Industrial Pollution in Economic Development: Kuznets Revisited
Industrial Pollution in Economic Development: Kuznets Revisited
Industrial Pollution in Economic Development: Kuznets Revisited
net/publication/23548929
CITATIONS READS
114 2,652
3 authors, including:
Muthukumara Mani
World Bank
37 PUBLICATIONS 2,115 CITATIONS
SEE PROFILE
Some of the authors of this publication are also working on these related projects:
All content following this page was uploaded by Muthukumara Mani on 06 January 2014.
Front Cover:
Industrial water pollution stabilizes with economic development, but there is no evidence of a
decline.
Abstract:
This paper uses new international data to test for an inverse U-shaped, or ‘Kuznets,’
relationship between industrial water pollution and economic development. Hettige, Mani and
Wheeler measure the effect of income growth on three proximate determinants of pollution: The
share of manufacturing in total output; the sectoral composition of manufacturing; and the
intensity (per unit of output) of industrial pollution at the end-of-pipe. They find that the
manufacturing share of output follows a Kuznets-type trajectory, but the other two determinants
do not. Sectoral composition gets ‘cleaner’through middle-income status and then stabilizes. At
the end-of-pipe, pollution intensity declines strongly with income. The authors attribute part of
this to stricter regulation as income increases, and part to pollution-labor complementarity in
production. When they combine the three relationships, they do not find a Kuznets story.
Instead, total industrial water pollution rises rapidly through middle-income status and remains
approximately constant thereafter.
To explore the implications of their findings, Hettige, Mani and Wheeler simulate recent
trends in industrial water pollution for industrial economies in the OECD, the NIC’s, Asian LDC’s
and the ex-COMECON economies. They find approximately stable emissions in the OECD and
ex-COMECON, moderate increases in the NIC’s and rapidly-growing pollution in the Asian
LDC’s. During the 1980’s, their estimates suggest that the latter group displaced the OECD
economies as the world’s largest generator of industrial water pollution. Overall, however, the
negative feedback from economic development to pollution intensity was sufficient to hold total
world pollution growth to around 15% during a twelve-year sample period.
INDUSTRIAL POLLUTION IN ECONOMIC DEVELOPMENT:
KUZNETS REVISITED
by
Hemamala Hettige*
Muthukumara Mani
David Wheeler
December, 1997
*
The authors are respectively Economist, Asian Development Bank, and Consultant and Principal
Economist in the Infrastructure and Environment Unit, Development Research Group, World
Bank. Our thanks to the many Bank staff members, consultants and officials of national
environmental protection institutions who made the data for this study available to us. We
gratefully accept responsibility for any errors of transcription or interpretation.
1. INTRODUCTION
A number of recent studies have explored the relationship between economic development
and environmental quality. Theoretical papers by Gruver (1976), John and Pecchenino (1992),
and Seldon and Song (1995) have derived transition paths for pollution, abatement effort and
development under alternative assumptions about social welfare functions, pollution damage, the
cost of abatement, and the productivity of capital. Empirical studies (Hettige, et. al. (1992),
Shafik (1994), Seldon and Song (1994) and Grossman and Krueger (1995)) have searched for
systematic relationships by regressing cross-country measures of ambient air and water quality on
various polynomial specifications of income per capita. This extensive body of work has been
motivated by several related questions: Does pollution follow a ‘Kuznets’curve, first rising and
then falling as income increases? At what income level does the turnaround occur? Do all
pollutants follow the same trajectory? Is pollution reduction in developed economies due
The theoretical work has shown that a Kuznets, or inverted-U, relationship can result if a few
plausible conditions are satisfied as income increases: Constant or falling marginal utility of
consumption; rising marginal disutility of pollution; constant or rising marginal pollution damage;
and rising marginal abatement cost. Of course, actual turnaround points depend on the relative
magnitudes of the underlying parameters, as well as their signs. Although they are not explicitly
captured by the theoretical models, structural change in the economy and more effective
The empirical results are roughly consistent with a Kuznets curve for conventional air
pollutants such as suspended particulates and sulphur dioxide, but the results for water pollution
are mixed. In most cases, however, the implied trajectories are sensitive to inclusion of higher-
1
order polynomial terms in income whose significance varies widely. Structural interpretation of
the estimates remains ad hoc, since the existing studies have incorporated almost no evidence
This paper attempts to advance the state of the art, using new data on industrial water
emissions in developed and developing countries. Our analysis decomposes total industrial
pollution into four proximate determinants: National output; the share of industry in national
output; the share of polluting sectors in industrial output; and end-of-pipe pollution intensities in
the polluting sectors. As most of the previously-cited work has noted (without being able to
resolve the issue), declining pollution at higher levels of development must be driven by some
We investigate these changes in three econometric exercises. Using international panel data,
we estimate the effects of economic development on industry’s share of total output and the
pollution intensity, we have collected factory-level data on industrial water pollution from national
and regional environmental protection agencies (EPA’s) in twelve countries: Brazil, China,
Finland, India, Indonesia, Korea, Mexico, Netherlands, Philippines, Sri Lanka, Taiwan (China),
Thailand and the US. Controlling for sectoral differences, we use these data to investigate the
effects of income per capita, regulatory strictness and relative input prices on factory-level
regulatory strictness to a cross-country labor intensity equation to test for the impact of regulation
1
A partial exception is the work of Seldon and Song (1994), whose regressions employ air emissions instead of
ambient air quality measures. The lack of monitoring information forces the authors to estimate air
emissions from secondary sources: National fuel use data and fuel-based pollution parameters which are
adjusted for conditions in countries at varying income levels. Data scarcity in developing countries is clearly
2
on the demand for labor. For our international pollution accounting exercise, these results
provide two inputs: A measure of average water pollution intensity for each industry sector (an
input to our study of income-related changes in polluting sectors), and an estimate of the change
We combine our econometric results to simulate the total effect of economic development on
industrial water pollution. In this case, we do not find an overall inverse U-shaped relationship.
