects, or (dP) in equation (3).

The middle line in Figure 1 - line 3 - is the result of multiplying each industry's
value of shipments in each year (yit) by the TEAM emissions coefficient for S02 in 1997
(zi) and aggregating across industries. It represents what S02 emissions would be in each
year if each separate manufacturing industry produced its concurrent output, but used the
production technique that generated the same amount of S02 per dollar of output that it
did in 1997. This combines the scale and composition effects, which grew by 19 percent
from 1972 to 2001.

These three numbers (scale up 71 percent, S02 down 66 percent, scale and
composition up 19 percent) are listed in the first three columns of Table 4. Columns (4)
through (6) contain the share of the total cleanup that can be attributed, respectively, to
the scale, composition, and technique effects. Manufacturing grew by 71 percent and
S02 emissions fell 66 percent. Therefore the scale effect more than completely offsets
the decline in pollution (or minus 107) percent of the decline S02 in emissions can be
attributed to changes in the scale of manufacturing).

The composition effect is simply the difference between lines (1) and (3) in
Figure 1. Scale added 71 percent to S02 emissions and scale and composition together
added 19 percent. As a result, the composition effect alone amounts to a 52 percent drop
in S02 emissions (relative to the 1972 baseline). This reduction accounts for 79 percent
of the total emissions decline of 66 percent.

Finally, the technique effect is simply the difference between lines (2) and (3) in
Figure 1. Scale, composition, and technique together result in an S02 emissions

9 This calculation is based on the summation of the value of shipments of six-digit NAICS codes, where
each industry's value shipped is indexed using an industry-specific price deflator. If instead I aggregate
across industries before applying an economy-wide PPI, manufacturing output grows by 55 percent.

-------
reduction of 66 percent. Scale and composition alone result in an increase of 19 percent.
Therefore the technique effect alone must amount to a 85 percent drop in emissions, or
128 percent of the total emissions decline of 66 percent. Changes in technique from 1972
to 2001 have reduced pollution from the 2001 manufacturing sector by more than the
total amount of pollution the sector emitted in 1972.

As a summary statistic for these calculations, Column (7) in Table 4 reports the
ratio of the technique effect to the composition effect, 1.62. This means that the S02
emissions reductions resulting from changing production technologies are more than 1.5
times as large as the emissions reductions resulting from the changing mix of industries
that comprise manufacturing.

Table 4 also presents this same calculation for the other three criteria air
pollutants individually. Even though we can see differences based on the patterns of
pollution emissions and growth across industries, the overall result is consistent. In each
case, the technique effect (Column (6)) accounts for more than 100 percent of the
cleanup. For VOCs, the composition effect is small, and as a consequence the ratio of
technique to composition is large (2.71). But in absolute terms the technique effect in
Column (6) is approximately the same size as for the other pollutants. The bottom row of
Table 4 contains the same set of calculations for the unweighted sum of all four
pollutants. Overall, changing technology reduced air pollution from manufacturing by
1.44 times as much as changing the composition of the manufacturing sector.

Table 5 presents the ratio of technique to composition for some alternative
versions of these calculations. Column (1) contains the baseline for comparison,
repeating Column (7) of Table 4. In Column (2) of Table 5,1 conducted the same
exercise, but used the PPI to deflate industry output instead of using industry-specific
price deflators. As expected, this greatly increases the estimate of technique relative to
composition, mostly by exaggerating the growth of pollution from energy-intensive
industries.

In column (3) I address the opposite concern, that the industry-specific index for
computers overstates their importance in the cleanup of manufacturing. I conduct the
entire analysis without the eight SIC codes where the industry-specific price index falls

-------
between 1972 and 2001.10 Deflating output from those eight industries suggests their
output grew by more than as measured by nominal prices, and if they do not pollute much
may overstate their role in the composition effect on pollution. As a consequence,
dropping these eight industries results in a large estimate of the technique effect. The
ratios of technique to composition effect in column (3) suggest the corresponding
calculations in column (1) should be interpreted as lower bounds on the role of
technology. Those baseline numbers inflate the role of composition change by
exaggerating the shift towards relatively non-polluting high-tech industries. The baseline
calculation in column (1) thus represents a conservative estimate of the role of
technology in reducing pollution, because it uses industry-specific price deflators which
exaggerate the shift towards relatively clean computers and electronics industries.

Columns (4) and (5) describe the change in this ratio over time. Looking back at
Figure 1, the technique effect is the difference between the bottom line (scale,
composition, and technique) and the middle line (scale and composition). The gap
appears to grow quickly during the first decade or so and then slow down, suggesting that
the technique effect has diminished in importance over time. To document this, Table 5
reports the ratios of technique to composition for two separate time periods: 1972-1985
and 1985-2001. In each case, the ratio of technique to composition is smaller during later
years. Technique appears to have played a larger role than composition in the cleanup of
U.S. manufacturing, but the ratio falls over time.

