1. science
2. ecology
3. pollution

Democracy and Environmental Quality
Y. Hossein Farzin*
and
Craig A. Bond
Abstract
We develop and estimate an econometric model of the relationship between
several local and global air pollutants and economic development while
allowing for critical aspects of the socio-political-economic regime of a
State. We obtain empirical support for our hypothesis that democracy and its
associated freedoms provide the conduit through which agents can exercise
their preferences for environmental quality more effectively than under an
autocratic regime, thus leading to decreased concentrations or emissions of
pollution. However, additional factors such as income inequality, age
distribution, education, and urbanization may mitigate or exacerbate the net
effect of the type of political regime on pollution, depending on the
underlying societal preferences and the weights assigned to those
preferences by the State.
JEL Classification: O13, Q28, H40, D78
Key Words: Political institutions, societal preferences, economic development,
environmental quality
* Y. H. Farzin is professor and C.A. Bond is a PhD candidate at Department of
Agricultural and Resource Economics, University of California at Davis, Davis, CA 95616.
All correspondences should be addressed to Y. H. Farzin, Tel. (530)752-7610, Fax
(530)752-5614, Email: [email protected].
Acknowledgment: For useful comments and suggestions, we especially thank two anonymous referees
and Lant Pritchett, Co-Editor, of this journal. We also thank the participants at the Annual Conference
of Public Economic Theory, PET 04, Peking, China, August 25-29, 2004, the 13th Annual Conference
of the European Association of Environmental and Resource Economists, EAERE, Budapest, Hungry,
June 25-28, 2004, and the 6th Heartland Environmental and Resource Economics (HERE) Workshop,
Iowa State University, Ames, Iowa, September, 19-20, 2004, and at the seminars at Kyoto University
University of California at Davis, and University of Southern California.
Democracy and Environmental Quality
1. Introduction
The examination of the relationship between national income and environmental
quality has become of great interest to economists, policy-makers, and the public at large.
This interest is not only expressed within national boundaries but also is reflected by
growing conflicts between global environmental concerns and global economic
development policy, as seen by frequent uprisings at WTO meetings. Previous literature
has focused on the so-called Environmental Kuznets Curve (EKC), which hypothesizes an
inverted-U shape when pollution indicators are plotted against income per capita.
Explanations for this hypothesis generally focus on several primary factors that interact to
produce the shape. Among these are: (1) changes in the composition of aggregate output as
economies evolve from agricultural to industrial to service-based goods and services, (2)
technological progress, and (3) increases in demand for environmental quality as income
grows (Anderson and Cavendish, 2001; Ansuategi and Escapa, 2002; Grossman and
Kruger, 1995; Heerink, et al., 2001; Panayotou, 1997).
However, as most authors tend to agree, the relationship between environmental
quality and economic development is not formed in isolation from political institutions that
govern the process of policy making in a particular country. Thus, for example, Dasgupta
and Mäler (1995, P. 2412) have aptly emphasized: “The connection between environmental
protection and civil and political rights is a close one. As a general rule, political and civil
liberties are instrumentally powerful in protecting the environmental resource-base, at
least when compared with the absence of such liberties in countries run by authoritarian
regimes”. This observation raises several important questions: How does pubic
environmental policy influence the relationship between per capita income and pollution,
and how does that public policy represent the citizens’ preferences for environmental
quality?
Several authors have included explanatory variables to explicitly account for the
role of political institutions in the income-environment relationship. For example, in their
study of the causes of deforestation in Latin America, Africa, and Asia, Bhattarai and
Hammig (2001) use a measurement of institutional quality, measured by an index of
political rights and civil liberties, to account for the role of different policy regimes. Torras
1
and Boyce (1998) use a similar technique for a panel data survey of a variety of air and
water pollution indicators measured as ambient concentrations of toxins. However, these
studies did not explicitly decompose the demand for environmental policy based on
heterogeneous population characteristics, nor did they address the potential for distortions
of the environmental quality preferences by the political regime.
This paper proposes an empirical econometric model of the relation between several
local and global air pollutants and economic development, measured by national income
per capita. The model explicitly accounts for critical aspects of the socio-politicaleconomic regime of a State. Specifically, we develop a model that directly incorporates the
relationship between societal preferences and provision of public pollution abatement,
utilizing a measure of quality of governance as a proxy for weights on those preferences.
We hypothesize that democracy and its associated freedoms provide the conduit through
which agents can exercise their preferences for environmental quality more effectively than
under an autocratic regime, thus leading to reduced concentrations and/or emissions of
pollution. However, additional variables such as income inequality, age distribution, and
urbanization may mitigate or exacerbate the net effect of the type of governance on
pollution, depending on the underlying preferences of the population at large and the
weights assigned to those preferences by the State.
Section 2 provides a reduced-form model of the relationship between
environmental preferences, the degree of democratization of political institutions, and
realized environmental policy. Section 3 describes the structure and key variables included
in the empirical models, and the likely effects of several socio-economic variables on
demand for environmental quality. Section 4 describes the data used in the analysis and the
estimation issues that arise. Section 5 presents and discusses the results.
Section 6
concludes.
2. Relationship between Environmental Policy, Governance, and Preferences
One of the major determinants of environmental policy is the political regime of a
particular country, or “governance” as phrased by Rivera-Batiz (2002). Specifically, Lopez
and Mitra (2000) argue that corruption and rent-seeking behavior can influence the
relationship between income and the environment. They provide a theoretical model that
2
shows corruption causes the turning points of an EKC to rise above the socially optimum
level. Additionally, Magnani (2000) suggests that well-defined property rights, democratic
voting systems, and respect of human rights can create synergies that lead to increased
levels and efficacy of environmental policy.
We propose a simple explanation of the role of political structure on the relationship
between income and environmental quality, based on the relationship between the demand
and supply of environmental quality. Because environmental quality is mostly a public
good, and in many cases the capital costs of the required infrastructures to abate pollution
are huge, individuals or groups within a society are unable to effectively provide them. 1 As
such, it is usually the state that provides these goods. However, the state’s environmental
policy is at least partly influenced by the society’s preferences for environmental quality. In
particular, the relationship between the public’s preferred level of environmental quality
and the level actually supplied by the state depends on the weights the policymakers place
on the various heterogeneous societal preferences. These weights can be thought to reflect
a competitive political equilibrium of the interplay between the state and various citizens
groups with a stake in environmental policy, and reflect the degree of democratization and
quality of political institutions in place. 2
To illustrate, consider the following simple model. Suppose that the population of N
individuals in society can be aggregated into two distinct groups, i = 1, 2, based on some
distinguishing characteristic, denoted by γ i . The categorical distinction γ i can be made
along any number of characteristics that are likely to affect environmental preferences; for
example, “rich vs. poor”, “educated vs. non-educated”, “urban vs. rural”, or “young vs.
old”. In the case of “rich vs. poor”, γ i represents the income of the ith group, while in the
other cases, it represents the number of individuals in each category. The citizens groups’
environmental preferences are assumed to be reflected by their preferred levels of pollution
abatement expenditure to be undertaken by the state.
It is reasonable to think that in deciding how much abatement efforts to undertake at
a particular time t (subscript omitted hereafter for clarity), the State may take the
1
As opposed to private environmental quality, say, in the home as a result of inadequate ventilation.
Here we do not intend to present a formal game-theoretic model that could underlie such a political
equilibrium. For examples of such models, the interested reader may refer to seminal work of Persson and
Tabellini (2000) and references cited therein.