The three factors have very different relationships with income, and their joint product with total
output is asymptotic, not parabolic. Industrial water emissions rise until countries attain middle-
income status, and then remain approximately constant as they grow richer.
While our results do not support the Kuznets hypothesis for industrial water pollution, they
results suggest that pollution and labor intensities with respect to output decline continuously, and
at almost exactly the same rate, as income increases. Thus, sectoral pollution/labor ratios remain
approximately constant during the development process. This finding provides useful leverage
for the analysis of pollution trends across countries and over time. As an illustration, we combine
our estimated sectoral pollution/labor ratios with panel data on sectoral employment to simulate
international trends in industrial water pollution during the past two decades.
The remainder of the paper is organized as follows. Section 2 develops the models which
link our three pollution factors to economic development. Section 3 introduces the data used for
estimation. Section 4 discusses the results and their implications, while Section 5 provides
a problem for this exercise. Of thirty countries in the estimation sample only four (China, India, Thailand,
Turkey) are LDC’s.
3
illustrative estimates of recent water pollution trends in a number of developed and developing
The first stages of economic development typically witness the rapid growth of industrial
activity and declining environmental quality in densely-populated urban areas. When new
industries are pollution-intensive, their emissions can increase local ambient pollutant
(1) P = m( y )Qρ( y )η ( y )
where
development: The manufacturing share of total output (m), the pollution intensity of
manufacturing (ρ), and the degree of pollution abatement by industry (η). In this decomposition,
the effect of economic development on pollution depends on the signs and the magnitudes of the
have suggested an inverted-U relationship between the manufacturing share of output (m) and
4
income per capita (y).2 During the first phase of economic growth, m increases as industry
expands more rapidly than agriculture. As the economy begins to mature, rapid growth in
services becomes the dominant factor and m declines. Over the existing range of national incomes
Our empirical analysis uses cross-country evidence for the past two decades to estimate this
relationship and test its intertemporal stability. We focus particularly on changes in δm/δy as
development proceeds. Controlling for growth in total output, large movements inm will have a
The sectoral composition of industrial activity has an important effect on its average pollution
intensity, or pollution per unit of output. Industrial processes differ greatly in their production of
waste residuals which, in turn, have varying potential for creating environmental damage.
Abatement costs also differ significantly by industry sector (Dasgupta, et. al. (1996), Hartman, et.
al., (1997)). Even in well-regulated economies, these factors cause significant intersectoral
differences in pollution intensity. For example, metals and cement are generally intensive in
harmful air pollutants; food and paper production are disproportionate emitters of organic water
Anecdotal evidence suggests that the sectoral composition of industry follows a ‘clean’trend
industries, which generate heavy pollution loads as they convert bulk raw materials into primary
2
For a discussion of structural change in development, see Syrquin (1989).
5
inputs (e.g. metals, paper, cement, sugar). During the development process, primary industries
may lose output share to cleaner industries (e.g. vehicle and electronics assembly, instruments). 3
In this paper, we test the clean trend hypothesis for industrial water pollution by fitting the
~ = β + β log y + β (log y )2
ρ jt 0 1 jt 2 jt
(3) where
∑ k k
~ = s ρ∃
ρ
k
The marginal cost of abating pollution from industrial sources is a function of the scale of
activity, pollutant concentration in process influent,4 the degree of abatement, and local input
prices. 5 In static partial equilibrium, cost-minimizing firms with flexible abatement choices will
control pollution to the point where their marginal abatement costs equal the ‘price’exacted for
pollution by affected parties. 6 Characteristic production scale and process effluent intensity differ
3
See Mani and Wheeler (1997) for further discussion.
4
‘Influent’refers to emissions from industrial processes before treatment (or abatement); ‘effluent’refers to
emissions to air, water or land after treatment; ‘concentration’refers to the quantity of pollutant per unit
volume of the waste stream.
5
For recent empirical evidence, see Hartman, et. al. (1997) and Dasgupta, et. al. (1996).
6
These may, according to the circumstances, include local administrators, pressure groups, national regulators,
stockholders, and ‘green consumers.’ Each group is in a position to impose some cost on a firm or plant if its
emissions exceed the norms adopted by that group. Thus, even where pollution charges are in effect, there is
no single ‘price’of pollution. For a detailed discussion, see Afsah, et. al. (1996).
6
significantly by sector, and abatement costs differ by location. Differences in the groups affected
Where the environment is ‘cheaper’or abatement is more expensive, the pollution intensity of
production in a particular sector should be higher, ceteris paribus. However, data scarcity has
made it difficult to test the magnitude of these effects, as well as the impact of spatial variation in
the prices of capital, labor, energy and materials. At present, we have sufficient data to
ln P = α + ∑ δ D + α ln R + α ln W + α ln W + α ln W + α ln W + α ln Q
j 0 SP SP R j K Kj L Lj E E M M j Q j
(4) S j
ln L = β + ∑ β D + β ln R + β ln W + β ln W + β ln W + β ln W + β ln Q
j 0 SL SL P j K K L Lj E Ej M Mj Q j
S j
P = Plant-level pollution
L = Plant-level employment
D = Vector of dummy variables for S sectors
R = An index of regulatory strictness
WK,L,E,M = Prices of capital, labor, energy and materials
Q = Plant-level output
Our dataset, described in the following section, combines information from several sources:
plant- and sector-level data on emissions and employment from national and regional EPA’s;
sector-level information on output and employment from national census bureaus and the World
Bank’s international database (BESD); and data from BESD on national income, population, and
7
For recent evidence, see Pargal and Wheeler (1996), Wang and Wheeler (1996), and Hettige, et. al. (1997).
