Economic Development and Environmental Quality: An Econometric AnalysisAuthor(s): Nemat ShafikReviewed work(s):Source: Oxford Economic Papers, New Series, Vol. 46, Special Issue on EnvironmentalEconomics (Oct., 1994), pp. 757-773Published by: Oxford University PressStable URL: http://www.jstor.org/stable/2663498 .Accessed: 16/02/2012 20:14

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .http://www.jstor.org/page/info/about/policies/terms.jsp

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range ofcontent in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new formsof scholarship. For more information about JSTOR, please contact support@jstor.org.

Oxford University Press is collaborating with JSTOR to digitize, preserve and extend access to OxfordEconomic Papers.

http://www.jstor.org

Oxford Economic Papers 46 (1994), 757-773

ECONOMIC DEVELOPMENT AND ENVIRONMENTAL QUALITY: AN ECONOMETRIC ANALYSIS

By NEMAT SHAFIK The World Bank, 1818 The H Street NW, Washington, DC 20433, USA

1. Introduction

THE RELATIONSHIP between economic growth and environmental quality has been a source of great controversy for a very long time. At one extreme has been the view that greater economic activity inevitably leads to environmental degradation and ultimately to possible economic and ecological collapse. At the other extreme is the view that those environmental problems worth solving will be addressed more or less automatically as a consequence of economic growth. The longevity and passion of this debate has, in part, been a reflection of the lack of substantial empirical evidence on how environmental quality changes at different income levels. Compilation of such evidence has been constrained by the absence of data for a large number of countries. While data remain a problem, the situation is much improved and this paper takes a first step at systematic analysis of what data are available (see Appendix I for details).

A number of caveats are in order. The data on environmental quality are patchy at best, but are likely to improve over time with better monitoring. Comparability across countries is affected by definitional differences and by inaccuracies and unrepresentative measurement sites. At this stage of knowl- edge, this paper has the modest objective of opening up the empirical debate using a relatively simple modeling technique applied on a consistent basis to a large number of environmental quality indicators and countries.

The relationship between income and the costs and benefits associated with any given level of environmental quality is complex because it operates through a number of different channels, such as preferences, technology, and economic structure. The types of environmental degradation that occur depend on the composition of output, which changes with income. Some income levels are often associated with increases in certain polluting activities such as the development of heavy industry whereas economies with large service sectors may generate less pollution. There is a view that rising incomes imply that the cost of environmental degradation is greater because the wages used to value the opportunity cost of illness or work days lost are higher. This would imply increases in marginal benefits as incomes rise. But the poor are often the most exposed and vulnerable to the health and productivity losses associated with a degraded environment. There are some environmental problems where thresholds like survival are at stake. Here, the willingness to pay to avert damage is close to infinity and the level of per capita income only affects the capacity, not the willingness, to pay. With other environmental issues, most of

9() Oxford University Press 1994

758 ECONOMIC DEVELOPMENT AND ENVIRONMENTAL QUALITY

the costs are external (such as transnational pollution or global climate change) and the private benefits of averting damage are small. It is also necessary to consider the intrinsic values of some natural resources. There is a general perception that higher incomes enable the relative luxury of caring about amenities such as landscapes and biodiversity. But many societies with very low incomes, such as tribal peoples, place a very high value on conservation (Davis 1992). Thus, it is not necessarily a question of different preferences between the rich and the poor, but rather one of different budget constraints.

At a theoretical level, it is not possible to predict how environmental quality will evolve with changes in per capita incomes, particularly where public goods are involved. The question is more tractable empirically where we observe some clear patterns. The evidence suggests that, while there is no inevitable pattern of environmental transformation with respect to economic growth at an aggregate level, there are clear relationships between specific environmental indicators and per capita incomes. Where environmental quality directly affects human welfare, higher incomes tend to be associated with less degradation. But where the costs of environmental damage can be externalized, economic growth tends to result in a steady deterioration or environmental quality.

2. Specification of an empirical model

It is hypothesized that there are four determinants of environmental quality in any given country: (i) endowment such as climate and location; (ii) per capita income which reflects the structure of production, urbanization, and con- sumption patterns of private goods including those environmental goods and services which have the characteristics of private goods and services; (iii) exogenous factors such as technology which are available to all countries but change over time; and (iv) policies that reflect social decisions about the provision of environmental public goods depending on the sum of individual benefits relative to the sum of individuals' willingness to pay. The focus of this paper is on the relationship between environment quality and per capita income, taking into account these other determinants of environmental quality.

2.1. Endowment

In the case of endowment, location-specific characteristics that affect environ- mental quality at any given level of income can be accounted for by the inclusion of fixed-effects that allow each country to have its own intercept in regression estimates. It is also possible to include dummies that take into account the characteristics of a specific city or river and the location of measurement sites (such as residential vs commercial, urban vs suburban).