2
3
preferences of different citizen groups into account. However, the extent to which it does
so depends on the democratic structure of political institutions. Formally, let
E1 (γ 1 ) and E2 (γ 2 ) denote the citizen groups’ preferred abatement levels, and Ed the State’s
own preferred (discretionary) level. Note that these preferences may, in fact, vary by
pollutant type, so that group γ i may prefer more abatement than does group γ j for, say,
greenhouse gasses, but the reverse might be true for primarily ozone-contributing
emissions. The realized (or actual) abatement level provided through the State, Eˆ , may be
expressed as a function of the preferred abatement levels and the State’s degree of
democratic representation of the citizens’ preferences, i.e.
Eˆ = E ( E1 (γ 1 ), E2 (γ 2 ), Ed ; α ) .
(1)
Here the vector of parameters α = (α1 , α 2 , α 3 ) can be viewed as reflecting the type
of the existing political structure and processes that determine the political equilibrium
weights of the citizens groups and the State in the environmental policy outcome.
Accordingly, α1 and α 2 represent the equilibrium weights of citizens groups’ preferences
in the final decision (reflecting the degree of democratic representation of citizens’
preferences), while α 3 may be interpreted as the weight of the State’s own preferences for
abatement level (reflecting the degree of State’s autocracy). 3,4 This formulation is in line
with the tradition of public choice school, which emphasizes the importance of lobbying
and rent-seeking activities by organized interest groups in shaping public policy (see
Mueller (1997) for a survey of the literature). It is also in line with the tradition of rational
choice school, which treats the State’s as the agent of citizen voters and views public
decisions as outcomes of rational individuals’ choices, but recognizes that, due to free-rider
problems in the democratic electoral process itself, rational voters may fail to choose
3
Technically speaking, the parameters α1, α2, and α3 determine, at the margin, the rate at which the State
trades off one citizen group’s preferred abatement level for that of the other group and of its own preferred
level when deciding to provide a target level of abatement expenditure.
4
It should be noted that relationship (1) holds at each point in time. In particular, the weights of the State’s
and citizen groups’ preferences in deciding about public abatement expenditure may well change over time
for various reasons. For example, as the State’s environmental knowledge (including information on the
sources of harmful pollutants, their relative degree of threat to human communities and ecosystem, and the
actual degree of exposure suffered by victims of pollution) might evolve and interact with citizen groups’
preferences and knowledge. Similarly, the form and quality of political institutions may change over time.
However, the question of how such changes may alter the weights is a complex issue and beyond the scope of
this paper.
4
politicians who will serve their interests. This school of thought uses the tools of noncooperative game theory to study collective choice within specific political institutions
(see, for example, Ordeshook (1986)).
A linear functional form for this simple model results in the specification
Eˆ = E ( E1 (γ 1 ), E2 (γ 2 ), Ed ; α ) = α1 [ e1 (γ 1 )γ 1 ] + α 2 [ e2 (γ 2 )γ 2 ] + (1 − α1 − α 2 ) Ed ,
(2)
where ei (γ i ), i = 1, 2, is the ith group’s average propensity for public expenditure on
pollution abatement. From Equation (2), three extreme political structures can be
immediately distinguished.
One extreme case is when a political structure is totally
corrupt, so that one interest group or the other captures the State’s environmental policy.
Such a case is presented either by α1 = 1 and α 2 = α 3 = 0, or by α 2 = 1 and α1 = α 3 = 0. At
the other extreme, the existing political structure and institutions may be totally
democratic, so that the state’s policy is solely determined by the citizens’ preferences, i.e.
α 3 = 0. As a special case of this, one may consider an egalitarian democracy, represented
by α1 = α 2 = 1/ 2 and α 3 = 0. In reality, of course, there is a wide spectrum of alternative
political structures, each of which can be thought to imply a specific set of political
equilibrium weights α = (α1 ,α 2 , α 3 ) . In particular, even in democratic countries, the role of
the State is rarely limited to a merely benevolent public agency. For various reasons, such
as asymmetry of information between the citizen groups and the State about the details of
the State’s specific policies, there may be a divergence between politicians’ incentives to
present public interests and protect their own private interests, implying 0 < α 3 < 1 .
Noting that the aggregate characteristic γ = γ 1 + γ 2 , we have from Equation (2)
Eˆ ( ⋅)
γ
= [α1e1 (γ 1 ) − α 2 e2 (γ 2 ) ]
E
γ1
+ α 2 e2 (γ 2 ) + (1 − α1 − α 2 ) d .
γ
γ
(3)
The left-hand side of Equation (3) can be interpreted as public abatement per unit of
characteristic; for example, public abatement expenditure per dollar of national income in
the case of an income measure, or abatement expenditure per person in the case of a
personal characteristic. The change in abatement per unit as society develops along a
particular attribute is then given by
5
(
∂ Eˆ / γ
) = α e (γ ) − α e (γ ).
∂ (γ 1 / γ )
1 1
1
2 2
2
(4)
Note that the direction and magnitude of this change depend crucially both on the
intensities of preferences and the weights that represent the existing political structure and
the state’s sensitivity to incidences of a particular pollutant. For example, assume that the
population, γ , of a particular society consists of γ 1 number of “educated” citizens and
γ 2 = γ − γ 1 number of “uneducated” ones, and that the educated category has a greater
average propensity for environmental quality, e1 (γ 1 ) > e2 (γ 2 ). In that case, it follows from
(4) that the effect of an increase in the education level of the population, as measured by
γ1
, on per capita abatement expenditure will be given by
γ
⎧> 0 for an egalitarian or pro-educated democracy ⎫
∂ ( Eˆ / γ )
⎪
⎪
= α1e1 (γ 1 ) − α 2 e2 (γ 2 ) ⎨<0 for a strongly pro-uneducated democracy
⎬ . (5)
∂ (γ 1 / γ )
⎪=0 for a complete autocracy
⎪
⎩
⎭
Similar results also hold for the effects of changes in other socio-economic
attributes (such as income distribution, degree of urbanization, and age distribution) of the
population. For instance, in the case of income distribution, if we assume that the rich
people have a greater average propensity for pollution abatement than the poor, then it can
be shown that an increase in income inequality (as measured by a larger income share of
the rich in national income) will lead to a lager abatement expenditure/income ratio for a an
egalitarian or pro-rich democracy, whereas it will result in a lower ratio for a strongly propoor democracy, or will have no effect for a totally autocratic society.
While highly stylized, this model of public environmental quality provision
highlights the importance of the type of political regime for the relationship between
economic development and environmental quality. It suggests that environmental quality
expenditures are partly a function of the citizens groups’ preferences, but these preferences
are subject to political distortions, misrepresentation, or neglect by the State. The more
open and democratic are the political institutions, however, the more likely it is that the
citizens’ preferences will be reflected in actual policy decisions, and environmental quality
6
as a normal public good will increase. The simple model also facilitates the economic
interpretation of some of the econometric results that are reported in Section 5.
3. Preference Shifters and the Empirical Models
The econometric models use a reduced-form parametric approach that implies no
feedback from environment to economy, with the structure allowing for flexibility via
higher-order terms in GNP per capita and population density (Grossman and Kruger, 1995;
Coondoo and Dinda, 2002). On the demand side, the preferences described in the preceding
section are incorporated through the addition of structural variables for the quality of
governance and interactions with the socio-economic characteristics hypothesized to affect
the demand for environmental quality.