8
Data-gathering in this context is not a simple task. Even assembly of the relatively sparse dataset used for this
exercise has required a massive canvass of World Bank project files, consultants’reports, and emissions
reports from many national environmental protection institutions. The data are briefly surveyed in the
following section, with a more detailed description in the Appendix.
7
a number of other variables. For cross-country consistency, we use summary data by sector.9
Plant-level relations between scale (Q) and pollution intensity (P/Q) are not relevant for sectoral
estimation of the pollution and labor equations in intensity form, with P/Q and L/Q as the
dependent variables.
Using the BESD database, we estimate sectoral average L/Q ratios for each sample country.
We construct sectoral average P/L ratios (sectoral pollution intensities w.r.t. labor) from the data
provided by national and regional EPA’s. We estimate P/Q (pollution intensity w.r.t. output) by
multiplying L/Q and P/L for each sector and country. The results permit us to estimate the
following equations:
P
j
ln =α + ∑ δ D + α ln R + α ln W + α ln W + α ln W + εj
(5) Qj 0 SP SP R j K Kj L Lj E E
S j
L
j
ln = β + ∑ β D + β ln R + β ln W + β ln W + β ln W + νj
Qj 0 SL SL P j K K L Lj E Ej
S j
9
See Appendix I for a description of data sources in each country.
10
Marginal abatement costs decline with treatment scale for most pollutants, because abatement capital is lumpy.
Thus, the estimated output elasticity of emissions in a plant-level equation is generally less than one. For
evidence from Asia, see Pargal and Wheeler (1996). At the sectoral level, however, the constant-returns
assumption seems appropriate for cross-country work. It is possible that characteristic plant scale is larger in
countries with greater sectoral output, but we have no way to test this proposition with the available data.
8
are complements in production, while the pairs KE and LM are gross substitutes.11 If these
relations hold, a wage increase should have the following effects on emissions:
(1) Materials use and the volume of polluting residuals should decline;
(2) Labor use should decrease in both processing and pollution abatement activities, with some
increase in pollution from the latter effect. However, our prior expectation is that the materials-
reducing effect should dominate: A wage increase should reduce water pollution intensity.
Energy Price: If labor and energy are gross substitutes in production, then an increase in
the price of energy should increase the labor intensity of production. In the case of pollution, an
energy price increase will reduce energy use for both processing and pollution abatement.
Abatement activity should therefore fall, and water pollution intensity should rise.
Capital Price: A capital price increase should also increase labor intensity. For pollution,
an increase in the interest rate or the price of equipment should reduce capital and energy use as
well as pollution abatement, while increasing the use of labor and materials in processing. Both
reduced abatement and increased materials use should lead to more water pollution.
Regulatory strictness and some input prices (e.g., wages) change systematically as per capita
income increases. To assess the overall impact of economic development, we also estimate our
P
j
ln = ρj = γ + ∑ δ D + ρy ln y + ε j
(6) Qj 0
S
SP SP
L
j
ln = λj = γ + ∑ γ D + γy ln y + ν j
Qj 0 SL SL
S
These equations have two specific roles to play in our analysis. First, they provide an
estimate of ρy, the elasticity of end-of-pipe pollution intensity with respect to income per capita.
We use this to construct the index η in Equation (1). The results also provide estimates of
5
See Christensen, et al. (1973).
9
average sectoral pollution intensities (δSP) across countries. We combine these intensities with
3. DATA
To our knowledge, this is the first comparative international study of industrial pollution
which uses direct observations on emissions. We have obtained the data from environmental
protection agencies in Brazil, China, Finland, India, Indonesia, Korea, Mexico, Netherlands,
Philippines, Sri Lanka, Taiwan (China), Thailand and the US. Descriptions of the data sources
We use the pollution information and complementary employment data to estimate emissions
intensities by industry sector in kilograms per day per employee. We focus on organic water
pollution because it provides the most plentiful and reliable source of comparable cross-country
emissions information. Water pollution data are the most plentiful because developing countries
have traditionally begun industrial pollution control programs with regulation of organic water
emissions. They are relatively reliable because sampling techniques for measuring water pollution
are more widely understood and much less expensive than those for air pollution.
Some comparable measure of regulatory strictness is necessary for estimation of our cross-
country equations. However, credible indices of environmental regulation are difficult to find.
Even in the US, comparative analyses of state-level regulatory ‘outputs’have generally used
10
of inspectors.12 Such measures may have at least some justification for within-country analyses,
since quality- and price-adjustment problems are not too serious. For international comparisons,
however, they would be problematic even if comparable data were available. Most developing
countries do not have such data, so input-based comparisons are not possible in any case.
A more promising approach has been taken by recent econometric work on the sources of
variation in regulatory strictness. This work is helping to identify robust proxies which can be
used as instruments in cross-country comparisons. The best instrument is undoubtedly per capita
income, which has been shown to affect both formal and informal regulatory pressure on polluters
in the US and Asia (McConnell (1992), Pargal and Wheeler (1996), Hartman, et. al. (1997),
Hettige, et. al. (1996) and Wang and Wheeler (1996)). Dasgupta et. al. (1995) have advanced
the state of the art by developing quantitative indices of regulatory development from reports filed
for the U.N. Conference on Environment and Development (UNCED - Rio de Janeiro, 1992).