All four of the air pollutants show a marked decline in the technique effect over
the three decades since the 1970s. There may be several reasons for this. First, marginal
abatement costs could be increasing, and all of the least expensive abatement
technologies might have been employed in the response to the 1970 and 1977 Clean Air
Acts.11 However, it seems equally likely that there are increasing marginal costs

10 These are electronic computers (3571), computer storage devices (3572), computer terminals (3575),
computer equipment n.e.c. (3577), calculating and accounting machines (3578), household audio and video
equipment (3651), semiconductors (3674), and magnetic and optical media (3695).

11 Cost-minimizing polluters will employ abatement technologies in the order of increasing marginal cost,
which means that as more is done, abatement becomes more costly. If polluters were mandated to reduce
pollution, but left on their own to choose the method (output reduction, international outsourcing, or
technological abatement), we might expect to see the pattern depicted in Table 4. Of course, polluters were
not free to choose because technologies were mandated by the 1970, 1977, and 1990 Clean Air Acts. Those
laws, however, may reflect the underlying realities of increasing marginal cost and mandate successively
smaller increments in abatement.

-------
associated with altering the composition of U.S. manufacturing or relocating industries
abroad. Some industries have inelastic demand and are immobile, and their share of
output is unlikely to decline. Others have elastic demand or are mobile, and U.S.
production should decline in the face of increasing costs. Whether increasing marginal
costs of technological abatement are larger or smaller than the increasing marginal costs
of compositional change is an open question. It is not obvious a priori that the ratio of
technique to composition effects will necessarily fall over time.

A second explanation for the falling ratio of technique to composition is that
regulators may have demanded successively smaller increases in abatement technology.
If there is a trend away from U.S. production of polluting goods that is unrelated to
regulations, that effect will come to dominate over time as technological advances shrink.

Third, and perhaps most likely, the diminishing importance of technology in
abating polluting may be due to the fact that energy prices increased sharply during the
1970s and early 1980s, and decreased thereafter. High energy costs gave manufacturers
incentives to increase energy efficiency, which also reduced pollution. The high energy
costs were experienced world-wide, and could not be avoided by relocating overseas.
Hence in the 1970s and early 1980s we see technique dominating composition. Once
energy costs start to fall in the mid-1980s, energy efficiency improvements slowed down,
and composition changes took on greater importance.

Before drawing general conclusions about the analysis so far in Part 1, one
important caveat deserves mention: intra-industry composition.

One final caveat: Intra-industry composition

The analyses in tables 4 and 5 rely on a decomposition of the U.S. manufacturing
sector into 469 six-digit NAICS industry codes. These codes represent a relatively fine
categorization of industries. Twenty-five different six-digit NAICS codes comprise the
primary metals industries, ranging from iron foundries to rolled steel shape
manufacturers. Eighteen codes make up paper manufacturing, ranging from pulp mills to
envelope manufacturers. Despite the level of disaggregation, a concern remains - that
heterogeneity within six-digit NAICS codes may be driving what I describe here as a
technique effect. If six-digit NAICS industries contain subindustries of varying pollution
intensity, and the composition of individual industries has shifted toward less-polluting
subindustries over time, some of what I have described as a technique effect may actually
be another form of composition effect, at a level of disaggregation too fine to see with the

-------
six-digit NAICS classifications. However, composition changes within NAICS codes
would have to be implausibly large relative to those across NAICS codes to overturn the
basic result of this analysis - that technology accounts for more pollution reduction than
composition change.

Part 1 Conclusion: The Role of Technology in Pollution Abatement

Even being cautious about the role of relative price indices and the possibility of
intra-industry composition effects, the implications of the simple decomposition here are
stark. Air pollution from manufacturing has declined significantly in the United States
despite increases in manufactured output. Most of this improvement seems to arise from
changes in the way goods are produced (technique), rather than the types of goods
produced (composition), although the role of technique may be waning.

A substantial composition effect does remain, however. Changes in the mix of
industries have reduced air pollution from manufacturing by as much as 36 to 60 percent
relative to the amount in 1972.12 And that composition effect grows in importance as the
technique effect fades. This change in manufacturing composition could itself come from
one of two sources - changes in consumption or changes in net imports. If the change
comes from imports, important concerns arise for the environment in the origin countries,
especially the least developed countries. In the future, these countries will not be able to
repeat U.S. success in abating pollution without even less-developed countries from
which to import polluting goods. In Part 2,1 ask explicitly how much of the pollution
reductions from composition change can possibly be explained by the increase in net
imports of pollution-intensive manufactured goods.

Part 2: International Trade - An Input-Output Approach to Measuring Embodied
Pollution

Part 1 shows that although most of the pollution reduction from U.S.
manufacturing has come from technology, a significant and growing part has also come
from shifting over time toward production of less-polluting goods. A logical next
question to ask is how much this green shift in U.S manufacturing can be explained by
increasing imports of more-polluting goods. For comparison with Part 1,1 continue using

12 This is the difference between Columns (2) and (3) in Table 4.

-------
the four pollutants studied earlier. But because I do not need emissions data back to 1972
for this part of the paper (I only need the industry-specific emissions coefficients (z) for
one year), I expand the analysis to study 9 different air, water, and toxic pollutants.