2.2. Income

Per capita income serves to measure directly the relationship between economic growth and environmental quality and measures indirectly the endogenous

N. SHAFIK 759

characteristics of growth. Thus the impact of rising industrialization and urbanization at middle-income levels and the growing importance of services in high income economies are typical patterns that are proxied by per capita income.'

2.3. Technology

For exogenous factors such as technology that can affect environmental quality directly, a time trend can be used as a proxy. The disadvantage of a time trend is that it also captures other exogenous factors that display a trend.

2.4. Policy

Measuring the direct impact of policy, is complicated by the lack of data on a consistent basis for a large sample of countries. While it is possible to construct indicators of macroeconomic policies, such as the trade regime, indebtedness or investment rates, these are not likely to be the policies that matter most for environmental quality. Meanwhile, those policies that are most likely to alter the pattern of environmental degradation, such as regulations governing emissions, energy taxation, or land use, do not lend themselves to such aggregate analysis. Therefore in the case of policy, it is only possible to cautiously infer where we observe a close association between certain types of policies and levels of per capita income.

2.5. The empirical mnodel

Indicators of environmental quality were used as dependent variables in panel regressions based on ordinary least squares estimates using data from up to 149 countries for the period 1960-90.2 The ultimate sample size depended on the availability of data for the relevant variables. The details on data used are provided in Appendix 1. The environmental quality indicators analyzed are the lack of clean water, lack of urban sanitation, ambient levels of suspended particulate matter (SPM), ambient sulfur oxides (SO2), change in forest area between 1961-86, the annual rate of deforestation between 1962-86, dissolved oxygen in rivers,3 fecal coliforms in rivers,4 municipal waste per capita, and

' The inclusion of variables that directly measure urbanization or industrialization would generate multicollinearity and would undermine the objective of evaluating both the direct and indirect effects of growth. See Holtz-Eakin and Seldon (1992) for a similar approach in the case of carbon emissions.

2 Cross-section regressions were also tried for a number of years, but the results were less robust. Because the number of observations and the country coverage varied widely across years, the specifications, coefficient estimates and significance levels varied among cross-section regressions for any given year. The panel results, however, have greater degrees of freedom and provide more consistent results.

3 Low levels of dissolved oxygen, usually caused by human sewage or agro-industrial effluent, reduce the capacity of rivers to support aquatic life.

4 High levels of fecal coliforms result from untreated human wastes that often carry disease.

760 ECONOMIC DEVELOPMENT AND ENVIRONMENTAL QUALITY

carbon emissions per capita.5 Six of these indicators (water, sanitation, SPM, S02, dissolved oxygen, and fecal coliforms in rivers) appropriately measure the quality of a stock of natural resources. For the remaining indicators (deforesta- tion, municipal waste and carbon emissions), reliable data are not available on either the size or the quality of the stock.6 For these variables, changes in the flow that contribute to degradation of the stock are used as a proxy measure of environmental quality.

Three basic models were tested log linear, quadratic, and cubic-to explore the shape of the relationship between income and each environmental indicator

Eit = oco + al lnyat + 02 + 3Fieit (1)

Eit = #0o + f,B lnyat + #2(ln it) 2+ #3Tt + fl4Fi + eit (2)

Eit = 00 + 0, In yt + 02(ln 1t)2 + 03(ln it)3 + 04T + 05Fi + eit (3)

where Ej, is an indicator of environmental quality for country i at time t, Y is per capita income, T is a time trend, F is the fixed effect for site-specific factors, and e is a stochastic error term. Per capita income was defined in purchasing power parity terms.7 All variables are in logarithms unless otherwise specified in the data appendix. Where city variables were used on the left-hand-side (in the case of local air pollutants like SPM and S02), national income figures were used to proxy city incomes.8 Interactive dummies for the city and measurement site were also included in the air pollution and river quality regressions.9

3. Econometric results

The panel regression results for all the environmental indicators are reported in Table 1 and graphical depictions of the patterns of environmental change and per capita income are presented in Fig. 1 based on the coefficient estimates

' The choice of these variables was largely determined by data availability. There are a number of other environmental indicators, such as lead concentrations or species loss, for which sufficient data are not available to begin to analyze systematically.

6 In the case of forests, there are some data on global forest stocks, but the issue of quality is complicated by questions of biodiversity associated with different forest types. For municipal waste, there are no data on total stocks and a true measure of environmental quality would have to take the efficiency of disposal facilities into account. On carbon emissions, there is no clear consensus on the optimal stock of this greenhouse gas.

7 The core model was also estimated using conventional GDP measures and the results were not substantially different, although the PPP measure of income did tend to perform better.