Their results suggest that international differences in pollution regulation are well-explained by a
model which incorporates the effects of per capita income, urbanization, population density, and
manufacturing share in national output. We have adopted the Dasgupta model to produce a
cross-country pollution regulation index for this paper. Six of our thirteen country cases have
actually been scored by the Dasgupta exercise. For the remaining seven cases, we have
calculated the pollution regulation index values using the Dasgupta equation.
12
See for example Levinson (1994) and Beede (1993).
11
3.3 Input prices
We have computed wages (in $US 1990 per worker) by ISIC sector from UNIDO’s reported
sectoral totals for employment and payrolls. Our electricity tariff rates for the OECD and
developing countries have been drawn from International Energy Agency data and the World
Bank’s Power Sheets database, respectively. The World Bank’s World Development Indicators
Estimation of equations (2), (3), (5) and (6) requires cross-country data on total output,
industrial output, employment, income, population and a number of other variables. We have
obtained the relevant panel data from the World Bank’s international database (BESD). We have
4. ECONOMETRIC RESULTS
Table 4.1 reports panel estimates for equation (2). We provide comparable results for OLS,
fixed-effects (without time dummies) and random effects models. We prefer the random effects
model, but the choice of estimator does not have a major effect on the results. They are
consistent with an inverted-U model for manufacturing share in national output. Our results also
suggest some structural change in the relationship, since the interactions of time with income and
income squared both satisfy classical significance criteria. During the past two decades, the
income class for all 1,717 observations in our sample. The result (Figure 4.1) suggests that
12
manufacturing share rises steeply with income until a country reaches middle-income status;13
from around 10% in countries with less than $1000 per capita (Summers-Heston income, in $US
1990) to around 25% in countries with incomes of $5,000-$6,000. Then the manufacturing share
Figure 4.1: Manufacturing Share in GDP vs. Per Capita Income, 1975-1994
25.0
Manufacturing Share (%)
20.0
15.0
10.0
5.0
0.0
500
1500
2500
3500
4500
6000
8000
12000
10000
14000
17000
20000
Table 4.2 reports results for our analysis of changes in sectoral composition. We have
employed panel techniques to estimate Equation (3) in log-log form, using the log of share-
weighted average BOD intensity as the dependent variable. Again, the fixed-effects and random-
effects estimates tell the same story: As income per capita increases, overall pollution intensity
declines because relatively ‘clean’sectors grow more quickly. However, our results suggest that
13
We have smoothed the series with a three-interval moving average: Each share observation on the graph is the
average for the previous, corresponding, and succeeding income groups.
13
the rate of decline also decreases. We find no evidence of a structural change, except for a very
In Figure 4.2, we provide an illustration of the relationship between overall pollution intensity
and income during the sample period.14 The figure is based on median values of overall intensity
for each income group in the set of 2,210 observations. It suggests that sector-weighted average
water pollution intensity declines from nearly 6 Kg. to 4Kg. Per $US 1 million per day, or about
30%, as income increases to around $5,000 per capita. Then it remains approximately stable over
5.5
4.5
3.5
3
1500
2500
3500
4500
6000
8000
10000
12000
14000
17000
20000
500
14
We have also used three-interval smoothing for Figure 4.2. See Footnote 13.
14
4.3 End-of-Pipe Pollution Intensity
Tables 4.3 - 4.5 report cross-country regression results for Equations (5) and (6). We use
dummy variables to control for sectoral differences in average pollution intensity; dummy variable
controls are also introduced for national differences in reporting procedures and measures of
organic water pollution. The majority of environmental protection agencies (EPA’s) have
reported emissions of biological oxygen demand (BOD), which is a measure of oxygen removal
from water by bacteria which are oxidizing organic materials. However, three EPA’s – for
China, Netherlands and Taiwan (China) – have reported COD (chemical oxygen demand). COD
incorporates the effect of other pollutants on the rate of oxidization; it is systematically larger than
BOD measures.
We have controlled for the measurement problem by introducing a dummy variable for COD-
based emissions reports. As expected, the estimated COD dummy is positive, large and highly
significant in all pollution intensity equations. Our sectoral dummy variable results are also in
accord with prior expectations: Food and Paper have the highest average organic water pollution
intensities; Metals and Mineral Products have the lowest. In the case of labor intensity, Textiles,
Food and Wood Products are highest (along with Other Manufacturing, the numeraire sector);
We have also controlled for the possible impact of differences in emissions reporting
procedures. In several cases (China, India, Indonesia, Netherlands, Philippines, Sri Lanka,
Taiwan (China), Thailand) the plant-level information provided by the EPA’s includes
employment data. This has enabled us to estimate sectoral pollution/labor ratios directly from the
EPA data. In the other cases (Brazil, Finland, Korea, Mexico and the US), the EPA’s have
provided summary pollution data by sector. We have obtained summary employment data by
15
sector from other national or regional sources, and have used the two summaries to calculate
We recognize the possibility of systematic differences in the results generated by these two
approaches. EPA’s in developing countries focus on large polluters, so the average pollution
intensity of these facilities will be reflected in estimates based on plant samples. The situation is
potentially quite different when EPA-reported sectoral emissions are divided by census-reported
sectoral employment. Plants which ignore pollution regulations (and whose reported pollution is
therefore zero) may nevertheless be registered in an employment census. This might impart a
downward bias to summary-based intensities. In addition, all five countries for which we employ
summary data (Brazil, Finland, Korea, Mexico, US) are in the middle or high income category.