A number of papers have examined this issue, but the standard approach has had
several shortcomings. First, all the papers to date use the World Bank's 1987 Industrial
Pollution Projection System (IPPS), which is now two decades old.13 If environmental
regulations and technological progress have succeeded in reducing the pollution emitted
per unit of output from the dirtiest industries, using the IPPS exaggerates the degree to
which the change in the composition of the U.S. manufacturing sector has reduced
pollution.14 Moreover, the IPPS index is based on pollution data from 1987 that have
since been refined. For example, one input to the IPPS is the Toxics Release Inventory,
for which 1987 was the first year of data collection, with low participation rates by
industry. I use the TEAM data described in Part 1, which is based on emissions and
output data from 1997. These more-recent data are less likely to exaggerate the green
shift of U.S. manufacturing or imports.

More importantly, papers to date on this topic have focused only on the pollution
content generated by each industry directly, ignoring the pollution generated by the
manufacture of intermediate inputs to those industries. This is appropriate when
documenting changes in U.S. manufacturing, where intermediate good production is
counted separately. But for imports, where only the final good is observed, omitting
pollution from intermediate goods can significantly understate the potential role of trade
in the green shift of U.S. manufacturing. To address this factor content issue, I use the
BEA's benchmark input-output tables for 1997, along with a Leontief-style input-output
model to adjust the TEAM emissions content for the total pollution displaced by imports,
including their intermediate inputs.

13 The IPPS is described inHettige et al. (1995). Previous studies using the IPPS include Kahn (2003),
Schatan (2003), Cole (2004), Ederington et al. (2004), and Gamper-Rabindran (2006).

14 This is essentially an environmental version of the standard Laspeyres-Paasche price index problem.
Using an index (prices or pollution intensities) from earlier in the time beriod overstates the change (price
inflation or the composition effect). Using an index from later in the time period overstates the change.
Thus by using the 1997 TEAM data to measure pollution intensities, I overstate the relative cotribution of
composition change to reducing pollution, and provide a conservative estimate of the importance of
technolgy.

-------
I conduct two straightforward exercises. First, I show that despite the concerns of
trade agreement opponents and antiglobalization protesters, the pollution content of
imports to the United States has become cleaner over time, not dirtier, even accounting
for pollution from intermediate inputs to imports. In fact, imports have been getting
cleaner faster than domestic production. Second, in a simple thought experiment, I offset
the effect of trade on U.S. manufacturing by adding all increases in imports over the past
30 years back into U.S. production, and subtracting all increases in exports. I compare the
pollution that would have been generated by this hypothetical "no-trade-growth" U.S.
manufacturing base to that generated by the actual U.S. manufacturing base. Despite
enormous growth in the scale of imports, trade increases have not been large enough to
account for the green shift of U.S. manufacturing. For most pollutants, increased trade is
sufficient to explain no more than about half of the pollution reductions from the
composition change of U.S. manufacturing.

The Composition of Trade

The concerns voiced about the relationship between pollution and trade have a
common theme - that industries will relocate outside the United States to save on
environmental compliance costs. If this has been happening, over time we should see
relatively less manufacturing of highly polluting goods in the United States and relatively
more net imports of those goods from overseas. Part 1 documented the green shift of U.S.
manufacturing. Here, I apply a similar approach to imports.

Let P" be the amount of domestic pollution displaced by imports, by which I
mean the amount of pollution that would have been emitted in the United States had
those imported goods been produced domestically. can be written as the sum of the
pollution displaced by each industry,/?/1^ which in turn can be written as the total value
of imports, times the sum of each industry's share in that total, (&1 = i'/l/ Vu), times
each industry's emissions per dollar of shipments in the United States, (zi = pi/vi).

p" =Y.p",=v"\

-------
This is a direct analog to equation (1), which applies to U.S. domestic production. Data
on and 0U come from the Center for International Data (Feenstra 1996, 1997), and
have been translated from SIC codes to NAICS codes using the concordance described in
Part l.15 Note that Pu does not tell us the amount of pollution occurring overseas as a
consequence of producing goods for the United States, because other countries
presumably have different emissions coefficients (z). Rather, Pu is the amount of
pollution that would have been emitted in the United States had those imports been
produced domestically.

How much of the green shift of U.S. manufacturing can be explained by the
pollution displaced by imports, as measured by equation (4)? Figure 2 begins to illustrate
the analysis for S02 emissions. The top line in Figure 2 is an index (1972 = 100) of real
U.S. manufacturing imports, which grew 641 percent from 1972 to 2001. This is
analogous to the scale effect of manufacturing. If imported goods contain the same mix
of industries as domestic production, we could say that the amount of U.S. pollution
displaced by imports grew 641 percent.