8 National per capita income is a crude proxy for urban income, but sufficient data are not available on income at the city level. The proxy used here assumes that the ratio of urban to national per capita income remains stable. The use of national income may result in a downward bias of turning point estimates because urban incomes tend to be higher than the national average. See Seldon and Song (1992).

9 City dummies were included when air pollution data were available for more than one city in any country. Site dummies were divided into four categories-city central residential, city central commercial, suburban residential, and suburban commercial. City and site dummies were interactive based on the view that pollution from residential and commercial sites across cities might vary depending on the types of industries, local geography, and other site-specific factors.

N. SHAFIK 761

from the panel regressions. The results indicate that access to clean water and urban sanitation are indicators that clearly improve with higher per capita incomes. The addition of the quadratic or cubic terms does not add considerable explanatory power to either the water or sanitation regressions. The time trend is significantly negative in all the regressions, implying that, at any given income level, more people have access to water and sanitation services than in the past. The overall fit of the equations for clean water and urban sanitation is not high, implying that variables other than income also matter. This may reflect the priority given to water and sanitation, even at very low levels of income, because of the critical impact on health.

Not surprisingly, access to clean water and to adequate sanitation are environmental problems that are essentially solved by higher incomes. This reflects the degree to which water, and to a lesser extent sanitation, approximate private goods. In the case of water, the private benefits to provision are high (survival is at stake) and the social costs of provision are fairly low relative to the benefits. These characteristics of water supply have caused some to argue that access to clean water is not truly an 'environmental' problem. Environ- mental problems are not defined here as necessarily deriving from externalities. In the case of water supply, the environmental issues stem from concerns about scarcity and pollution both of which have major consequences for human health, productivity, and the ecosystem. With urban sanitation, the private benefits are not as high as with water, but the social benefits are due to the substantial externalities, particularly those related to health, associated with poor urban sanitation.

The case for deforestation is more complex. The first obstacle is the measurement of deforestation. In the case of other environmental indicators where flow measures were used (municipal wastes and carbon emissions), much of the damage is relatively recent and flows over the past 20-30 years are likely to be correlated with the quality of the stock. In the case of deforestation, some of the degradation, particularly in higher income countries, occurred in the distant past for which data are not available. The annual variation in deforestation rates is deceptive since countries that depleted their forests in the distant past and have slowed down more recently would appear to be doing better than countries with substantial forest resources that have only recently begun to draw down timber stocks. The change in total forest area over a 25-year period also does not capture deforestation that occurred in the distant past. Moreover, there are a number of serious controversies concerning the data on deforestation (for a discussion, see Allen and Barnes 1985) and the data are poor at capturing important differences between types of forest. Recognizing these problems, the disappointing results for both the change in forest area between 1962-86 and the annual rate of deforestation between 1961 and 1986 in Table 1 are not surprising. None of the income terms are significant in any specification. The best fit, relatively speaking, is the quadratic form. But one can only conclude from these results that, given the measurement problems per capita income appears to have very little bearing on the rate of deforestation.

TABLE 1

Environment indicators and income (PPP)

Adjusted R Number of 9

Dependent variables Intercept* Income Income squared Income cubed Time trend squared observations

Lack of safe water 71.36 -0.48 -0.03 0.43 86 n

(3.36) (7.39) (3.02) 0 62.87 1.59 -0.14 -0.03 0.46 86 Z

(3.00) (1.79) (2.34) (3.06) 16.97 19.27 -2.48 0.10 -0.03 0.47 86 n

(0.53) (2.04) (1.99) (1.88) (3.00)

Lack of urban sanitation 169.10 -0.57 -0.08 0.22 123 <

(2.43) (5.65) (2.32) r

167.53 1.07 -0.11 -0.08 0.22 123 0 (2.41) (0.82) (1.26) (2.39) 87.35 27.37 -3.44 0.14 -0.08 0.24 123 t

(1.12) (2.25) (2.24) (2.18) (2.25) H

Annual deforestation 3.49 -0.02 0.00 -0.00 1511 >

(0.79) (0.77) (0.83) 0.64 0.65 -0.04 0.00 -0.00 1511 m

(0.05) (0.85) (0.88) (0.81) 15.57 -5.34 0.76 -0.04 000 -0.00 1511 -

(0.64) (0.66) (0.70) (0.74) (0.23) 0

Total deforestation 2.99 -0.87 -0.01 58 g

(0.04) (0.65) z -9.74 3.33 -0.23 -0.00 58 H (0.95) (1.22) (1.25)

-41.42 16.04 -1.91 0.07 -0.02 58 ,C (0.56) (0.54) (0.49) (0.43)

Dissolved oxygen in rivers -0.18 0.00 0.99 566 t

(1.52) (0.16) -1.46 0.08 0.00 0.99 566 (1.42) (1.25) (0.24)