Thus, failure to control for the sampling difference might also produce a downward bias in the
We have introduced a dummy variable to control for this difference, but it is not significant in
our regressions. In fact, we are not overly surprised by this result because effective coverage of
As expected, the estimated wage-elasticity of labor intensity is large (around -.70) and highly
significant. The wage elasticity of pollution intensity is also negative, large (-1.71) and highly
significant. In the pollution intensity equation, our results are consistent with the hypothesis that
15
Both approaches may underestimate ‘true’ sectoral pollution intensities in developing countries, because
existing research suggests that medium and large plants have lower pollution per unit of output than smaller
facilities (ceteris paribus). Since smaller plants are covered by regulators in developed economies, our
econometric result may actually understate the effect of income on pollution intensity. We accept the
plausibility of this hypothesis, but we have no way to test it at present.
16
labor and pollution are complements in production. However, the converse is not true. Our
While the latter result is not particularly surprising, we also find that our regulatory strictness
index is not significant in pollution intensity regressions which control for wages. Does this
imply that market forces alone drive pollution, and that regulation is irrelevant? Although our
results are consistent with this interpretation, we reject it for several reasons. First, our wage and
regulation variables are highly collinear because they are both correlated with per capita income.
As Table 4.4 shows, each variable is significant in equations which exclude the other. Second, a
large body of empirical work suggests that industrial pollution is responsive to pressure from local
communities (Pargal and Wheeler, 1996; Hettige, et. al., 1997, Hartman, et. al., 1996), as well as
formal regulation. Both forms of regulation are strongly affected by income, reflecting
increasing preferences for environmental quality and higher valuation of pollution damage. We
believe that the estimated wage elasticity in our pollution intensity regression is capturing cross-
country income effects on formal and informal regulation, as well as the effect of complementarity
between these two effects. However, their joint effect clearly shows the impact of rising income
on pollution intensity.
Our results for energy and capital prices are considerably weaker. Surprisingly, neither
variable is significant in the labor intensity equation when both are included. In the pollution
intensity regression, the estimated electricity price elasticity is positive, large and highly
significant. The real interest rate elasticity is also positive, and close to significance at the 5%
level. However, these results are not robust to changes in right-hand variables or sample
composition. Dropping the real interest rate increases the sample size, because we do not have
17
real interest rate data for Mexico, Brazil and Taiwan (China). However, with the larger sample
the electricity price elasticity loses significance in the pollution intensity equation, while becoming
large, negative, and highly ‘significant’in the labor intensity equation. We conclude that our
results for capital and energy prices are highly sensitive to outliers, and we see no reason to draw
We have also estimated reduced-form intensity equations which control for per capita
income, sector and COD reporting. The results are summarized in Table 4.5 for three intensities:
labor/output, pollution/output and pollution/labor. In all three equations, the results for the
sectoral dummies replicate the pattern of results in Tables 4.3-4.4. As before, the dummy variable
The results for per capita income suggest a striking regularity across countries. The income
elasticities of pollution/output and labor/output are both negative, and not significantly different
from one. In the third equation, we test for the equality of pollution and labor elasticities (w.r.t.
income) by regressing pollution/labor on the same set of right-hand variables (this amounts to
differencing the coefficients in the first two equations). The resulting elasticity of
pollution/labor with respect to income per capita is not significantly different from zero. Of
course, we cannot generalize from one sample for one pollutant to all industrial emissions.
However, for industrial water pollution, our results suggest that sectoral emissions/labor ratios
are approximately constant across countries at all income levels. Developing economies
generate much more pollution per unit of output than developed economies, but they also employ
much more labor per unit of output, and in the same proportion.
18
Figure 4.3 and Table 4.6 portray the estimated relationship between pollution intensity (per
unit of output) and income per capita. For ease of interpretation, we normalize to an intensity
value of 100 for the poorest income category ($500 per capita). The cross-country evidence
suggests a sharp drop in pollution intensity with income growth, as manufacturers respond to
higher wages and regulatory pressures with end-of-pipe abatement and process change. From an
emissions index value of 100 at $500 per capita, pollution abatement is about 60% at $1,500,
Income %
Per Capita Abatement
1,500 60
3,000 80
7,000 90
15,000 95
100
Normalized Pollution Intensity
90
80
70
60
50
40
30
20
10
0
0 2000 4000 6000 8000 10000 12000 14000 16000 18000 20000
Per Capita Income ($US)
19
5. IMPLICATIONS OF THE RESULTS
Our estimation exercises have suggested three distinct patterns of response to economic
development. Industry’s share of national output rises sharply through middle-income status and
then slowly declines. Sectoral composition follows a ‘clean’trend for low-income developing
countries, but exhibits little or no trend beyond the middle income range. End-of-pipe pollution
We use simulation to project the net result of changes in these three factors. Our four
simulation variables are in columns 1-4 of Table 5.1. Column 1 includes a broad range of
incomes, from $US 500 to $US 20,000 per capita. Columns 2 and 3 replicate the information on
manufacturing output shares and average pollution intensities in Figures 4.1 and 4.2. Column 4
reproduces the pollution intensity index in Figure 4.3, re-normalized to one for the lowest income
level.
We assume a unit population for convenience, so income per capita also serves as a measure
of total output. We simulate the overall relationship between economic development and
industrial pollution by multiplying the four column entries in each row. The result combines the
effects of changes in total output, manufacturing share, sectoral composition, and end-of-pipe
pollution intensity. Column 5 and Figure 5.1 portray the total pollution estimate, which has been
normalized to an index value of 100 at the lowest income level. Our result suggests that the
inverted U-shaped story is only half right for industrial water pollution: Total emissions rise
sharply in the range [$500 - $7,000], but remain constant as income increases further.