Of course, the composition of imported goods has also been changing over time.
Multiplying each industry's imports by its 1997 S02 emissions coefficient and then
aggregating across industries for each year generates the bottom line depicted in Figure 2.
S02 emissions displaced by imports would grow by 146 percent. Or, put differently, S02
emissions displaced by imports were 67 percent lower than the pure scale effect of
imports would predict.16 Column (2) of Table 6 contains these calculations of the
composition effect of imports for nine individual pollutants. The composition effects
range from 61 percent for VOCs to 82 percent for biological oxygen demand (BOD), a
measure of water pollution.17

Column (1) of Table 6 contains these same calculations for the effects of scale
and composition on U.S. pollution, using equation (1). These effects are much smaller

15 The Center for International Data at the University of California Davis (UCD) can be found at
http://cid.econ.ucdavis.edu. The data run from 1972 to 2001.

16 This calculation is (1 + 1.46)/(1 + 6.41) - 1.

171 could, in theory, apply this analysis to any of the more than 1,000 chemicals documented in the TEAM
data. Here I use the criteria pollutants, for comparison with Part 1, along with two common measures of
water pollution (BOD and total suspended solids [TSS]) and two aggregations of toxic chemicals (released
to air and released to water).

-------
than the calculations for imports in Column (2). The changing composition of U.S.
manufacturing reduced the benchmark four criteria air pollutants by 30 percent; the
changing composition of imports (by this calculation) reduced the pollution displaced by
those imports by 66 percent. This begins to suggest that imports have been shifting
toward cleaner goods faster than domestically produced goods, and that increased trade
may not account for (or even contribute to) the green shift of U.S. manufacturing. And
here is where the previous literature stops.

There are two problems, however, with stopping at this point and merely pointing
out that the composition effect as measured using the simple emissions coefficients tilts
imports toward cleaner industries. First, the simple emissions coefficients neglect the
intermediate inputs to those imports. Second, focusing solely on the composition effect
ignores the fact that imports grew so much more in absolute terms than domestic
production. I address each in order.

Recent prior work has ignored pollution from intermediate inputs. A simple
example may help to explain the problem. Suppose that in an initial time period the
United States produces automobiles. Each car requires one ton of steel as an input, and
steel is entirely produced domestically. In the second time period, the United States
imports one more car and produces one fewer. How much of the decline in U.S. pollution
can be accounted for by the increase in imports? In the example, we can account for 100
percent by construction. But if we simply multiply the change in imports by the
respective direct emissions coefficients (the calculation in equation (4) and Figure 2), we
understate the pollution displaced by imports because there are no steel imports. The
change in steel production occurs abroad and is embedded in the car. Using the direct
emissions coefficients ignores the pollution generated by intermediate inputs to imports,
and makes it appear as though import composition is shifting toward cleaner goods faster
than it is in reality. The direct emissions coefficients therefore understate the amount of
U.S. pollution reduction that is merely the consequence of increased imports.

To correct this, we need to account for not only the pollution embodied in the
intermediate inputs to imports, but also the pollution embodied in the intermediate inputs
to those intermediate inputs, and so on ad infinitum. (The steel used to make cars itself

-------
requires inputs that may produce pollution, and so on.) For this calculation, I rely on a
basic Leontief input-output framework.18

Suppose that xt represents the total output of sector including intermediate
inputs to other industries and final output to either consumption or export. The total
dollar amount of good i required directly in the production of one dollar's worth of good j
is Cjj. Final output is>7. Total output, x, which is a vector of n outputs - one from each
industry - is the sum of output used as intermediate goods and final output.

X,"

~xi~

Xn_

c , ••• c

n\ nn _

Xn_

Or, in vector notation:

x = Cx + y

where C is an n x n matrix of direct requirements coefficients with elements c,,
representing the dollar value of input industry i needed to produce one dollar's worth of
output industry j.

When we examine U.S. production, we are observing x, the value of all
shipments, including both intermediate inputs and final products. We can appropriately
estimate pollution, then, by multiplying x by a vector of direct emissions coefficients z,
such as those from TEAM. But when we examine imports, we see only final product y,
without all of the intermediate production. In this case, we need a set of total pollution
coefficients. These coefficients must embody all the pollution generated by all of the
inputs to y, all the inputs to those inputs, and so on. To calculate this, we can solve
equation (6) for x to get

x = [I-C]1y

where I is the identity matrix. The matrix T = [I - C]-1 is the Leontief total requirements
matrix. Each element % contains the dollar amount of industry i necessary to produce one
dollar of output industry j, including the amount of i used in all other industries that are
used in j, as well as the amount of i used in the inputs to those industries, and so forth.

18 See, for example, Miller and Blair (1985).

-------
The vector x represents the total amount of manufactured goods necessary to produce
output y. To generate the total pollution coefficients, I simply premultiply the Leontief
total requirements matrix by the z vector from TEAM as follows:

z = z'T = z'[l-C]1

The only new piece of information we need to construct z is C, the matrix of
direct requirements coefficients. The BEA publishes an input-output table for the United
States, and I use the 1997 version to create the matrix C for the manufacturing sector.
The BEA tables are organized by commodity rather than industry, but for the
manufacturing sector, commodity codes mostly map one to one into NAICS industry
codes. For those that do not, I aggregate up to the level of five-digit NAICS codes, and,
in 13 cases, to four-digit NAICS codes. The resulting C matrix is 344 x 344.