-11.55 1.34 -0.05 0.00 0.99 566 (0.98) (0.92) (0.86) (0.66)

Fecal coliform in rivers -1.87 0.17 0.96 402 (1.87) (5.20)

- 9.64 -0.74 0.17 0.96 402 (1.25) (1.50) (5.28)

- 256.38 - 31.47 1.27 0.12 0.96 402 (2.77) (2.74) (2.57) (3.10)

Ambient SPM - 008 -0.03 1.00 764 (0.53) (7.05)

- 4.64 -0.29 -0.02 1.00 764 (4.66) (4.62) (5.04)

- -8.09 1.36 -0.06 -0.02 1.00 764 (0.62) (0.80) (0.98) (4.44)

Ambient SO2 0.17 -0.06 0.99 729 (0.83) (9.05) 6.81 -0.41 -0.05 0.99 729 z (3.79) (3.72) (7.40) 37.20 -4.12 0.15 -0.05 0.99 729 (1.36) (1.24) (1.11) (7.39) ,

Municipal waste per capita 2.41 0.38 0.60 39 t (5.51) (7.69) 11.02 -1.70 0.13 0.63 39 (2.50) (1.60) (1.96

- 33.96 15.08 -1.95 0.08 0.64 39 (0.88) (1.05) (1.10) (1.17)

Carbon emissions per capita -15.46 1.62 0.00 0.85 3456 (5.27) (137.12) (0.56)

-22.44 3.22 -0.10 0.00 0.85 3456 (7.60) (22.02) (11.02) (0.82) 6.50 -6.91 1.17 -0.05 -0.00 0.86 3456 (1.56) (6.48) (8.80) (9.59) (0.01)

Note: Regressions without an intercept reported here include city and site or river dummies which allow each country to have its own intercept. 'T' statistics are reported in parentheses below the coefficient estimates.

764 ECONOMIC DEVELOPMENT AND ENVIRONMENTAL QUALITY

1 Lack of safe water C Lack of urban sanitation .o 100 ?0 100

- 1975 - -1980 OQ 1986 0 _ .*----- 1986 0 ~~~~~~~~~~0 E0 50 - -09 ;5 -a

cm~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~c co 0~~~~~~~~~~~~~~~~~~~~~~~~~~~~~c

0aP Jait iome 900 capita income

Annual deforestation Total deforestation 6.25 c c1 0 100 12 - ) 1962 a 2 1997

CD 6.15 - ...... 1 C 0 ...... 1986

6.05

Per~~~~~~~ caiaicoePrcpiaicm

0 0 E 50 -

0 O5.95 -)a Cn

5.575 0 a _ _ _ _ _ _ _ _

100 1000 10000 100000 100 1000 10000 10oooo Per capita income Per capita income

12 1.solved oxygen in rivers Fecal coliform in rivers 20

10 ....1 8. ....1 8 k ~~~~~~~~E

U) 8k o'1

0).Y L~~~~~~~~~

100 1000 10000 100000 100 1000 10000 1odooo Per capita income Per capita income

Suspended particulate matter Ambient sulfur dioxide 120 - 1972 co -1972

o....1986 .....1986

40

23. 40~J -

100 1000 10000 100000 &) 100 1000 10000 100000 Per capita income Per capita income

Municipal solid waste per capita ~ C Carbon emission per capita 700 0 2 e - -1962 a)~~~~~~~~~~~~~~~~~~~~C