20
Figure 5.1 Industrial Pollution and Economic Development
300
Water Pollution Index
250
200
150
100
50
500
20000
3500
4500
6000
1500
2500
8000
10000
12000
14000
17000
Per Capita Income ($US)
To assess the contribution of each factor to the overall result, we perform three
counterfactual simulations which are tabulated in columns 6-8 and illustrated in Figure 5.2 Each
simulation allows one of columns 2-4 to vary while holding the other two constant at the lowest-
income level. The experiment in column 6 holds sectoral composition and end-of-pipe pollution
intensity constant, while allowing the share of manufacturing to vary with income. The result is
rapid growth of pollution over the whole income range, and an estimated pollution load at
$20,000 which is eighty times the initial load. To produce column 7, we allow sectoral
composition to vary while holding the other two factors constant. Pollution growth is
considerably moderated by comparison with column 6, but the projected load at $20,000 is still 40
times the initial load. Finally, we test the effect of end-of-pipe change in column 8. This
experiment clearly identifies the most important factor: Projected emissions at $20,000 are only
1.8 times the initial load, if manufacturing share and sectoral composition are held constant.
21
We conclude that pollution levels off in the middle income range because end-of-pipe
pollution intensity responds to rising wages and stricter regulation. By comparison, the
manufacturing share and sectoral composition are minor players. For industrial water pollution,
the inverted-U pattern does not emerge because declining pollution intensity almost exactly
balances output growth, while manufacturing share and sectoral composition remain constant
8,000
7,000 Variable:
Water Pollution Index
Pollution Intensity
6,000
Manuf. Share
5,000
Sector Mix
4,000
3,000
2,000
1,000
0
10000
12000
14000
17000
20000
1500
2500
3500
4500
6000
8000
500
To explore the real-world implications of our results, we estimate pollution loads for a set of
large industrial economies during the period 1977 - 1989. Powerful leverage is provided by our
finding that sectoral pollution per unit of labor (P/L) remains approximately constant across the
entire range of incomes. This allows us to use commonly-available sectoral labor/output (L/Q)
ratios to predict international changes in industrial water pollution. As an illustration, we use the
World Bank’s BESD database to estimate sectoral L/Q ratios for fifteen countries during the
22
period 1977-1989. To estimate BOD loads by sector, we multiply the L/Q ratios by sectoral P/L
We have chosen the fifteen countries to represent large industrial economies in four major
groups: OECD (represented by the US, Japan, France and Germany (former F.R.); the NIC’s
(Mexico, Brazil, Taiwan, Korea, South Africa, Turkey); Asian LDC’s (China, India, Indonesia);
and the ex-COMECON countries (Poland, former USSR). The results are tabulated in Table 5.3
(Appendix II) and summarized in Table 5.2 below. Taken together, they illustrate the main
In the OECD, despite modest continued economic growth, estimated BOD emissions remain
almost constant. In our view, this reflects the countervailing effects of output growth and
increases in wages and regulation; manufacturing shares and the ‘clean’sector share change very
little. The COMECON economies are in relative stagnation during the sample period, so there is
The story for the NIC’s is quite different. Their estimated pollution increases by about 25%
during the sample period – substantially less than their growth in per capita income. The increase
is relatively moderate because rapid output growth is offset by three factors: the negative impact
of increased wages and regulation on industrial pollution intensity; the first stage of the decline in
manufacturing share; and the last stage of the ‘clean’trend in sectoral composition.
The Asian LDC experience is also distinctive. Estimated BOD emissions grow by
approximately 55% in these lower-income economies, because rapid output growth and
16
For this application, we regress log (P/L) on dummy variables for COD and the industry sectors. Since income
is insignificant, we impose a parameter value of zero by dropping it from the equation. To calculate sectoral
P/L ratios, we assume the BOD case (COD=0), add the constant term to the estimated parameters for the
sector dummies, and calculate the antilogs of the results.
23
increasing manufacturing share dominate the clean compositional trend and the first effects of
rising wages and regulation on pollution intensity. To a striking degree, BOD growth in our
The overall result of these changes is a significant shift in group shares of total pollution.
Table 5.2 shows that the OECD and COMECON countries drop significantly in share; the NIC’s
increase marginally; and the Asian LDC’s jump from 29% to 37% of the total. Equally
impressive, however, is the apparently moderate growth in total BOD emissions during the
period when world concern over environmental damage was reaching a peak. While economic
development was sparking greater interest in pollution, is was also setting the stage for real
industrial BOD emissions grew by only 16% in these fifteen major industrial countries.
% of Sample Total
1977 1980 1983 1986 1989
OECD 36 34 32 30 30
NIC's 10 11 11 12 12
ASIAN LDC'S 29 30 32 34 37
COMECON 26 25 25 23 22
TOTAL 100 100 100 100 100
24
6. SUMMARY AND CONCLUSIONS
In this paper, we have used new international data to investigate the relationship between
industrial pollution and economic development. To test for a Kuznets effect, we measure the
manufacturing in total output; the sectoral composition of manufacturing; and the intensity (per
unit of output) of industrial pollution at the end-of-pipe. We find that the manufacturing share
follows a Kuznets-type trajectory, but the other two determinants do not. Sectoral composition
gets ‘cleaner’through middle-income status and then stabilizes. At the end-of-pipe, pollution
intensity declines strongly with income. We attribute part of this to stricter regulation as income
Our results suggest that income elasticities of both pollution- and labor-intensity are
approximately minus one. The remarkable implication is that a sector’s pollution/labor ratio is
constant across countries at all income levels. Our findings motivate two illustrative simulation
exercises. First, for a set of income benchmarks, we simulate total pollution by combining
pollution abatement. We do not see a Kuznets-type story in the result, since total pollution rises
rapidly through middle-income status and remains approximately constant thereafter. In three
determinants. Our results highlight the dominance of end-of-pipe reductions as wages and
regulation increase with development. The combined influence of changes in manufacturing share
Our second simulation uses international panel data to explore the implications of constant
sectoral pollution/labor ratios. We estimate recent trends in water pollution for fifteen major
25
industrial nations in the OECD, the NIC’s, Asian LDC’s and the ex-COMECON economies. We
find approximately stable emissions in the OECD and ex-COMECON, moderate increases in the
NIC’s and rapidly-growing pollution in the Asian LDC’s. During the 1980’s, our estimates
suggest that the latter group displaced the major OECD economies as the world’s largest
generator of organic water pollution. Overall, however, the negative feedback from economic
development to pollution intensity was sufficient to hold total world pollution growth to around
In closing, it is worth asking whether these results are cause for optimism or pessimism. The
appropriate answer seems to be ‘both.’ It is comforting to see that industrial water emissions
level off in richer economies because pollution intensity has an elastic response to income growth.