Using the total emissions coefficients, z ^ in place of the direct emissions
coefficients, z, captures all of the pollution generated by intermediate goods, and does not
understate displaced pollution. There is, however, a further complication. If the steel used
in the production of automobiles in the United States is entirely imported, importing an
automobile from abroad displaces no U.S. steel pollution, and the appropriate emissions
coefficient is the direct one (z). Suppose, however, that 10 percent of the steel used in
U.S. automobile production is imported in the first period. If the U.S. imports one car in
the second period, U.S. steel production declines by 0.9 tons and steel imports decline by
0.1 tons. Pollution in the United States declines by the amount emitted from
manufacturing one automobile and 0.9 tons of steel.

The only way to solve this problem is to adjust the pollution coefficients to
account for the imported fraction. We must multiply the direct requirements coefficients
by the fractions of goods in each industry that are produced domestically. In other words,
we need to replace the C matrix in the inverse Leontief calculation with diag{d)C, where
d is an n x 1 vector whose elements are the share of each industry supplied by domestic
production.19

19 The domestic share of supply is defined as 1 - imports/(domestic production + imports - exports).

-------
z* = z' [I - diag{A)C\

(9)

These emissions coefficients, z*, are total domestic requirements emissions coefficients.20

Figure 2 demonstrates the effect of this adjustment. The bottom line depicts the
amount of U.S. pollution displaced by imports, using the direct emissions coefficients
that fail to account for intermediate goods (z). The middle line uses the total domestic
requirements emissions coefficients (z* in place of z). Multiplying each industry's imports
by its Zi, and then aggregating across industries for each year, shows us that S02
emissions displaced by imports would grow by 385 percent. If we use z* instead of z, we
see that emissions displaced by imports are 44 percent lower than the 641 percent
increase that would have been predicted by the growth in the scale of imports alone.

Table 6 contains this calculation - the composition shift of imported goods using
the total domestic emissions coefficient (Column 3) - along with similar calculations for
each of nine individual pollutants. The numbers range from 28 percent for CO to 81
percent for BOD. In every case, the displaced pollution estimated using the total domestic
requirements coefficients is larger than the estimate obtained using the direct
requirements coefficients (because the coefficients are by definition larger).

Consequently, the estimated green shift of imports is smaller.

U.S. imports are dominated by trade with other industrialized countries, but
pollution haven worries focus on developing countries. To address this, column (4) of
Table 6 presents the same calculation for imports from outside of the OECD. It turns out,
however, that the green shift in imports from non-OECD countries toward cleaner goods
is approximately as large as the green shift in imports in general, and typically much
larger than the green shift in U.S. manufacturing.

Adjusting for factor content diminishes the previous result (that imports shifted
toward cleaner goods faster than goods manufactured domestically), but does not

20 Note that this assumes that the fraction of any input that is imported is the same, regardless of which
industry uses it.

-------
eliminate it. Most of the import growth seems to have come from industries that were not
pollution intensive in the United States, even after accounting for pollution from
intermediate inputs. Even though previous studies - in which analysts used direct
requirements coefficients - got the magnitude wrong, the basic finding remains valid.
The U.S. manufacturing sector has shifted away from polluting goods, and imports have
shifted even further away from polluting goods. The green shift of U.S. manufacturing
has not been accompanied by a corresponding "brown" shift in imports to the United
States. Instead, imports have been shifting toward cleaner goods faster than those
produced domestically.

Results like these have been interpreted as evidence that U.S. environmental
policies are not pushing U.S. polluting manufacturers overseas, but that conclusion does
not necessarily follow. We do not know what the composition of imports would have
been without changes in U.S. environmental regulations. Perhaps imports would have
shifted toward less-polluting goods even faster. The best we can do with these data is ask
whether the overall size of import growth is sufficient to account for the pollution
reductions resulting from the green shift of U.S. manufacturing. If import growth is small
or composed of clean industries (with clean intermediate inputs), the composition-related
pollution reductions in the United States cannot possibly be explained by international
trade.

Whether international trade increases have been sufficient to account for the green
shift of U.S. manufacturing depends on both the composition and scale of imports, and
these effects work in opposite directions. Although imports became 28 to 81 percent
cleaner from 1972 to 2001, those imports increased by more than five times. This might
leave ample room for imports to replace pollution generated during domestic production
because the overall pollution content of imports (counting both the composition and
scale) has increased.