. .... 1986

E C i

0cc -1972

~~~0. ~~~~~1986

Per capita income Per capita income

N. SHAFIK 765

This is fairly consistent with other attempts to estimate the causes of deforesta- tion which find little role for per capita incomes and a significant role for agricultural settlement and timber use (Allen and Barnes 1985; Johnson 1991; World Bank 1991).

The two measures of river quality tend to worsen with rising per capita income. Dissolved oxygen seems to be linear with a negative slope implying a tendency for worsening river quality with economic growth. Growing effluent pollution associated with industrialization may play a role in reducing dissolved oxygen at higher incomes. In the case of fecal coliform, the cubic model fits the best implying that fecal content of rivers worsens, then improves, and then deteriorates again at very high income levels. The initial worsening of fecal content, which occurs up to a per capita income level of about $1,375, is probably associated with growing urbanization and consequent pressures on sanitation. The improvement results when urban sanitation services are intro- duced. The increase in fecal coliform at high income levels, which begins at an income level of $11,400, is more difficult to explain. The cubic shape of fecal content is not an artifact of the functional form. The increase in fecal pollution which occurs at incomes above $11,500 per capita is based on 38 observations from seven rivers in three countries (Australia, Japan, and the United States). This cubic relationship held even after some extremely high observations of fecal content from the Yodo river in Japan were dropped from the sample. The increased fecal coliform may reflect improvements in water supply systems where people no longer depend directly on rivers for water and therefore may be less concerned about river water quality. Some sample bias may exist for both measures of water quality because only the most polluted rivers may be monitored in high income countries. Because of these caveats, the results for river water quality must be treated with caution.

Local air pollution follows a 'bell-shaped' curve. Suspended particulate matter (SPM), which causes respiratory illness and mortality, is largely the result of energy use. The regressions for SPM in Table 1 indicate that the quadratic model fits best, implying that pollution from particulates gets worse initially as countries become more energy intensive, and then improves. The improvement begins at a per capita income level of around $3,280. The middle part of Fig. 2 shows changes in the elasticity of SPM with respect to per capita income. SPM is found to be inelastic to changes in per capita income in the range of $570-$18,750. Below $570, a 1% increase in per capita income would lead to more than 1% increase in the SPM. Above $18,750 a similar increase would lead to more than 1% decline in SPM.

Sulphur dioxides, which affect human health and contribute to ecosystem acidification, are also largely the product of energy use, particularly the burning of fuels with a high sulfur content. The results in Table 1 also confirm a quadratic relationship for sulfur oxides with a turning point of around $3,670 per capita. The bottom part of Fig. 2 shows changes in the elasticity of sulfur dioxide concentrations with respect to per capita income. Sulfur dioxide is inelastic to changes in per capita income levels in the range of $1,000-$12,240.

766 ECONOMIC DEVELOPMENT AND ENVIRONMENTAL QUALITY

Elasticity of fecal coliform with respect to income

7-

Z% 3 - .5 ',U00A G 1

-10C E

-3- -1.5 l IlI

100 1000 10000 100000 Income per capita (dollar, log scale)

Elasticity of ambient SPM with respect to income 1.5

1.0C

0.5 -

-1.0

100 1000 ~~~10000 100000 Income per capita (dollar, log scale)

ElsIcit of Camintesulu dneniroxidena wlsiithiespc wto income

2) .0

10.5-

-1.0

*-3 .5

100 1000 ~~~~10000 100000 Income per capita (dollar, log scale)

FlsiG.t 2. Cambgesntslu enironmdentleatcte with rsett income

N. SHAFIK 767

Below $1,100 per capita, a 1% increase in income results in a more than 1% increase in ambient sulfur dioxide. Similarly, the same 1% increase in per capita income for an economy with income above $12,240 would lead to a more than 1% decline in sulfur dioxide concentrations.

In the case of local air pollution, the pattern seems to be one of an initial deterioration of environmental quality as industrialization and energy intensity increases, followed by an improvement as cleaner technologies are used and fuel switching occurs. Technology, proxied by the time trend, appears to have played a favorable role in making improved local air quality possible at an earlier stage of development. In the case of particulates, ambient air quality improves by about 2% a year; ambient sulfur dioxide tends to decline by 5% a year.

These results are broadly consistent with the only other estimates of this type done by Grossman and Krueger (1993) and Seldon and Song (1992) who explore the relationship between air pollutants and income using panel data for a large sample of countries. For two indicators of local air pollution SO2 and dark matter Grossman and Krueger conclude that a cubic functional form provides the best fit. They note, however, that there are only two countries (the US and Canada) in their sample with per capita incomes in excess of $16,000 where the cubic part of the functional form becomes relevant. Thus they conclude that the evidence of a renewed positive relationship between per capita income and SO2 and dark matter at the highest income levels is relatively weaker than the inverted 'U' shape found at lower income levels. Their turning point for SO2 of about $5,000 per capita is somewhat higher than that of $3,670 found here and may reflect differences in sample size, but their turning point estimate is broadly consistent with the conclusion that at middle incomes, local