Unfortunately, unitary elasticity implies that total emissions remain constant unless other factors
intervene. Of course, industry tends to deconcentrate over time as infrastructure improves and
prosperity spreads. Constant total emissions may therefore be consistent with improving water
quality in at least some areas. However, the continued existence of many seriously-polluted
waterways, even in the most prosperous countries, suggests that economic development remains
26
REFERENCES
Afsah, Shakeb, Benoit Laplante and David Wheeler, 1996, “Controlling Industrial Pollution: A
New Paradigm,” World Bank, Policy Research Department Working Paper
Beede, David, David Bloom and David Wheeler, 1992, “Measuring and Explaining Cross-
Establishment Variation in the Generation and Management of Industrial Waste, World Bank
(mimeo.)
Birdsall, Nancy and David Wheeler, 1993, “Trade Policy and Industrial Pollution in Latin
America: Where are the Pollution Havens? Journal of Environment and Development, 2, 1,
Winter, 137-149.
Dasgupta, Susmita and David Wheeler, 1996, “Citizen Complaints as Environmental Indicators:
Evidence from China,” World Bank, Policy Research Department Working Paper, December
Dasgupta, Susmita, Ashoka Mody, Subhendu Roy and David Wheeler, 1995, “Environmental
Regulation and Development: A Cross-Country Empirical Analysis," World Bank, Policy
Research Department Working Paper, No. 1448, April
Dasgupta, Susmita, Mainul Huq, David Wheeler and C.H.Zhang, 1996, “Water Pollution
Abatement by Chinese Industry: Cost Estimates and Policy Implications,” World Bank, Policy
Research Department Working Paper
Grossman, Gene and Alan Krueger, 1995, “Economic Growth and the Environment,” Quarterly
Journal of Economics, May, 353-377
Gruver, G.W., 1976, “Optimal Investment in Pollution Control Capital in a Neoclassical Growth
Context,” Journal of Environmental Economics and Management, 5, 165-177.
Hartman, Raymond, Mainul Huq and David Wheeler, 1997, “Why Paper Mills Clean Up:
Determinants of Pollution Abatement in Four Asian Countries,” World Bank, Policy Research
Department Working Paper, Number 1710.
Hartman, Raymond, Manjula Singh and David Wheeler, 1997, “The Cost of Air Pollution
Abatement,” Applied Economics (forthcoming)
Hettige, H., R.E.B. Lucas and D. Wheeler, 1992, “The Toxic Intensity of Industrial Production:
Global Patterns, Trends and Trade Policy,” American Economic Review Papers and
Proceedings, 82, 478-481.
27
Hettige, Hemamala, Paul Martin, Manjula Singh and David Wheeler, 1995, "IPPS: The Industrial
Pollution Projection System," World Bank, Policy Research Department Working Paper,
February.
Hettige, Hemamala, Mainul Huq, Sheoli Pargal and David Wheeler, 1996, "Determinants of
Pollution Abatement in Developing Countries: Evidence from South and Southeast Asia," 1996,
World Development, December.
Hettige, Hemamala, Manjula Singh, Sheoli Pargal and David Wheeler, 1997, “Formal and
Informal Regulation of Industrial Pollution: Evidence from the US and Indonesia,” World Bank
Economic Review, Fall.
John, A. and R. Pecchenino, 1992, “An Overlapping Generations Model of Growth and the
Environment,” Department of Economics, Michigan State University (mimeo.).
Mani, Muthukumara S. and David Wheeler, 1997, “In Search of Pollution Havens? Dirty
Industry in the World Economy, 1960-1995,” World Bank, Policy Research Department
Working Paper (forthcoming)
Mani, Muthukumara S., 1996, “Environmental Tariffs on Polluting Imports: An Empirical Study,”
Environmental and Resource Economics, 7, 391-411
Mody, Ashoka and David Wheeler, Automation and World Competition: New Technologies,
Industrial Location, and Trade (London: Macmillan Press, 1990).
Pargal, S. and D. Wheeler, 1996, “Informal Regulation in Developing Countries: Evidence from
Indonesia,” Journal of Political Economy, December.
Robison, David H., 1988, “Industrial Pollution Abatement: The Impact on the Balance of Trade,”
Canadian Journal of Economics, 21, 702-706.
Seldon, Thomas and Daqing Song, 1994, “Environmental Quality and Development: Is There a
Kuznets Curve for Air Pollution Emissions?” Journal of Environmental Economics and
Management, 27, 147-162.
Seldon, Thomas and Daqing Song, 1995, “Neoclassical Growth, the J Curve for Abatement, and
the Inverted U Curve for Pollution,” Journal of Environmental Economics and Management,
29, 162-168.