The Pollution Content (Scale and Composition) of Trade

Figure 1 illustrates the last step in this analysis, explaining the fraction of U.S.
emissions reduction that can be explained by changes in the combined scale and
composition of imports. Recall that the gap between lines (1) and (3) depicts the green
shift in U.S. manufacturing away from pollution-intensive goods. How much of this shift
can be explained by increased imports? If we multiply each industry's net imports
(imports minus exports) by the relevant total domestic requirements pollution

22

-------
coefficients, (z,*), we obtain an estimate of the amount of U.S. pollution displaced by net
imports. To show the change over time, use the difference between each year's net
imports and net imports in 1972. This series is plotted as line (4) in Figure 1. Holding
technology fixed as of 1997, this line represents what air pollution from manufacturing
would have been if every bit of increased net imports since 1972 had instead been
manufactured in the United States. As shown in the figure, adding back into U.S.
emissions the amount displaced by imports (using the total emissions factors that account
for pollution from intermediate goods) accounts for roughly half of the green shift of U.S.
manufacturing.21

I recognize, of course, that this "no-net-trade-growth" scenario represents an
extremely unlikely partial-equilibrium thought experiment. Had there been no net trade
growth, surely the mix of goods consumed by the U.S. economy would have been much
different. In no sense does the "without net imports" line in Figure 1 represent what the
pollution content of U.S. manufacturing actually would have been absent trade growth.
Instead, I think of the no-net-trade growth scenario as an accounting exercise - asking
what fraction of the composition-related cleanup of U.S. manufacturing can be matched
to increased imports of pollution-intensive products, not what pollution in the U.S. would
have been absent the ability to import those products. In fact, if the U.S. imports
pollution-intensive goods because they are less expensive when produced overseas, in the
absence of trade growth the United States would likely consume and produce fewer
pollution intensive goods, changing the baseline (pollution) and the overall scale
(manufacturing output).

Table 7 summarizes this calculation. Adding back pollution displaced by net
imports and accounting for intermediate goods accounts for 53 percent of this green shift
for S02 pollution (column (2)). Column (2) of Table 7 contains this same no-trade-
growth thought experiment for each of nine individual pollutants. For each of the
pollutants except CO, the net pollution embodied in the increased trade accounts for no
more than about half of the green shift of U.S. manufacturing. For CO, trade accounts for
99 percent of the change. For BOD, exports of pollution-intensive goods grew enough

21 Note that this no-trade-growth scenario exaggerates the role of trade, because it holds imports constant at
their 1972 levels in real terms, which means that imports' share of manufacturing would decline over time.
An intermediate case would hold the import growth rate equal to the growth of U.S. manufacturing.

-------
that the net effect of trade makes the U.S. industrial composition more pollution
intensive, rather than less.22

Table 7 contains two other sets of calculations for comparison. Column (1)
conducts the analysis using the direct requirement coefficients, failing to account for the
pollution content of intermediate inputs to imports. The amount of the green shift of U.S.
manufacturing explained by trade growth is much smaller (5 percent rather than 53
percent for S02), but of course this understates the pollution content of imports by
ignoring intermediate inputs. More significantly, Column (3) conducts the analysis for
trade with non-OECD countries, which are of most concern to those worried about
pollution havens. For most pollutants, trade growth with these non-OECD countries
appears to account for only about one-quarter of the green shift of U.S. manufacturing.

Displaced Pollution and the Pollution Havens Hypothesis

These analyses demonstrate that the increase in the pollution content of imported
goods is typically insufficient to explain the decline in U.S. manufacturing pollution
resulting from the changing composition of U.S. industries. When we import a product,
the pollution from its manufacture occurs abroad, not in the United States. I have focused
here on the resulting decline in U.S. pollution, not the increase in overseas pollution. The
results do not tell us what has happened to the environment in countries from which the
United States imports goods, only what pollution in the United States would have been
had those goods been produced at home - what I have called "displaced" pollution.

Furthermore, the analysis here implies no causality. I have not asked why the U.S.
composition changed nor why imports increased. I have only attempted to show that the
scale of imports is insufficient to account for the green shift of U.S. manufacturing, even
including the pollution caused by intermediate inputs to imports.

22 These outliers are explained by a relatively small number of industries. Each industry's effect on the total
depends on three items - the size of its imports and exports, the change in its imports and exports, and its
emissions coefficient (z,*). For CO, automobile imports alone grew by enough to displace 14 percent of
U.S. CO emissions in 1972. Adding those imports back into U.S. production explains a large fraction of the
decline in U.S. CO emissions. For BOD, pulp and paper exports alone grew enough to increase U.S.
emissions by 27 percent. Subtracting those exports would have made the U.S. manufacturing sector
significantly cleaner. No other industry-pollutant combination comes close to these two in terms of
explaining the results in Table 6.

-------
A large body of literature does address these causal relationships. In particular,
analysts have tried to assess the degree to which increasing environmental regulations in
the United States have caused either the green shift or the increase in imports. The results
of the analysis in this paper have no bearing on that literature. My finding that the
composition of imports has been shifting toward clean goods faster than the composition
of goods produced domestically does not mean that there is no pollution haven effect.
Perhaps the green shift of imports would have been larger and the green shift of U.S.
manufacturing smaller in the absence of U.S. environmental laws. The finding that the
pollution content of imports is sufficient to offset only about half of the pollution changes
resulting from the scale and composition of U.S. manufacturing does not mean that these
changes were themselves caused by U.S. environmental laws. The changes in industry
composition and imports are the result of many concurrent trends in addition to
environmental costs, such as changes in labor, energy, shipping, and tariff costs, among
others. Sorting out which costs drive changes in U.S. industrial composition and imports
is a job for another paper. This paper merely documents that fact that the growth and
composition of imports to the United States is sufficient to explain at most about half of
the pollution reductions achieved from producing a cleaner mix of goods at home.

Conclusion

Separating the decline in manufacturing emissions into its three components
(scale, composition, and technique) is important for several reasons. Most U.S.
environmental regulations have been designed explicitly to affect production
technologies, not to depress manufacturing or alter the mix of goods manufactured. And
most measures of the costs of environmental regulations focus on easily measured costs
of abatement technologies, not the diminished consumer or producer surplus from
reduced or relocated production. If pollution reductions result from changes in the overall
scale or composition of U.S. manufacturing, there could potentially be adverse
consequences. Environmental improvements could then be said to have imposed large,
unmeasured economic costs; to have imposed large changes in goods we consume; or to
have shifted pollution from the United States to other countries. Furthermore, none of
these changes would be replicable by all countries indefinitely. If the pollution reductions
come from technological progress, however, there is nothing suggesting that the trend
cannot continue indefinitely and be repeated around the world.

-------
The good news, then, is that most of the pollution reduction over the past 30 years
has come from changes in technology, rather than from changes in imports or changes in
the types of goods produced domestically. Criteria air pollutants collectively declined 58
percent from 1972 to 2001, despite a 71 percent increase in manufacturing output. The
cleanup was accomplished by changing the mix of goods produced and by altering the
technologies used to produce those goods. For a typical pollutant, technology accounts
for a large majority of the cleanup. Moreover, although some of the improvement is due
to the changing composition of industries, that change cannot be explained by increases
in imports. For the typical pollutant, increased international trade explains at most half of
the pollution reductions from composition changes in U.S. manufacturing. Those
composition changes in turn explain less than half of the overall reduction in U.S.
manufacturing pollution.

Together these findings suggest that the environmental concerns of
antiglobalization protesters have been overblown, and that the pollution reduction
achieved by U.S. manufacturing will replicable by other countries in the future. Most of
the environmental improvements in the United States have come from technology, not
from relocating polluting industries overseas. That good news must be tempered
somewhat by the fact that the role of technology appears to be shrinking, making
composition changes increasingly important to pollution reduction.

-------
References

Abt Associates, Inc. 2004. "Trade and Environmental Assessment Model: Model
Description." Prepared for U.S. Environmental Protection Agency (EPA),
National Center for Environmental Economics. Cambridge, MA: Abt Associates.

Ang, B.W. 1999. "Decomposition Methodology in Energy Demand and Environmental
Analysis." Pages 1146-1163 in Handbook of Environmental and Resource
Economics. Edited by J. van den Bergh. Northampton, MA: Edward Elgar.

Ang, B.W., and F.Q. Zhang. 2000. "A Survey of Index Decomposition Analysis in

Energy and Environmental Studies." Energy - The International Journal. 25(1):
1149-176.

Bartelsman, E.J., and W. Gray. 1996. "The NBER Manufacturing Productivity

Database." NBER Technical Working Paper No. 205. Cambridge, MA: National
Bureau of Economic Research (NBER).

Brunnermeier, S., and A. Levinson. 2004. "Examining the Evidence on Environmental
Regulations and Industry Location." Journal of the Environment and
Development. 13(1): 6-41.

Cole, M.A. 2004. "U.S. Environmental Load Displacement: Examining Consumption,
Regulations and the Role of NAFTA." Ecological Economics 48(4): 439-450.

Copeland, B., and M.S. Taylor. 2005. Trade and the Environment: Theory and Evidence.
Princeton, NJ: Princeton University Press.

Creason, J., M. Fisher, S. Semenova, and S.F. Stone. 2005. "The Environmental Impacts
of Trade Liberalization: A Quantitative Analysis for the United States Using
TEAM." Agricultural and Resource Economics Review 34(1): 90-103.

Ederington, J., A. Levinson, and J. Minier. 2004. "Trade Liberalization and Pollution
Havens." Advances in Economic Policy and Analysis 4(2): Article 6.

Available at: http://www.bepress.com/bejeap/advances/vol4/iss2/art6.

Fare, Rolf, Shawna Grosskopf, Knox Lovell, and Carl Pasurka. 1989. Multilateral
Productivity Comparisons when Some Outputs are Undesirable. Review of
Economics and Statistics. 71,No.l (February), 90-98.

27

-------
Fare, Rolf, Shawna Grosskopf, and Carl Pasurka, 2001, Accounting for air pollution
emissions in measures of state manufacturing productivity growth, Journal of
Regional Science, 41, 381-409.

Feenstra, R.C. 1996. "NBER Trade Database, Disk 1: U.S. Imports, 1972-1994: Data and
Concordances." NBER Working Paper No. 5515. Cambridge, MA: NBER.

	. 1997. "NBER Trade Database, Disk 3: U.S. Exports, 1972-1994, with State

Exports and Other U.S. Data." NBER Working Paper No. 5990. Cambridge, MA:
NBER.

Gamper-Rabindran, S. 2006. "NAFTA and the Environment: What Can the Data Tell
Us?" Economic Development and Cultural Change 54: 605-633.

Grossman, G.M., and A.B. Krueger. 1993. "Environmental Impacts of a North American
Free Trade Agreement." in The Mexico-U.S. Free Trade Agreement (edited by
P.M. Garber, Cambridge, MA: MIT Press) reprinted in Law and the Environment:
An Interdisciplinary Reader (edited by R.V. Percival and D.C. Alevizatos,
Philadelphia: Temple University Press, 1997).

Hettige, H., P. Martin, M. Singh, and D. Wheeler. 1995. "The Industrial Pollution
Projection System." World Bank Policy Research Working Paper No. 1421.
Washington, DC: World Bank Policy Research.

Hoekstra, R., and J. van den Bergh. 2002. "Structural Decomposition Analysis of

Physical Flows in the Economy." Environmental and Resource Economics 23(3):
357-378.

	. 2003. "Comparing Structural and Index Decomposition Analysis." Energy

Economics 25(1): 39-64.

Kahn, M.E. 2003. "The Geography of U.S. Pollution Intensive Trade: Evidence from
1959 to 1994." Regional Science and Urban Economics 33(4): 383-400.

Koo, A.Y.C. 1974. "Environmental Repercussions and Trade Theory." Review of
Economics and Statistics 56(2): 235-244.

Levinson, A., and M.S. Taylor. Forthcoming. "Unmasking the Pollution Haven Effect."
International Economic Review.

Leontief, W. 1970. "Environmental Repercussions and the Economic Structure: An
Input-Output Approach." Review of Economics and Statistics 52(3): 262-271.

28

-------
Leontief, W., and D. Ford. 1972. "Air Pollution and the Economic Structure: Empirical
Results of Input-Output Computations." Pages 9-29 in Input-Output Techniques.
Edited by A. Brody and A.P. Carter. New York: North-Holland.

Meade, D.S. 2000. "Information Workers in the 'New Economy': Has IT Investment Had
a Favorable Effect on the Demand for Skilled Labor?" Paper presented at the XIII
International Conference on Input-Output Techniques at the University of
Macerata, Italy, August 21-25.

Metcalf, G.E. 2007. "Energy Conservation in the United States: Understanding its Role in
Climate Policy" Tufts University Economics Department working paper.

Miller, R.E., and P.D. Blair. 1985. Input-Output Analysis: Foundations and Extensions.
Englewood Cliffs, NJ: Prentice-Hall.

Pasurka, C.A. 2003. "Changes in Emissions from U.S. Manufacturing: A Joint
Production Perspective." Washington, DC: U.S. EPA.

Rose, A. 1999. "Input-Output Structural Analysis of Energy and the Environment." Pages
1164-1179 in Handbook of Environmental and Resource Economics. Edited by J.
van den Bergh. Northampton, MA: Edward Elgar.

Schatan, C. 2003. "The Environmental Impact of Mexican Manufacturing Exports under
NAFTA." In Greening NAFTA. Edited by D. Markell and J. Knox. Stanford, CA:
Stanford University Press.

Turner, K., M. Lenzen, T. Wiedmann, and J. Barrett. 2007. "Examining the Global

Environmental Impact of Regional Consumption Activities - Part 1: A Technical
Note on Combining Input-Output and Ecological Footprint Analysis." Ecological
Economics 62(1): 37-44.

U.S. Census Bureau. 2000. "Bridge between NAICS and SIC 1997." EC97x-C53.
Washington, DC: U.S. Census Bureau.

U.S. EPA. 1998. "National Air Pollutant Emission Trends Procedures Document, 1900-
1996." EPA-454/R-98-008. Research Triangle Park, NC: U.S. EPA Office of Air
Quality Planning and Standards.

	. 2000. "National Air Pollutant Emission Trends, 1900-1998." EPA-454/R-00-

002. Research Triangle Park, NC: U.S. EPA Office of Air Quality Planning and
Standards.

29

-------
Wiedmann, T., M. Lenzen, K. Turner, and J. Barrett. 2007. Examining the Global

Environmental Impact of Regional Consumption Activities - Part 2: Review of
Input-Output Models for the Assessment of Environmental Impacts Embodied in
Trade." Ecological Economics 61(1): 15-26.

Wier, M., and B. Hasler. 1999. "Accounting for Nitrogen in Denmark: A Structural
Decomposition Analysis." Ecological Economics 30(2): 317-331.

30

-------
Table 1. National Emissions Inventory Major Source Categories