air pollution tends to rise. The only major difference between Grossman and Krueger's findings and those presented here lies in particulates where they get conflicting results in contrast to the significant 'bell-shaped' relationship found here. They find a monotonically declining relationship between particulates and income in a random-effects model while their fixed-effects model generates a monotonically increasing relationship. Seldon and Song's empirical results for air pollutants are also broadly similar, but they tend to find higher turning points. This is probably a reflection of the fact that they use aggregate emissions flow rather than measures of ambient urban air pollution stocks, which are used here because they more directly measure environmental quality and health impact. Because urban incomes tend to be higher than national incomes, turning point estimates based on urban pollution and national income are likely to be higher than those for aggregate emissions and national income.

Municipal waste per capita is one environmental indicator that unambiguously worsens with rising incomes. The log linear specification works best. Unlike air pollution, which is generalized and affects everyone who steps outdoors, solid waste can be disposed of in isolated localities and, if disposed of properly, can have a relatively small impact on human health. Because solid waste disposal can be transformed into a localized and potentially harmless problem, particu-

768 ECONOMIC DEVELOPMENT AND ENVIRONMENTAL QUALITY

larly in areas that are not densely populated or are low income com- munities, higher incomes are not associated with reductions in waste generation.

Carbon emissions per capita, like solid waste, do not improve with rising incomes because widespread awareness of the problem of climate change is relatively recent and because the costs are born externally. The log linear specification has virtually all the explanatory power, although the quadratic and cubic terms are also significant. The turning point on the quadratic specification occurs at an income level that is well outside the sample range of per capita income.'0 The explanation for the exponential increase in carbon emissions per capita with rising incomes is that it is a classic free rider problem. There are no major local costs associated with carbon emissions all the costs in terms of climate change are borne by the rest of the world-and the local benefits in the near term are small in most cases. Technology has not helped, evidenced by the insignificant time trend, because no significant incentives to reduce carbon emissions exist. It is interesting to note that carbon emissions per unit of capital stock have declined over time as countries have moved to cleaner burning fuels and technologies. But this movement to cleaner fuels has been motivated largely by concerns about local, not global, pollutants (see Diwan and Shafik 1992, for an analysis).

4. Changes in income, changes in environmental quality

The elasticity of each environmental indicator with respect to changes in per capita income are provided in Table 2. The elasticities (E) are calculated for three income groups-low, middle, and high based on the coefficient estimates of the best fitting functional form for per capita income reported in Table 1

Linear: E = a 1 (4)

Quadratic: E = 3,B + 2/32 In Y (5)

Cubic: E = a, + 202 In Y+ 302 In y2 (6)

The environmental variables characterized by linear functional forms safe water, urban sanitation, municipal waste, and carbon dioxide emissions-have constant elasticities over changes in income. Access to clean water and sanitation have elasticities of -0.48 and -0.57 respectively, implying that a 1% increase in income results in about 0.5% more people in the population served by improved facilities. Municipal waste has an elasticity of 0.38 with respect to income. The greatest linear income elasticity is carbon dioxide emissions per capita. A 1% increase in income results in 1.62% increase in carbon dioxide emissions hence the exponentially increasing line in Fig. 1.

The elasticities for the local air pollutants follow slightly different patterns, although both are quadratic (Fig. 2). Particulates increase at low incomes (with an elasticity of 0.69), but begin to decline slowly at middle income levels. Once

0 The turning point for CO2 emissions, estimated by Holtz-Eakin and Seldon (1992) of $35,000 in 1985 dollars is also very high and is treated with caution by the authors for similar reasons.

N. SHAFIK 769

TABLE 2 Environmental elasticities-income effects

Low income Middle income High income

Lack of safe water -0.48 -0.48 -0.48 Lack of urban sanitation -0.57 -0.57 -0.57 Annual deforestation Total deforestation Dissolved oxygen Fecal coliform 4.08 -4.20 -0.11 Ambient SPM 0.74 -0.03 -0.70 Ambient SQ2 1.17 0.04 -0.93 Municipal waste per capita 0.38 0.38 0.38 Carbon emission per capita 1.62 1.62 1.62

Notes: The elasticities of environmental indicators with respect to income are calculated for low, middle, and high incomes defined as $900, $3,500, and $11,250 in PPP dollars respectively. These income groups represent the average PPP per capita income equivalents of the World Bank's country classification of low, middle and high income countries. Average low, middle, and high per capita income levels are used to calculate elasticities based on the coefficient estimates of the best fitting model in Table 1.

- Indicates the effects of the right-hand side variable on the environmental indicator are not statistically significant at the 5% level.

countries reach high incomes, the decline is rapid. Sulfur dioxides increase with respect to income at twice the rate of particulates (the elasticity at low incomes is 1.23) and continue to rise, albeit more slowly at middle incomes. At higher incomes, sulfur dioxide concentrations decline more quickly than particulates. Thus the inverted 'U' shape for sulfur dioxides is later and more peaked than for particulates.

Fecal coliform is the only cubic shaped environmental indicator, but the elasticities show that the largest effects are at low and middle incomes. The elasticity of fecal coliform with respect to income is positive at income levels below $1,375, indicating that a rise in income would lead to a rise in the level of fecal coliform (Fig. 2). The rise in fecal coliform is more than proportionate to the rise in income below the per capita income of $1,220 (point A in Fig. 2). Between points A and B the elasticity is positive but inelastic, and therefore an increase in income would lead to a decline in fecal coliform. Between points C and E, a 1% increase in income would imply more than 1% decline in fecal coliform. This improvement in fecal content is greatest as per capita income approaches about $3,950 (point D). However, as per capita income continues to increase beyond $11,400 the elasticity switches sign again and a further rise in per capita income would not improve environmental quality as measured by fecal coliform in rivers. Beyond $12,820 (point G) the elasticity of fecal coliform with respect to income is elastic and positive, which implies worsening fecal coliform levels at the highest income levels.

5. Conclusion

Some very clear patterns of environmental degradation emerge from the previous analysis. Some environmental indicators improve with rising incomes

770 ECONOMIC DEVELOPMENT AND ENVIRONMENTAL QUALITY

(like water and sanitation), others worsen and then improve (particulates and sulfur oxides) and others worsen steadily (dissolved oxygen in rivers, municipal solid wastes, and carbon emissions). The turning points at which the relationship with income changes varies substantially across environmental indicators.

The functional forms seem to reflect the relative costs and benefits that individuals and countries attach to addressing certain environmental problems at different stages of economic development. Water and sanitation, with relatively low costs and high private and social benefits are among the earliest environmental problems to be addressed. Local air pollution, which imposes external costs locally, but is relatively costly to abate, tends to be addressed when countries reach a middle income level. This is because air pollution problems tend to become more severe in middle income economies, which are often energy intensive and industrialized, and because the benefits are greater and more affordable. Where environmental problems can be externalized, as with solid wastes and carbon emissions, there are few incentives to incur the substantial abatement costs associated with reduced emissions and wastes.

But dynamics obviously matter for explaining the patterns-technological innovations alter the cost-benefit calculus at any point in time. The level of technology, as proxied by the time trend, tends to have a positive impact on environmental quality, controlling for the level of income. Figure 1 also shows how the relationships have changed over time using the coefficient on the trend to generate patterns for different years. Some indicators have unambiguously improved over time, such as water, sanitation, particulates, and sulfur oxides. But others, such as fecal coliform in rivers, have unambiguously worsened. Dissolved oxygen and carbon dioxide emissions display no change over time. Thus, where the costs of degradation are local and trigger demand for improvements (such as water, sanitation, and air pollution), technology is critical. Where the costs are diffused or knowledge about detrimental effects is uncertain, such as with carbon emissions, there is little demand for technological innovations that reduce environmental degradation.

The evidence suggests that it is possible to 'grow out of' some environmental problems. But there is not necessarily anything automatic about this-in most countries, environmental improvement has required policies and investments to be put into place to reduce degradation. Further detailed research on both the structural and policy determinants of each environmental quality indicator could show more conclusively the nature of the relationship between economic development and differing environmental policy regimes. The econometric results presented here do seem to indicate that most societies choose to adopt policies and to make investments that reduce environmental damage associated with growth. Action tends to be taken where there are generalized local costs and substantial private and social benefits. Where the costs of environmental degradation are borne by others (by the poor or by other countries), there are few incentives to alter damaging behaviour.

N. SHAFIK 771

ACKNOWLEDGEMENTS

I am grateful to William Cavendish, Sweder van Wijnbergen, David Wheeler, two anonymous referees and participants in seminars at Columbia and Princeton Universities, Swathmore College, and the University of Pennsylvania for their helpful comments. Special thanks go to Sushenjit Bandyopadhyay for his outstanding research assistance.

REFERENCES

ALLEN. J. and BARNES, D. (1985). 'The Causes of Deforestation in Developing Countries', Annals of the Association of American Geographers, 75, 2.

DAVIS, S. (1992). 'Indigenous Views of Land and the Environment', World Bank Working Paper, Washington, D.C.

DIWAN, I. and SHAFIK, N. (1992). 'Investment, Technology and the Global Environment: Towards International Agreement in a World of Disparities', in P. Low and R. Safadi (eds), Trade Policy and the Environment, The World Bank, Washington, D.C.

GROSSMAN, G. and KRUEGER, A. (1993). 'Environmental Impacts of a North American Free Agreement', in P. Gabor (ed.), The US-Mexico Free Trade Agreement, MIT Press, Cambridge, MA.

HOLTz-EAKIN, D. and SELDON, T. (1992). 'Stoking the Fires? CO2 Emissions and Economic Growth', Mimeograph, Syracuse University, Syracuse, NY.

JOHNSON, B. (1991). Responding to Tropical Deforestation. An Eruption of Crisis and Array of Solutions, Conservation Foundation, London; World Wildlife Fund, Washington, D.C.

KNEESE, A. V. and SWEENEY, J. (1985). Handbook of Natural Resource and Energy Economics, North-Holland, Amsterdam.

SELDON, T. and SONG, D. (1992). 'Environmental Quality and Development: Is there a Kuzuets Curve for Air Pollution?', Mimeograph, Syracuse University, Syracuse, NY.

WORLD BANK (1991). 'The Forest Sector: A World Bank Policy Paper', Washington, D.C.

APPENDIX

1. Data sources and definitions

Most of the variables cited here are included in the environmental data appendix to the World Bank's World Development Report, 1992. Because of data limitations, the actual sample size varied depending on availability. Whenever an indicator was not available for all sample countries in the period under consideration, a range is specified below. The sample size for each regression is specified in the last column of the relevant tables in the main text. All variables are in logarithms, unless otherwise specified.

Income per capita. Real per capita gross domestic product in purchasing power parity (PPP) terms were used for the years 1960-88 for 95 to 138 countries (variable RGDPCH in Penn. World Table Mark 5). The chain base method of indexing was used to take into account the changing production bundle over the period. GDP data were not available for all countries for all years. Source: Summers and Heston 1991.

Lack of safe water was measured by the percentage of population without access to safe drinking water. In urban areas access to safe water was defined as access to piped water or a public standpipe within 200 meters of a housing unit. In rural areas, it implies a family member need not spend a disproportionate part of the day fetching water. 'Safe' drinking water includes untreated water from protected springs, boreholes and sanitary wells, as well as treated surface water. Data for this measure were available for only two years, 1975 and 1985, for 44 and 43 countries respectively. Source: World Bank.

772 ECONOMIC DEVELOPMENT AND ENVIRONMENTAL QUALITY

Lack of urban sanitation was defined as percentage of urban population without access to sanitation. Access to sanitation was defined as urban areas served by connections to public sewers or household systems such as pit privies, pour-flush latrine, septic tanks, communal toilets, and other such facilities. Data were available for 1980 and 1985 for 55 and 70 countries, respectively. Source: World Bank.

Annual deforestation reflected the yearly change in forest area for 66 countries between 1962 and 1986. The variable was defined as

log [FA,1 -FAJ

where FA is forest area in thousands of hectares and t takes value 1962-86. Source: World Bank.

Total deforestation was the change in forest area between the earliest date for which substantial data were available, 1961, and the latest date, 1986. The variable was measured as

((FA61 -FA86) * 100) lg

FA61

Total deforestation data were available for 77 countries. Source: World Bank.

Dissolved oxygen data measured in milligrams per cubic meter, were available for 57 rivers distributed in 27 countries for intermittent years between 1979 and 1988. Dissolved oxygen measures the extent to which aquatic life can be supported. Low levels of dissolved oxygen can result from large amounts of industrial effluent or fertilizer run-off from adjacent agricultural land. Source: CCIW 1991.

Fecal coliform data measured in numbers per 100 milliliter, were available for 52 rivers distributed in 25 countries for intermittent years between 1979 and 1988. Fecal coliform measures the level of biological refuse in the river water. High levels of fecal coliform are associated with high incidence of water borne disease in the affected area. Data from five rivers were excluded from the sample due to extremely high reported levels of fecal coliform (exceeding 700,000 per 100 milliliter). These rivers are the Atoyac, Balsas, and Lerma in Mexico, San Pedro in Ecuador, and Yodo in Japan. The effective sample size for coliform was reduced from 434 to 402 observations. Source: CCIW 1991.

Sulfur dioxide. Data on ambient levels of sulfur dioxide measured in micrograms per cubic meter were available for 47 cities distributed in 31 countries for the years 1972 to 1988. Source: MARC 1991.

Suspended particulate matter. Data on ambient levels of suspended particulate matter measured in micrographs per cubic meter were available for 48 cities in 31 countries for 1972 to 1988. Source: MARC 1991.

Municipal solid waste per capita was computed in kilograms, on the basis of available city level information for 39 countries compiled for the year 1985. Source: OECD 1991, and WRI 1990.

Carbon emissions per capita were expressed in metric tons per person per year, for 118 to 153 countries between 1960 and 1989. Source: Marland 1989.

REFERENCES

CCIW (1991). Unpublished data from Canada Centerfor Inland Water, Burlington, Ontario. MARC (1991). Unpublished data from Monitoring and Assessmnent Research Centre, London. MARLAND, G. (1989). Estimnates of CO2 Enmissions from Fossil Fuel Burning and Cemant Manufacturing,

Based on United Nations Energy Statistics and the U.S. Bureau of Mines Cemnent Manufacturing Data, Oak Ridge National Laboratory, Oak Ridge, TN.

OECD (1991). Environmnental Indicators: A Prelimtinary Set, Organization for Economic Co-operation and Development, Paris.

N. SHAFIK 773

SUMMERS, R. and HESTON, A. (1991). 'The Penn World Table (Mark 5): an Expanded Set of International Comparison, 1950-1988', The Quarterly Journal of Economtics. 327-68.

WORLD RESOURCES INSTITUTE (1990). World Resources 1990-1991, Oxford University Press, New York. WORLD BANK (1992). 'World Development Report 1992: Environmental Data Appendix', Development

Economics, Washington, D.C.