Following Panayotou (1997), we begin the decomposition by assuming that policy
can be instrumented solely by the democratic/autocratic structure of political institutions of
the country, resulting in the following model
3
3
j =1
k =1
X it = κ i + ∑ β jYitj + ∑ β k +3 Pitk + β 7 Dit + β8 DitYit + δ t + ε it ,
(6)
where, for country i in year t, Xit denotes the pollutant measure, Yit is GDP per capita, Pit is
population density, and Dit is the democracy measure. The variable t is a linear time trend
to account for technological change over time, βj and δ are regression coefficients, κ i ’s are
country-level fixed effects and εit is an error term. Note that the political regime
variable, Dit , enters both additively and multiplicatively, allowing for non-constant
marginal effects on pollution measures. Equation (6) constitutes a benchmark in the
analysis, and is used to test the EKC hypothesis and the effect of political institutions on
environmental quality. More specifically, results from this specification decouple the direct
effects of income and political regime on the pollution measure, separating the pollution
trend into more-or-less “automatic” responses and choice-based policy responses, the latter
approximated by a regime’s degree of democracy. 5
5
As pointed out by an anonymous reviewer, the extent to which income itself might affect political regime,
and more generally any of the other regressors, is omitted from the analysis here. These indirect effects could
be captured by additional structural equations in a suitably specified theoretical model. However, this is
outside the scope of the current analysis.
7
Equation (6) is subsequently expanded to include variables expected to be
correlated with the heterogeneous preferences of society at large, thus explicitly
incorporating some of the potential determinants of demand for environmental quality into
the specification. It is assumed that these preferences can be exercised only through the
political system of the state, and as such the democracy variable, Dit, is used to interact with
each shifter. This specification implies that strongly autocratic regimes will be
unresponsive to societal preferences relative to more democratic societies, with the
magnitude of the coefficients reflecting the actual weights ( α ) assigned to citizens groups’
preferences ei (γ i )
Formally decomposing Equation (6) to account for the preference shifters results in
the following model:
3
3
j =1
k =1
X it = κ i + ∑ β jYitj + ∑ β k +3 Pitk + β 7 Dit + β8 DitYit + β9U it + Dit ( β10Git + β11 Ait
(7)
+ β12U it + β13 I it ) + DitYit ( β14Git + β15 Ait + β16U it + β17 I it ) + δ t + ε it ,
where Uit is a measure of urbanization, Git is the GINI coefficient measuring inequality, Ait
is the proportion of the population under age of 15, Iit is the illiteracy rate of the adult
population as a proxy for the education level, and all other variables are as previously
defined. We maintain the assumption that the measure of democracy and preference
shifters can affect both the intercept and slope of the income-environmental quality (EKC)
curve, resulting in a significantly flexible functional form with marginal effects that may
depend on the current state of development.
It should be emphasized that both the basic and expanded estimating Equations (6)
and (7) are reduced-form equations. As such, they only estimate the net effects of the
explanatory variables without tracing through the specific channels of the effects. Further,
it should be noted that the simple theoretical model of section 2 relates to the estimating
equations by relating the political equilibrium weights, α ’s, and the environmental quality
propensities, ei (.) , to the expected signs and magnitudes of the coefficients in the latter
equations. For example, a relatively large weight on preferences of educated people,
coupled with a higher average propensity for environmental quality compared to those of
uneducated people, implies a positive sign for the derivative of the per capita abatement
expenditure with respect to the ratio of educated people to total population
8
⎛ ∂ ( Eˆ / γ )
⎞
= α1e1 (γ 1 ) − α 2e2 (γ 2 ) ⎟ > 0 . Linking this to the relevant coefficients in equation (7)
⎜
⎝ ∂ (γ 1 / γ )
⎠
(i.e. β13 and β17 ), we expect a net positive association between the illiteracy variable ( I )
and
the
measure
of
pollution
concentration/emissions
( X it )
(i.e.,
∂X it
= β13 Dit + β17 DitYit > 0 ). Furthermore, we expect a positive correlation between the
∂I
marginal (negative) effect of democracy on pollution emissions and illiteracy
(
∂ ⎛ ∂X it ⎞
⎜
⎟ = β13 + β17Yit > 0 ); that is, a higher level of education (lower level of I )
∂I ⎝ ∂Dit ⎠
reinforces the pollution mitigating effect of a more democratic political structure.
To facilitate economic interpretations of the econometric results that follow, we
now discuss the likely effects of several socio-economic factors on citizen groups’
environmental preferences, which may in turn influence environmental policy and thus
environmental quality outcomes. These so called “preference shifters” include income
inequality, the age composition of population, degree of urbanization, and education level.
Income Inequality and Environmental Quality
The relationship between income inequality and environmental quality is unclear.
According to one argument, environmental quality is primarily a concern of well-to-do (the
upper and upper-middle income) groups who have already satisfied their basic needs and
enjoy relatively high living standards. It is held that for these groups, environmental
preferences are largely focused on conservation of environmental amenities to support their
life styles. At least in rich democratic countries, this view seems to be supported by
statistics on membership in environmental organizations, visits to parks and recreational
areas, and on participation in activities such as hunting, fishing, backpacking, bird
watching, and hiking. Coupled with the argument that the rich groups are politically more
active and more able to influence politicians than are the poor, this view suggests that as
the share of high income-groups in total income rises (i.e. as income inequality increases)
one may expect improvements in quality of environmental amenities valued and demanded
mostly by the rich people. Scruggs (1998) finds empirical support for this view, showing
that higher levels of education and wealth, implying greater income inequality, are
associated with “pro environment” preferences.
9
A contrasting argument holds that poverty forces the poor to choose employment
over environmental quality, as evident by their choices of occupations and residential
locations that are considerably more exposed to industrial toxic pollutants and wastes.
Increased employment opportunities for the urban poor are often associated with the
emergence or growth of highly polluting industries. Accordingly, to the extent that the
growth of these industries increases urban employment and lowers income inequality, one
may expect a negative correlation between income inequality and measures of certain types
of urban pollution. Of course, this negative correlation will be less strong if politicians are
particularly sensitive to the adverse impacts of urban (industrial) pollutions on the wellbeing of the urban poor, and if they respond to them by undertaking severe pollution
control measures. Conversely, it will be stronger if politicians are indifferent to the plight
of pollution victims, either because the victims are unable to effectively organize or
because of absence of (or lack of access to) channels to express their environmental
demands and influence policymakers. 6 Empirical evidence of a negative correlation
between demand for environmental quality and individual’s income and/or power due to
ownership and consumption patterns is provided in Torras and Boyce (1998). Further
empirical support for this hypothesis are presented by Bimonte (2002) with respect to
public lands, and by Magnani (2000) who finds a higher income inequality to be associated
with a lower public expenditure on pollution abatement.
On the other hand, Heerink, et al. (2001) question the existence of any general
relationship between income inequality and environmental quality, arguing that it is the
convexity and concavity properties of the individual household EKC that determine the
marginal effects of income redistribution when they are aggregated to examine the effect at
the macroeconomic level.
Age Composition of Population and Environmental Quality
Several arguments and counter arguments blur the effect of the age distribution of
population on environmental politics and environmental quality. Some of the arguments
suggesting a positive correlation between population youth and environmental quality
include: (1) Because young people expect to live much longer than old people, they have
6
Mueller and Stratmann (2002) find that the government serves the interests of the upper classes in Latin and
Central American countries with weak democratic institutions.
10
larger stake in environmental quality, and therefore may have stronger preferences and
demand for public investment in long term environmental quality than do the old. (2)
Young people are better educated about the environment and are more aware of
environmental political issues. (3) Sports and outdoor activities are strongly associated with
young people and are complements with environmental amenities. (4) To be “green” or an
environmental advocate is regarded by the young to bring reputation. In other words,
environmentalism is regarded like a social value or cause in itself, with which the young
people identify themselves more than the old people do. (5) As a consequence of the
arguments (1)-(4), the young seem to be more vocal about their environmental preferences,
and, given their superior networking capabilities, they seem to more actively participate in
environmental legislative and regulatory processes through organizing environmental
pressure groups (green NGO’s) or membership in green parties or environmental advocacy
groups.
The counter arguments include: (i) The young can probably bear more risk of
environmental health hazards than can the old. Stated differently, the old are more
vulnerable to environmental health hazards than the young, implying that the young may
have a larger option value of waiting for future environmental quality improvements and a
relatively higher environmental discount rate to apply to future damages from
environmental pollution. (ii) Even if, because of their short life time horizon, the old may
not care as much about long term environmental quality as the young do, they still may
have intense environmental preferences because their strong desire to leave a better
environmental quality for their children and grandchildren (intergenerational environmental
altruism). (iii) At least in western democracies, older people may be financially better off.
So, even if they do not directly participate in environmental political activities, indirectly
they can influence environmental policies through their financial support for environmental
activist groups. (iv) Older people have relatively more spare time to spend on
environmental political activities, particularly on organizational networking, using
information technologies and pre-existing networks.
There is surprisingly very little in the empirical literature regarding the age
distribution effects. One notable exception in this regard is the work of Ono and Maeda
(2001), who study the effects of an aging population on environmental quality. They
11
conclude, based on the interaction of income and substitution effects from a change in
longevity, that the marginal impact of age distribution on environmental quality depends on
the risk-aversion properties of individual agents’ preferences. Particularly, if relative risk
aversion with respect to consumption is less than one (i.e. implying that the population is
not terribly averse to intergenerational inequality in consumption), then aging may be
beneficial to the environment. Yet, as alluded to in the introductory remarks, the recent
demonstrations at WTO meetings to protest the environmental consequences of
globalization have been largely organized and led by the young environmentalist groups.
Based on casual observations about the dominant presence of young people in
environmental advocacy groups, and in view of the arguments (1)-(5) noted above, we
hypothesize that a younger population tends to have a larger propensity to demand
environmental quality.
Education Level of Population and Environmental Quality
There are three main arguments for a possible positive association between the level
of education in a country and the indicators of environmental quality. First, the educated
are more aware of environmental issues and thus are likely to have more intense
preferences for environmental quality and behave more consistently with protection of the
environment. Bimonte (2002) finds strong positive correlation between the level of
education and the demand for environmental amenities. Second, those with more education
are more willing and able to use the existing channels to express their environmental
preferences, organize advocacy groups or participate in them, and thus be vocal about their
demands for public environmental policies. Third, a more educated population is more
likely to generate an environmentally progressive civil service, and therefore have
democratically-minded public policymakers and organizations that are more receptive to
public demands for environmental quality. Rivera-Batiz (2002) provides empirical
evidence in support of this hypothesis. Because these arguments reinforce each other, it
seems reasonable to expect lower pollution levels to be associated with higher education
levels.
Urbanization and Environmental Quality
On purely theoretical grounds, urbanization can have a mixed effect on
environmental policy making and hence on environmental quality. On the one hand, a
12
greater degree of urbanization is likely to be associated with a greater per capita and per
unit area consumption of fossil fuels and industrial chemical inputs, thus increasing the
concentration of toxic chemicals and pollutants including CO2, SO2, NOx (Panayotou,
1997). On the other hand, urbanization has been a reason for emergence of citizens’
environmental action groups to influence politicians and policymakers to mitigate urban
pollution. In fact, as documented by Miller (2002) for the United States, many of the
environmentalist groups and movements formed in the past few decades (in the forms of
environmental organizations, alliances, and coalitions) have been in response not so much
to the traditional conservationist concerns of the rich, suburban groups but primarily to
urban environmental issues such as clean air, water, and land, which have been the central
environmental concerns of the inner-city poor and middle–income populations.
Furthermore, urbanization brings pollution victims into close contact with policymakers
and facilitates channels through which the environmental preferences of urban population
can be voiced out to government agencies. This is partly because urbanization lowers both
the transportation costs and transaction costs of people getting together to coordinate their
actions and organize groups to signal their environmental preferences to responsible state
agencies (Rivera-Batiz, 2002). This tends to increase the weights of citizen groups’
preferences in public policy making. It also generates economies of scale effects with
regards to pollution abatement (Torras and Boyce, 1998). We include urbanization in the
expanded version of Equation (6) to capture the foregoing effects and to account for
preference differences between rural and urban populations.
4. Data and Estimation Issues
We estimate the regression models for a number of diverse environmental quality
indicators for air quality, using data from several sources. Anthropogenic CO2 is a
greenhouse gas emitted from the burning of fossil fuels and cement production, and
emissions data is estimated by the United Nations for over two hundred countries for the
period 1980-1998 (United Nations, 2003; World Bank, 2003, United States EPA, 2003).
These estimates are derived by the Carbon Dioxide Information Analysis Center, based
primarily on energy statistics collected by the United Nations and reported in the Statistical
Yearbook (Marland, et al., 2003; United Nations, 2004). As such, this dataset may
13
incorporate considerable measurement error in emissions, and might more accurately
reflect fossil fuel consumption. Nevertheless, the breadth of the sample, especially with
regard to the variance in political institutions, is attractive, although the authors urge
caution in interpretation of results derived from this data series.
Under the 1979 Convention on Long Range Transboundary Air Pollution, a number
of data series are collected through the Programme for Monitoring and Evaluation of the
Long-Range Transmission of Air Pollutants in Europe (EMEP), including emissions of
nitrogen oxides (NOX), non-methane volatile organic compounds (VOC), and sulfur
dioxide (SO2) from 1980 through 1996. 7 These chemicals contribute to ground level ozone
(smog) from industrial and transportation sources, with SO2 concentrations deriving from
primarily fixed-point industrial sources and also contributing to acid rain. There are
currently 49 parties to the monitoring agreement, mostly European and North American
countries. As such, the political structures are more homogeneous in these series, but the
reliability of the emissions estimates are likely considerably higher.
Finally, ambient SO2 concentration data, as opposed to SO2 emissions data, from
the Global Environmental Monitoring System/Urban Air Pollution Monitoring and
Assessment Programme (EPA, 2002), is used in the analysis. The data series contains an
unbalanced panel data set for mean ambient concentrations of SO2 for different monitoring
stations in forty-five countries over the period 1972 – 1994. We use the median
concentration for each country in each year, as in Panayotou (1997), for comparability with
the other pollutants, as well as minimizing bias due to the relatively larger number of
stations in high-income countries.
We next turn to the independent variables used to estimate the equations. The
political regime variables are taken directly from the Polity IV Database (2003), a project
of the Integrated Network for Societal Conflict Research Program, Center for International
Development and Conflict Management, University of Maryland. 8 Polity IV contains,
amongst
many
other
variables,
yearly
composite
indicators
measuring
both
“institutionalized democracy” and “autocracy” for just about every independent nation with
a population over 500,000 on an additive eleven-point scale. Institutionalized democracy is
7
This data was originally retrieved from World Resources Institute (2003), and is also available on the EMEP
website at http://www.emep.int/index.html.
8
See www.bsos.umd.edu/cidcm/inscr/polity.
14
defined on the basis of weighted measures of political participation, openness and
competitiveness of executive recruitment, and constraints on the chief executive, while
autocracy is a similarly weighted measure of those variables plus regulation of
participation. A summary “polity” measure is then defined as the difference between the
democracy and autocracy scores, with 10 indicating “strongly democratic” and –10
indicating “strongly autocratic”.
This paper uses the polity measure plus ten as an
explanatory variable indicating the political regime, with increasing values indicating
greater levels of democratic freedom over both time and between nations.
The World Bank provides the balance of the data used in the analysis, including the
WDI for GDP, population density, illiteracy, urbanization, and share of the total population
under 15. The income measure is defined as gross domestic product per capita in current
dollars, adjusted into PPP terms to account for relative purchasing power in each nation.
The illiteracy rate is used as an (inverse) proxy for the education level of the general
populace, and the share of total population under the age of fifteen years captures
differences in preferences as a result of the demographic structure of the population.
Finally, income inequality is proxied by the updated estimates of Gini coefficients
compiled by Deininger and Squire (1996) 9 . Summary measures of the variables, as they
enter the empirical model, are provided in Table 1.
Issues with regard to specification, estimation, and generalizations abound, as
evidenced by the special issue of Ecological Economics (1998) dedicated to the EKC.
Empirically, Stern and Common (2001) argue that evidence supporting the EKC hypothesis
may be sample-specific and dependent on estimation methods, some of which may lead to
omitted variable bias in the parameter estimates. Significant differences have also been
found between cross-section studies and those that focus on individual countries (Roca, et
al., 2001; Vincent, 1997).
In addition, there are several methodological issues that must be addressed in the
estimation procedure, including regressor endogeneity due to parameter heterogeneity,
measurement error, and multicollinearity amongst regressors. Given that we are lucky
enough to have a panel data set, the endogeneity and omitted variable problem for time9
As this dataset is incomplete, a variety of techniques were applied to fill in the missing data, including
truncated linear interpolations and regression of the included GINI values on the rest of the exogenous
variables in the system. The results are quite robust to both methods.
15
invariant variables is handled through estimation of a fixed-effects (or within) model, in
which the models described by Equations (6) and (7) are estimated via ordinary least
squares on deviations from the mean data values for dependent and independent variables
(Greene, 2000). The parameter estimates are thus calculated from variance within groups,
rather than between groups, and group-specific intercept terms are fixed parameters to be
estimated.
Measurement error is most severe for the Gini coefficients, which are relatively
incomplete in a time-series sense in relation to the dependent pollutant data. Ideally, one
would like to instrument the inequality variables to avoid potential correlation with the
error term in the parametric regression. However, data limitations here prevent such an
exercise. Instead, assuming that inequality admits a small variance over time and the series
are relatively smooth over time, linear interpolation is used between interior years to
complete the series, while incomplete years on either end of the time scale are held
constant to the closest year’s actual value. The latter technique is employed in order to
minimize the introduction of arbitrary trends outside of the original data set. 10 As with
many macroeconomic level socio-economic indicators, there is also the potential of
measurement error for the dependent and other independent variables, augmented by the
large number of interaction terms on the right hand side of the equations. Such
measurement error has the potential to render estimated coefficients inconsistent and
potentially biased (Greene, 2000).
Finally, multicollinearity is a problem resulting from the introduction of the power
terms in the parametric regressions, interaction terms between the independent variables,
and partial collinearity of the explanatory indicators, leading to inflated standard errors on
the highly collinear terms. As such, model specification is performed using orthogonalized
independent variables with the correlation to lower-order terms removed through auxiliary
regression on deviations from means. 11 For hierarchical completeness, insignificant lower
order terms are preserved whenever higher order terms prove significant. Once a
specification has been chosen, the within model on the original data is estimated and
10
Of course, any procedure to fill in missing values may introduce either bias or inefficiency into the
coefficient values or standard errors. However, the results reported below are robust to alternative techniques,
such as regression of the known GINI values on the exogenous variables of the system. Nevertheless, the
reader is advised to be cautious.
11
Results of these auxiliary regressions are available from the authors.
16
reported in the subsequent tables below. Given these potential statistical issues, the reader
is encouraged to be cautious in interpreting the econometric results.
5. Results
We begin by examining the results of Equation (6), as reported in Table 2. As in
Panayotou (1997), this specification assumes that environmental policy can be represented
solely by the polity variable, which captures the quality of institutions and openness of the
state to the environmental preferences of the populace. This allows us to test the hypothesis
of an inverted-U shaped relationship between the direct effects of national income per
capita and pollution indicators conditional on the institutional regime, population density,
and technology as proxied by the time variable, as well as the marginal effect of increasing
the quality of the public institutions on environmental quality. The lack of a change in sign
of the slope of the income-pollution curve suggests that economic growth alone, as
measured by a change in GDP, is insufficient to improve environmental quality. Rather,
conscious environmental policy emanating from the existing political institutions, as
represented by the polity variable, is necessary. Note, however, that this conclusion need
not necessarily hold if a structural relationship between income and political regime (i.e.,
increased growth improves quality of institutions) is assumed.
Of the five regressions in Table 2, only one (emissions of non-methane volatile
organic compounds) supports the EKC hypothesis of an inverted-U shaped relationship
once the proxy for environmental policy is included. The turning point for the VOC curve
is negatively related to the quality of political institutions (through the GDP*Polity
interaction variable), but tends to occur at GDP levels at the upper end of the distribution.
Emissions of the ozone-causing NOX and SO2 exhibit a cubic relationship with large,
positive slopes at low and high levels of income, but smaller marginal effects at moderate
levels, possibly turning negative for SO2 for highly democratic societies. The income and
CO2 emissions relationship is unambiguously positive and linear, while the curve for
ambient SO2 concentrations is increasing and convex. These results support the findings of
the previous literature; namely, growth in income per capita is not sufficient for increases
in pollution abatement as nations develop. As emphasized in the statement by Dasgupta
and Mäler (1995) quoted in the introduction, conscious choices of environmental policy
17
emanating from civil rights to express preferences are the key to understanding the
relationship between economic development and environmental quality.
Testing the effects of the policy proxy on the estimated relationship corroborates
this hypothesis. In all cases, the marginal effect of the polity variable with respect to the
pollutant is negative for the majority of the income range under consideration (more than
the 25th percentile of the sample data), suggesting that countries with more democratic
institutions have a greater tendency to reduce pollution. For those pollution measures for
which the effect is dependent on income levels (excluding NOx), the marginal effect of
democratization is intensified with income, as seen from the interaction term
(
∂ ∂X it
< 0 ). At very low levels of income for four of the five regressions, however, the
∂Yit ∂Dit
estimated marginal effect of democratization of political institutions could be positive
(
∂X it
> 0 ). This implies that, in very low–income countries, for most policy regimes, the
∂Dit
State and the people assign such a high priority to industrial development that despite
increased democratization of institutions and its positive impact on citizens’ opportunities
to voice out their environmental preferences, pollution emissions increase. However, this
effect is lessened as income per capita rises (
∂ ∂X it
< 0 ).
∂Yit ∂Dit
Recognizing that environmental policymaking considerations are of paramount
importance in describing the relationship between economic development and the
environment, we turn now to the results of the model described by Equation (7). This
model decomposes the environmental policy indicator variable in order to account for both
the heterogeneous preferences of the society and the mechanism through which these
preferences are translated into realized pollution abatement. As with the previous model,
the unrestricted full specification on orthogonalized data was estimated first, with
subsequent restrictions imposed given the results of individual and joint significance tests,
subject to hierarchy considerations.
Results of the restricted model are reported in Table 3, while Table 4 and Table 5
report point elasticities from the final model. Table 4 presents elasticities of each pollutant
with respect to the relevant variable evaluated at the estimating sample means, which differ
18
for each pollutant. Table 5 illustrates the relative magnitudes of the effects of polity and the
other preference shifters by reporting elasticities of the pollution measures with respect to
per capita income evaluated at mean, low (15th percentile), and high (85th percentile) levels
of the explanatory variables. 12 Figure 1 displays many of these calculations graphically.
As expected, the relationship between GDP per capita and the pollution measures is
similar to that estimated in the basic model, although the NOx model now admits an EKC
relationship with a turning point well outside the sample range of income when evaluated
at sample means for the preference shifters. Nevertheless, the similarity in conditional
results suggests that the decomposition is valid, and that demand considerations based on
heterogeneous societal preferences are an important determinant of overall environmental
quality. The estimated effects of the polity and preference interactions on the pollutionincome relationship can be seen in Table 5, and are generally realistic and fairly stable,
except for instances in which correlations are unlikely (for example, high polity scores and
high illiteracy percentages). Income growth conditioned on greater polity scores is
predicted to have a relatively smaller impact on increased emissions in four out of five
cases at sample means of the preference shifters, and except for CO2, this relationship
carries over to more extreme values of those variables. In fact, in some cases, the elasticity
with respect to income turns negative for high polity values. Furthermore, the marginal
effects of increasing polity on pollution remain negative at the mean sample values (Table
4), with elasticities ranging from a low of -0.08 for CO2 to a high of -2.51 for ambient
concentrations of SO2. Note from Table 1, however, that the samples used to estimate each
of these models differ considerably, and the interaction terms allow for heterogeneous
marginal effects depending on the state of the environmental preference shifters.
We now turn to the effects of the individual preference shifters on the pollution
indicators, conditional on the state of the institutions in place in a given society. As seen in
Tables 3 and 4, urbanization has an unambiguous net positive effect on all pollution
indicators with the exception of CO2 emissions at high levels of democracy and national
12
Elasticities defined as (∆Xi/Xi)/(∆GDPi/GDPi) for each pollutant i, with Xi set at the sample mean for each i.
In the first two numerical rows of Table 5 only the value of Polity varies; e.g., 2.541 is the elasticity of
GEMSSO2 with respect to GDP evaluated at the 15th Polity percentile, with other variables set at their sample
means. For the remaining point estimates, beside Polity, one preference shifter varies too; e.g., the elasticity
of CO2 w.r.t. GDP with Urbanization and Polity both set at their respective 15th sample percentiles, and other
variables at their sample means, is 0.596.
19
income. This suggests that the effects of increased fossil fuel use in urban societies mostly
dominate any economies of scale or preference effects. However, some preference effects
are evident through the negative interaction terms for CO2, NOx, and SO2 emissions,
though they are far from uniform.
Another abatement demand shifter widely discussed (and disputed) in the literature
is income inequality, as it is hypothesized that the distribution of income may play a role in
the income/environment relationship. In this application, the proxy for income inequality,
the GINI coefficient, is found to have a negative relationship with environmental quality in
three of the five regressions at the sample mean (with the exceptions being SO2 and NOx
emissions, the latter of which admits no relationship). In the case of ambient SO2
concentrations, the estimated coefficient of the interaction term between income inequality
and polity is positive, implying that increased inequality increases pollution levels, and the
effect is stronger the more democratic the political institutions. The remaining cases are
more complicated, with the sign of the effect determined by the level of per capita income.
Inequality exacerbates CO2 emissions if GDP per capita < $4,210, VOC emissions if GDP
per capita > $10,280, and SO2 emissions if GDP per capita > $13,495.
Interestingly, a distinction can be made here between greenhouse gasses (such as
CO2) and the ozone and acid-rain generating chemicals (such as SO2, VOC, and NOx). The
latter pollutants most often exhibit an EKC relationship because their consequent damages
are primarily local in nature, as opposed to carbon compounds which are global in their
environmental impacts (Stern and Common, 2001; Shafik, 1994; Ansuategi and Escapa,
2002). One explanation for this intriguing result may be the relationship between income
inequality and differences in environmental preferences of the poor and rich, as mentioned
in the discussion of the effect of income inequality on environmental quality in Section 3.
That is, the poor are the primary victims of many local air pollution because they can
neither afford the high locational rents associated with environmental amenities nor to
choose environmental quality over having a job that is overly exposed to pollution. They
often have to live and work immediately downstream and downwind, thus bearing a
disproportionate burden of local pollution. As such, the environmental preferences of the
poor are biased toward reduced local pollution. In contrast, the rich, who can afford, and
gain from, the rents of environmental amenities, have a lot of interest in amenity values
20
associated with protection of rain forest biodiversity, endangered species, and the like, and
little interest in some kind of local pollution. Thus, assuming that the political behavior of
each group (whether rich or poor) is self-interested, environmental quality outcomes
depend largely on which group’s environmental interests get served by the State, which in
turn depend on politicians’ sensitivity to the issue of environmental justice (both within and
between generations) and on the effectiveness of each group to influence them.
Accordingly, in the case of CO2, it may be that at relatively low per capita income levels,
as income inequality rises it is the environmental preferences of the poor (i.e., abatement of
local pollution), and not those of the rich (reduction of global pollution and conservation of
environmental amenities), that dominates the state’s environmental policy.
A similar pattern manifests itself in terms of the age distribution of the society, as
measured by the percentage of the population less than fifteen years of age. As seen in
Table 3, NOx and VOC emissions are negatively correlated with the proxy for youth,
independent of the level of national income. A similar result holds for SO2 emissions, but
the relationship reverses with high GDP per capita levels (achieved by only 18% of the
observations in the sample). This may partly reflect the empirical fact that infants and
young children are the main victims of local ground-level Ozone producing pollutants, and
the greater sensitivity of politicians to health hazards of these pollutants among the very
young. Again, however, the greenhouse gas CO2 emissions are predicted to increase with
the share of youngsters in the population, and this marginal effect is intensified with
increases in GDP per capita.
It thus appears that the nature of the pollutant may affect the policy weights given
to preferences and thus the rate at which the preferred environmental policy is translated
into actual policy. One can consider that environmental preferences differ significantly
among various citizen groups (particularly between poor and rich) and that the State may
serve the preferences of one group better than those of another depending on, among other
things, a group’s political influence, the incentives of members to free ride on each other’s
efforts to voice their group preferences, the existence of accountable public agencies and
ease of access to them, the deleteriousness of the effects of a particular pollutant, and the
length of time it takes before the pollution damages become noticeable. On these accounts,
it may very well be the case that at low income levels more weights are given to abatement
21
policies aiming at local pollutants (such as NOx, VOC, and SO2), whose main victims are
the inner-city low income groups and their damages become visible in a relatively short
period, than to policies aiming to abate global or regional air pollutants (such as CO2 ), or
more generally to those policies aiming to improve environmental amenities supporting the
lifestyle of, and benefiting, the rich. And this is more likely to be the case the more
democratic is the political regime of a society.
The last preference shifter under consideration is the education of the populace, as
proxied by the illiteracy rate for adults greater than fourteen years of age. As seen in Table
3, at least one education term is significant in all of the emissions regressions, though no
significant correlation could be determined for ambient SO2. The relationships all follow
the same pattern, with illiteracy positively correlated with emissions at relatively low levels
of income, but the marginal effect reversing sign at higher levels. First, note that for
income per capita levels generally less than the mean of the sample, illiteracy enters the
model with the expected sign (a negative correlation between illiteracy and environmental
policy). While in all four cases the marginal effect changes sign at relatively higher income
levels, this result has little economic meaning. Specifically, it is hard to imagine a scenario
of high per capita income and high illiteracy rates, nor there exist data points in the sample
consistent with this pattern. In addition, it may be that correlations between illiteracy and
the lack of industrialization are causing this counter-intuitive result. In any case, we
conclude that in the relevant range of per capita income levels, more education, as proxied
by a decline in the illiteracy rate, results in a greater demand for environmental quality, but
does so at a decreasing rate. Future research with richer data sets may shed more light on
this issue.
6. Conclusions
This paper has investigated the link between income per capita and environmental
quality. Recognizing that the often-cited “inverted U-shaped” relationship or EKC is not an
inevitable result of income growth, a model was developed that specifically accounted for
different environmental policy regimes, reflecting the demand for environmental quality as
a public good. The political regime was identified as a function of governance and
22
preference variables, with preferences for environmental policy exercised through
interactions with the political system.
Results of the exercise support the hypothesis that the qualities of political
institutions and several indicators of societal preference interact with each other to create
the inverted-U shape, which is frequently cited in the environment-development literature.
Estimates of individual effects for each of the included preference shifters support the
hypothesis that more democratic governments respond favorably to environmental
demands by the populace. Further research, especially with regard to mechanisms through
which preferred policy is related to actual policy (e.g., along the models developed by
Persson and Tabellini (2000)), is needed to determine the extent of these linkages. In this
regard, of particular interest is the structure of the relationship between the State and
alternative interest groups. Furthermore, the assumption of an exogenous and independent
political system, income distribution, and educational structure themselves may need to be
relaxed, particularly if national income is taken to directly affect these variables. Finally,
direct estimation of a structural system that accounts for feedbacks between the economy,
environment, and institutions in a given country, while fairly complex, could provide
valuable insights for formulation of environmental policy.
23
Table 1: Summary Statistics of Primary Explanatory Variables
Variable
Units
No. of Obs.
Mean
Std. Dev.
Min
Max
586
2,691.71
3,570.42
29.05
19,336.53
204
39.67
20.50
6.04
97.66
VOCa
emissions, kg per capita
emissions, '000 metric tons per
million persons
non-methane emissions, '000
metric tons per million persons
138
52.85
52.50
0.64
251.90
SO2a
emissions, '000 metric tons per
million persons
206
58.48
51.11
4.16
309.27
275
275
275
15.55
7.47
10.98
12.47
5.36
11.51
0.30
0.44
0.18
78.00
23.63
43.01
275
275
275
275
275
275
15.79
61.71
36.10
28.69
12.38
1982.55
6.63
23.64
8.16
8.02
16.42
4.37
1.00
14.18
19.90
18.38
0.00
1975
20.00
96.10
59.00
49.86
72.85
1992
CO2a
NOxa
GEMSSO2b
GDPb
Popdenb
Polityb
Urbanb
Ginib
Youthb
Illitb
Yearb
ambient concentration, parts per
million
000 Current PPP$
(persons / sq. km.) / 10
index, 0 (autocratic) - 20
(democratic)
% population
index, 0 (equal) - 100 (unequal)
% population 14 yrs or under
% population 15 yrs and above
a
CO2, NOx, VOC, and SO2 emissions statistics for full sample.
b
Remaining statistics summarized for GEMSSO2 observations.
24
Table 2: Basic EKC Model using Polity as Policy Proxy, Fixed Effects
CO2
GDP
NOx
521.49**
a
GDP2
(51.71)
--
GDP3
--
Popden
Popden2
Popden3
Polity
Polity*GDP
Year
R2
N
Mb
Dependent Variable
VOC
SO2
9.88**
10.68*
25.23*
3.87*
(1.61)
-0.40**
(0.10)
0.01*
(0.00)
--
(10.63)
-1.04*
(0.57)
0.03*
(0.01)
-103.75**
(29.38)
3.68**
(1.01)
-0.04**
(0.01)
3.23*
(1.68)
-0.59*
(0.31)
-3.41**
(1.10)
(3.36)
0.10**
(0.03)
--
0.439
206
30
13.44*
(8.31)
-10.98**
(2.11)
-34.27**
(7.90)
-1.69**
(0.24)
(4.90)
0.20*
(0.32)
-0.01*
(0.01)
-59.24*
(22.85)
1.72*
(0.77)
0.02*
(0.01)
2.27**
(0.85)
-0.54**
(0.14)
-0.24*
(0.60)
0.236
586
92
0.440
204
29
0.320
138
25
48.13**
(17.79)
-0.09**
(0.02)
--
GEMSSO2
---0.19*
(0.10)
--
4.06**
(1.16)
--0.41*
(0.93)
-0.33*
(0.14)
-0.77*
(0.39)
0.145
275
32
* denotes significance at the 5% level, ** at the 1% level. a Standard errors in parentheses. b Number of cross-section units.
25
Table 3: Extended EKC Model, Policy Regime Decomposition
Dependent Variable
VOC
SO2
CO2
NOx
GDP2
372.39**
(66.83)
---
14.33**
(2.26)
-0.09*
(0.04)
--
13.91*
(6.46)
--
GDP3
5.17**
(0.71)
-0.06**
(0.02)
--
Popden
--
--
Popden2
0.14
(0.10)
0.00
(0.00)
-422.27**
(80.53)
6.92
(14.47)
1.40*
(0.56)
3.33**
(1.27)
2.69**
(0.63)
2.35**
(0.58)
7.96
(11.27)
-0.33*
(0.13)
0.94**
(0.30)
-0.25**
(0.09)
-0.41
(0.22)
-14.23
(10.21)
-11.90*
(5.70)
0.96**
(0.28)
-0.01**
(0.00)
5.09**
(1.44)
3.17**
(0.42)
--
-1.19**
(0.42)
0.188
275
32
GDP
Popden3
Polity
Urban
Gini*Polity
Youth*Polity
Urban*Polity
Illit*Polity
Polity*GDP
Gini*Polity*GDP
Youth*Polity*GDP
Urban*Polity*GDP
Illit*Polity*GDP
Year
R2
N
Mb
0.331
585
92
--
0.00*
(0.00)
-0.01**
(0.00)
-2.54**
(0.23)
-0.06**
(0.01)
-3.01**
(0.51)
-121.67**
(31.76)
4.46**
(1.11)
-0.05**
(0.01)
18.20
(13.41)
8.37*
(3.84)
-0.27**
(0.06)
-0.31**
(0.11)
-0.02
(0.22)
0.56**
(0.15)
-1.80**
(0.41)
0.02**
(0.00)
0.02**
(0.00)
0.01*
(0.00)
-0.08**
(0.02)
-6.03**
(1.25)
0.641
204
29
0.660
138
25
0.585
206
30
-0.09**
(0.02)
-0.05*
(0.02)
0.03**
(0.02)
----
---13.87
(7.42)
-0.37
(2.52)
-0.07*
(0.03)
-0.23**
(0.04)
0.38**
(0.13)
0.36**
(0.12)
-0.66**
(0.12)
0.01**
(0.00)
---
GEMSSO2
5.42
(3.32)
0.09**
(0.03)
-3.95**
(1.16)
---0.58
(1.03)
0.79*
(0.34)
0.02*
(0.01)
----0.37**
(0.14)
-----
* denotes significance at the 5% level, ** at the 1% level.. a Standard errors in parentheses. b Number of cross-section units.
26
Table 4: Elasticity of Select Variables
Evaluated at Sample Means
Dependent Variable
VOC
SO2
CO2
NOx
GDP
0.63
0.70
0.35
0.25
0.43
Popden
0.03
1.46
0.00
-6.74
2.79
Polity
-0.08
-0.60
-0.34
-1.49
-2.51
Urban
0.46
3.79
8.52
11.41
3.12
Gini
0.02
0.00
0.13
-0.25
0.90
Youth
1.06
-0.85
-1.67
-0.63
0.00
Illit
0.08
-0.10
-0.25
-0.36
0.00
27
GEMSSO2
Table 5: Elasticities of Dependent Variables with respect to Per Capita GDPa
GEMSSO2b
CO2
Mean
0.434
Lowc
2.541
Highd
-0.320
NOx
VOC
Polity Values (all others at sample mean)
Mean
Mean
Mean
0.630
0.696
0.355
Low
High
Low
High
Low
High
0.567
0.696
0.720
0.644
0.413
0.125
Polity
Low
0.368
Polity
High
0.087
Polity
H
L
H
L
H
L
H
c
0.596
0.880
0.775
0.709
0.413
0.125
0.080
-0.234
d
H
0.538
0.511
0.650
0.562
0.413
0.125
0.676
0.429
Gini
L
H
0.581
0.552
0.785
0.603
0.720
0.720
0.644
0.644
0.285
0.584
-0.017
0.273
0.005
0.787
-0.317
-2.400
Youth
L
H
0.534
0.603
0.488
0.925
0.720
0.720
0.644
0.644
0.413
0.413
0.125
0.125
0.179
0.657
-0.123
0.408
Illit
L
H
0.607
0.513
0.949
0.351
0.859
0.707
0.808
0.629
0.924
0.130
0.698
-0.191
1.237
-0.203
1.052
-0.548
L
Elasticity defined as (∆Xi/Xi)/(∆GDP/GDP), with polity and preference shifters set at sample means, 15th (L),
or 85th (H) sample percentiles as indicated.
b
Mean
0.245
L
Urban
a
Polity
SO2
Final GEMSSO2 elasticities invariant to preference shifters.
c
"Low" value defined as 15th percentile value for pollutant-specific sample.
d
"High" value defined as 85th percentile value for pollutant-specific sample.
For example, countries in the L-L group for the GEMSSO2 sample (Polity ≤ 13, GDP ≤ 2.517) include
Chile, China, Egypt, Indonesia, Kenya, Pakistan, Philippines and Thailand. The H-H group (Polity = 20
and GDP ≥ 10.207) includes Australia, Belgium, Canada, Finland, Japan, Netherlands, New Zealand, Portugal,
Spain, Sweden, Switzerland, and the United States. Groups vary by pollutant.
28
Figure 1: Elasticities of Dependent Variables with respect to Per Capita GDP
Elasticity of CO2 w.r.t. GDP
1.000
0.800
L-L
0.600
L-H
H-L
0.400
H-H
0.200
0.000
Polity-Urban
Polity-Gini
Polity-Youth
Polity-Illit
Elasticity of NOx w.r.t. GDP
1.000
0.800
L-L
0.600
L-H
H-L
0.400
H-H
0.200
0.000
Polity-Urban
Polity-Gini
Polity-Youth
Polity-Illit
Elasticity of VOC w.r.t. GDP
1.000
0.800
L-L
0.600
L-H
0.400
H-L
0.200
H-H
0.000
-0.200
-0.400
Polity-Urban
Polity-Gini
Polity-Youth
Polity-Illit
Elasticity of SO2 w.r.t. GDP
2.000
1.000
L-L
0.000
-1.000
L-H
Polity-Urban
Polity-Gini
Polity-Youth
-2.000
-3.000
29
Polity-Illit
H-L
H-H
References
Anderson, D., Cavendish, W., 2001. Dynamic simulation and environmental policy
analysis: beyond comparative statistics and the environmental Kuznets curve.
Oxford Economic Papers 53 (4), 721--746.
Ansuategi, A., Escapa, M., 2002. Economic growth and greenhouse gas emissions.
Ecological Economics 40, 23--37.
Bhattarai, M., Hammig, M., 2001. Institutions and the environmental Kuznets curve for
deforestation: a cross country analysis for Latin America, Africa and Asia. World
Development 29 (6), 995--1010.
Bimonte, S., 2002. Information access, income distribution, and the environmental Kuznets
curve. Ecological Economics 41, 145--156.
Coondoo, D., Dinda, S., 2002. Causality between income and emission: a country groupspecific econometric analysis. Ecological Economics 40, 351--367.
Dasgupta, P., Mäler, K-G., 1995. Poverty, institutions, and the environmental resourcebase, in: Behrman, J., Srinivaan, T.N. (Eds.), Handbook of Development
Economics, Vol. 3A. Elsevier Science, Amsterdam, Chapter 39.
Deininger, K. and Squire, L., 1997. Measuring Income Inequality: a New Database.
Retrieved
2002,
from
the
World
Bank,
http://www.worldbank.org/research/growth/absineq.htm.
Ecological Economics (1998), Special Issue on the Environmental Kuznets Curve, 25 (2),
143-229.
Greene, W.H., 2000. Econometric Analysis, Fourth Edition. Prentice-Hall, New Jersey.
Grossman, G.M., Krueger, A.B., 1995. Economic growth and the environment. Quarterly
Journal of Economics 110 (2), 353--377.
Heerink, N., Mulatu, A., Bulte, E., 2001. Income inequality and the environment:
aggregation bias in environmental Kuznets curves. Ecological Economics 38 (3),
359--367.
Lopez, R., Mitra, S., 2000. Corruption, pollution, and the Kuznets environment curve.
Journal of Environmental Economics and Management 40 (2), 137--150.
Magnani, E., 2000. The environmental Kuznets curve, environmental protection policy and
income distribution. Ecological Economics 32 (3), 431-- 443.
30
Marland, G., Boden, T.A., Andres, R. J., 2003. Global, regional, and national fossil fuel
CO2 emissions. In Trends: A compendium of data on global change. Carbon
Dioxide Information Analysis Center, Oak Ridge National Laboratory, U.S.
Department of Energy, Oak Ridge, Tenn., U.S.A. http://cdiac.ornl.gov/trends/emis/
meth_reg.htm
Miller, N., 2002. Environmental Politics: Interest Groups, the Media and the Making of
Policy, Lewis Publishers, Boca Raton, Florida.
Mueller, D.C., 1997. ed. Perspectives on Public Choice: A Handbook. Cambridge
University Press, Cambridge, U.K.
Mueller, D. C., Stratmann, T., 2002. The economic effects of democratic participation.
CESifo Working Paper Series No. 656., January, http://ssrn.com/abstract=301260.
Ono, T., Maeda, Y., 2001. Is aging harmful to the environment? Environmental and
Resource Economics 20 (2), 113--127.
Ordeshook, P. C., 1986. Game Theory and Political Theory: An Introduction. Cambridge
University Press, Cambridge, U.K.
Panayotou, T., 1997. Demystifying the environmental Kuznets curve: turning a black box
into a policy tool. Environment and Development Economics 2, 465--484.
Persson , T., Tabellini, G., 2000. Political Economics: Explaining Economic Policy. The
MIT Press, Cambridge, Massachusetts.
Polity IV database, 2003. Compiled by The Integrated Network for Societal Conflict
Research Program, Center for International Development and Conflict
Management, University of Maryland. Available at http://www.cidcm.umd.edu/
inscr/polity/.
Rivera-Batiz, F.L., 2002. Democracy, governance, and economic growth: theory and
evidence. Review of Development Economics 6 (2), 225--247.
Roca, J., Padilla, E., Farré, M., Galletto, V., 2001. Economic growth and atmospheric
pollution in Spain: discussing the environmental Kuznets curve hypothesis.
Ecological Economics 39, 85--99.
Scruggs, L.A., 1998. Political and economic inequality and the environment. Ecological
Economics 26, 259--275.
Shafik, N., 1994. Economic development and environmental quality: an econometric
analysis. Oxford Economic Papers 46 (0), 757--773.
31
Stern, D.I., Common, M.S., 2001. Is there an environmental Kuznets curve for sulfur?
Journal of Environmental Economics and Management 41 (2), 162--178.
Torras, M., Boyce, J.K., 1998. Income, inequality, and pollution: a reassessment of the
environmental Kuznets curve. Ecological Economics 25 (2), 147--160.
United
Nations Statistics Division, 2004. 2000 Energy
http://unstats.un.org/unsd/energy/yearbook/default.htm.
Yearbook
United
Nations Statistics Division, 2003. Common Database.
http://unstats.un.org/unsd/cdb/cdb_help/cdb_quick_start.asp.
(webpage).
Available
at
United States Environmental Protection Agency (EPA), 2002. GEMS/AIRS Executive
database. Available at www.epa.gov/docs /airs/aexec.html.
United States Environmental Protection Agency (EPA), 2003. The Green Book. Available
at www.epa.gov/oar/oaqps/greenbk/.
Vincent, J.R., 1997. Testing for environmental Kuznets curves within a developing
economy. Environment and Development Economics 2 (4), 417--431.
World
Bank,
2003.
World
Development
Indicators.
http://www.worldbank.org/data/wdi2000/index.htm.
Available
at
World
Resources
Institute,
http://earthtrends.wri.org.
Available
at
2003.
Various
32
databases.