28
Syrquin, Moshe, 1989, “Patterns of Structural Change,” in Handbook of Development
Economics, Vol. 1, H. Chenery and T.N. Srinivasan, eds., (Amsterdam: North-Holland)
Tobey, James A., 1990, “The Effects of Domestic Environmental Policies on Patterns of World
Trade: An Empirical Test,” Kyklos 43, Fasc. 2, 191-209.
Wang, Hua and David Wheeler, 1996, “Pricing Industrial Pollution in China: An Econometric
Analysis of the Levy System,” World Bank, Policy Research Department Working Paper, No.
1644.
Wheeler, David and Ashoka Mody, 1992, “International Investment Location Decisions: The
Case of U.S. Firms,” Journal of International Economics, 33, 57-76.
29
Appendix I: DATA SOURCES
Brazil: The water pollution data for the Sao Paulo Metropolitan region of Brazil were collected
by CETESB, the environmental agency for Sao Paulo State. Our pollution estimates are based on
CETESB’s 1250-plant database, which includes measures of BOD loads in kg/day. The
corresponding employment data came from the Sao Paulo State Ministry of Labor, which
provided 2-digit sectoral information from 1991 on nearly 41,000 plants and 2.15 million workers.
China: Water pollution data for China were obtained from the National Environmental Protection
Agency (NEPA), which maintains a comprehensive database on major sources of industrial
pollution in China. Our estimates are based on NEPA’s 1993 emissions data for 269 factories
scattered throughout China.
Finland: The Finnish economic data, aggregated at the 3-digit ISIC level, were provided by the
Central Statistical Office, covering both white and blue collar workers for 1989. The pollution
data were provided by the Industrial Waste Water Office of the National Board of Waters and the
Environment. They cover water emissions in 1992 from 193 large water-polluting factories.
India: The India data are from the state of Tamil Nadu. Plant-level pollution data and
employment data for 1993-94 were provided by the Tamil Nadu Pollution Control Board, which
monitors air and water pollution for all the manufacturing units in the state.
Indonesia: The Indonesia data came from two different sources. The plant-level emissions data
were provided by BAPEDAL, Indonesia's National Pollution Control Agency in the Ministry of
Environment. The economic data are from Indonesia’s Central Statistics Bureau (BPS).
Korea: Korean pollution data were provided by the National Pollution Control Agency. They
cover water emissions by 13,504 facilities in 1991. Complementary employment data have been
drawn from Korea’s National Statistical Yearbooks and the ILO’s International Labor
Statistics, 1991.
Mexico: Data for water emissions in the Monterrey Metropolitan Area were provided by the
State Water Monitoring Authority. The data cover emissions from 7,500 facilities in 1994.
Complementary employment data were provided by Mexico’s Census Bureau (INEGI).
Netherlands: Water emissions and employment data for approximately 700 regularly-monitored
facilities in 1990 were provided by the Emissions Inventory System maintained by the Ministry of
Housing, Spatial Planning and the Environment (VROM).
Philippines: Water emissions and employment data for factories in the Metro Manila Area
(MMA) were provided the Philippines Department of Natural Resources (DENR) and the Laguna
Lake Development Authority.
30
Taiwan (China): Water emissions and employment data for 1,800 plants were provided by the
Water Quality Protection division of the Taiwan Environment Protection Agency.
Sri Lanka: Water pollution and employment data for Sri Lanka were obtained from a study of
waste water treatment options for the Ekala/Ja-ela Industrial Estate, which includes 143 industrial
establishments with 21,000 employees. The data were collected by a joint project of the World
Bank’s Metropolitan Environment Improvement Program and the Sri Lankan Board of
Investment. Ekala/Ja-Ela industrial estate is one of the two major industrial estates in Sri Lanka.
U.S.A.: The information for the United States were drawn from two main sources. The water
emissions data have been collected from regional databases which monitor industrial water
discharges as part of the U.S. Environmental Protection Agency’s NPDES system. Employment
data are from the U.S. Census Bureau’s Longitudinal Research Database.
31
Appendix II: Tables
Table 4.1: Log (Manufacturing Share of Total Output) vs. Log (Income Per Capita), 1975-1994*
Table 4.2: Log (Sector-Weighted BOD Intensity) vs. Log (Income Per Capita), 1975-1994*
32
Table 4.3: Intensity Equations for Pollution and Labor (in Prices and Regulation)
33
Table 4.4: Intensity Equations for Pollution and Labor (in Prices and Regulation)
34
Table 4.5: Intensity Equations for Pollution and Labor (in Income Per Capita)
35
Table 5.3: Estimated Industrial BOD Emissions
Selected Countries, 1977 - 1989 (‘000 Kg./Day)
36
THE WORLD BANK/IFC/M.I.G.A.
OFFICE MEMORANDUM
DATE: December 5, 1997
EXTENSION: 33401
Please find enclosed a copy of the paper, “Industrial Pollution in Economic Development:
Kuznets Revisited,” which we would like to submit to the Policy Research Working Paper
Series. As far as we know, this is the first empirical assessment of the relationship between
development and industrial emissions which uses actual plant-level water pollution data from
developing countries. The paper analyzes the impact of relative input prices, regulation and
income on three proximate determinants of total pollution: The share of manufacturing in
national output; the sectoral composition of manufacturing; and industrial emissions
intensity at the end-of-pipe. The paper tests for Kuznets-type effects in each dimension, and
combines the evidence in an assessment of the overall relationship between pollution and
development.
This paper should be of interest to the international environmental policy community, Bank
environment professionals, and other staff members who are working on industrial and
regulatory issues. It has been reviewed by Professor Judith Dean of SAIS and a number of
colleagues in the World Bank. We believe that country clearance is not necessary, since this
is an international study which does not report results for individual countries.
Enclosures: