Vous êtes sur la page 1sur 14

Ecological Economics 106 (2014) 6982

Contents lists available at ScienceDirect

Ecological Economics
journal homepage: www.elsevier.com/locate/ecolecon

Analysis

Environmental justice and air pollution: A case study on Italian provinces


Anna Rita Germani a,, Piergiuseppe Morone a,1, Giuseppina Testa b
a
b

DiGEF, Department of Law, Philosophy and Economic Studies, University of Rome La Sapienza, p.le Aldo Moro, 5, 00185 Rome, Italy
Department of Economics, University of Foggia, Largo Papa Giovanni Paolo II, 1, 70100 Foggia, Italy

a r t i c l e

i n f o

Article history:
Received 8 May 2012
Received in revised form 2 July 2014
Accepted 14 July 2014
Available online 4 August 2014
JEL classications:
Q56
K32
Keywords:
Environmental justice
Social inequalities
Environmental Kuznets curve
Air pollution emissions

a b s t r a c t
This paper provides an empirical analysis on the relationship between income, demographic characteristics and
concentrations of air industrial pollutants within the Italian provinces. Two general conclusions can be drawn
from the empirical results. First, the estimates obtained are consistent with an inverse U-shaped environmental
Kuznets curve: air pollution releases increase with income up to a turning point where the relation reverts.
Second, there is evidence that air releases tend to be higher in provinces with high concentration of females as
households' head and with high concentration of children. Since our ndings do not point to environmental
discrimination on the basis of ethnicity, this suggests that environmental justice issues in Italy are not likely to
manifest themselves along racial and ethnic terms but instead in terms of social categories and gender composition. We also nd that the proxy variables employed to measure the efciency or inefciency of law enforcement
are associated with higher levels of pollution. In terms of policy implications, this result suggests the need to
strengthen, all through the country, the local enforcement of environmental laws in order to possibly reduce
the negative effects on ambient air pollution.
2014 Elsevier B.V. All rights reserved.

1. Introduction
The relationship between income and environmental pollution has
been the subject of a large amount of both theoretical and empirical
studies2 since the pioneering works by Grossman and Krueger (1991,
1993, 1995). An assessment of the existing literature suggests that
there are two prevalent strands of literature focusing either on the
nexus between income and environmental pollutants or between
income, demographic composition of population and pollution, where
little effort has been made to join these two strands of literature
together. The aim of this study is an attempt to ll this gap. Basically,

Corresponding author. Tel.: +39 064991 0642; fax: +39 064991 0648.
E-mail addresses: annarita.germani@uniroma1.it (A.R. Germani),
piergiuseppe.morone@uniroma1.it (P. Morone), g.testa@unifg.it (G. Testa).
1
Tel.: +39 064991 0567; fax: +39 064991 0648.
2
In particular, Grossman and Krueger analyze the EKC through the discussion of three
different mechanisms: scale effect, composition effect and technique effect. Scale effect
shows that even if the structure of the economy and technology does not change, an
increase in production will result in an increase of pollution and environmental degradation. Economic growth through scale effect has a negative impact on the environment. On
the other hand, the authors argue that composition effect may have a positive impact on
the environment. Pollution increases in the earlier stages of development, while in the
later stages of development pollution decreases as the economic structure moves towards
services and light manufacturing industries. Therefore, composition effect could lower
environmental degradation through this change in the structure of production. Finally,
technique effect captures improvements in productivity and adaptation of cleaner technologies, which will lead to an increase in environmental quality.

http://dx.doi.org/10.1016/j.ecolecon.2014.07.010
0921-8009/ 2014 Elsevier B.V. All rights reserved.

the literature on environmental pollutants and income aims to test


the validity of the so-called environmental Kuznets curve (EKC)
hypothesis which postulates an inverted U-shaped relationship between the level of environmental degradation and income similar to
that suggested by Kuznets (1955) between income inequality and economic development. That is to say that the level of per capita income
has a negative effect on environmental quality measured by the levels
of pollution, but, beyond a certain level, per capita income has a positive
effect on environmental quality. A crucial issue becomes the estimation
of the turning point in such relationship (Grossman and Krueger, 1995;
List and Gallet, 1999). The second strand of the literature looks at the
causal relationship between income, on the one hand, and demographic
and socio-economic characteristics of population, on the other hand.
This gave rise to the environmental justice movement3 that deals mainly with the question of whether disadvantaged population groups, such
as racial and socio-economic minorities, suffer from living in more
polluted areas and whether demographic composition inuences the
amount of pollution. The key concept of environmental justice issues
is that low-income groups and ethnic minorities bear disproportionate

3
Environmental justice is a movement that emerged in the United States in the 1980s
and has become a concern in the US federal policy agenda in the early 1990s. In 1994, in
fact, environmental justice was institutionalized at federal level through an Executive Order, which focused attention on human health and environmental conditions in lowincome and minority communities (Title VI of the Civil Rights Act of 1964 prohibits discrimination on the basis of race, color, or national origin).

70

A.R. Germani et al. / Ecological Economics 106 (2014) 6982

environmental burdens, in the form of polluted air and water, unsafe


jobs, under-enforcement of environmental laws, etc. (Evans and
Kantrowitz, 2002; Ringquist, 1997).
A broad empirical literature has attempted to assess both hypotheses (i.e., the EKC and the environmental justice) nding, however,
sometimes contrasting results (Anderton et al., 1994; Arora and
Cason, 1999; Been and Gupta, 1997; Brown, 1995). There is, in fact,
no general agreement neither on the existence of an environmental
Kuznets curve, nor on whether minorities or disadvantaged groups
face more pollution, and if so which minorities (racial, age, gender)
are more at risk.
Moreover, most empirical investigations have focused on North
America, this leaving the European case partially under investigated.
Specically, there is only scant evidence in Italy on the impact of
socio-economic factors on environmental outcomes. In fact, empirical analyses on Italian data with focus on social inequalities in exposures to trafc emissions have only been done with regard to the
city of Rome (Forastiere et al., 2007), and on waste generation and
landll diversion (Mazzanti et al., 2009). This study attempts to ll
this gap in the empirical literature by investigating whether income
and the ethnic and social composition of population in Italy may
have a role in explaining air emissions. Moreover, following
Costantini et al. (2013) we implement our analysis taking into consideration three further elements which might affect polluting emissions, namely: technological factors, local environmental regulations
and spatial clustering effects. The analysis is conducted at provincial
level4 to investigate the existence of differences in the determination
of environmental pollution. Air pollution emission data (from 2005)
were combined with data from the 2001 Italian Census.
The results obtained show i) the existence of an inverted U-shaped
relationship between income and pollution, ii) no evidence of environmental inequity against the foreign component of the population but
signicant evidence that releases are higher in provinces with higher
percentage of both children and female-headed households, and
iii) that judicial inefciency is associated with higher levels of air pollution. These results imply that, in Italy, environmental injustices are more
likely to be observed in terms of social conditions that in terms of racial
discrimination.
The remainder of the paper is organized as follows. Section 2
presents a review of the key conceptual issues that are addressed in
the literature. Section 3 presents the theoretical framework and the
empirical model specications. In Section 4 the datasets used in the
analysis are discussed. Section 5 presents the estimations' results, and
Section 6 concludes.

2. Key References in the Literature


Since Grossman and Krueger's (1991) seminal paper, a large volume
of scientic research has been conducted to investigate the hypothesis of
an inverted U-shaped relationship between a measure of environmental
quality (such as, CO2 emissions per capita) and the gross domestic product (GDP) per capita. This hypothesis assumes that in the early steps of a
country's economic growth, pollution tends generally to increase rapidly
because the country's priority is production and minor attention is
devoted to the environmental impact of growth. This implies that
increases in economic production determine, throughout a scale effect, more pollution and environmental degradation (Beckerman,
1992; Bousquet and Favard, 2000; Dinda, 2004). However, in the following phases of growth, as national GDPs keep rising, the need of a cleaner
4
In Italy, a province is an administrative sub-division of a region, which is an administrative sub-division of the State. A province consists of several administrative subdivisions called comune. Italy was divided into 103 provinces at the time we collected
our data; as of 2011, there are 110 provinces. Provinces are equally distributed on the territory between northwest, northeast, center and south, even though the level of urbanization is higher in the northern part of the country.

environment increases as individuals, businesses and institutions are


willing to invest for improving the environmental quality (Bhagawati,
1993; Dinda, 2004; Kijima et al., 2010).
Since the early 1990s the empirical literature on the relationship
between economic growth and environmental pollution has dramatically increased with plenty of empirical studies nding evidence of the
existence of the EKC (Cole and Neumayer, 2005; Coondoo and Dinda,
2002; Copeland and Taylor, 2004; Grossman and Krueger, 1991, 1995;
Koop, 1998; Panayotou, 1993, 2000; Selden and Song, 1994; Shak
and Bandyopadhyay, 1992; Stern, 2004) but also with several studies
nding little evidence of the EKC (e.g., Dinda, 2001; Harbaugh et al.,
2002; Soytas et al., 2007). Given that the relationship between income
and environmental degradation has been observed to vary widely
according to the type of pollutant taken into account (Bruvoll and
Medin, 2003; Cole et al., 1997), several are the factors that have been
put forward to explain the different results, such as the timeframe considered, the countries (i.e., developed or developing), the measurement
method of emissions, and the estimation model.
Grossman and Krueger (1995) argue that knowing the shape of the
relationship between environment and income could help policy
makers in improving or developing new environmental policies. However, as de Bruyn et al. (1998) point out, studies on EKC are based on
reduced-form models. This means that the endogenous variable (environmental quality) is expressed only as a function of predetermined
variables, and no indication about the direction of causality (whether
growth affects the environment or vice versa) is known. As stated by
Cole et al. (1997, p. 401) reduced-form relationships reect correlation
rather than causal mechanism.
Another source of discontent associated with EKC studies refers to
the fact that several other variables might affect pollution levels. As
already mentioned, the so-called environmental justice approach aims
to expand the structural factors assumed to drive the environmental
Kuznets curve relationship, in order to better integrate economic and
social issues with environmental issues. Ethnic diversity and race have
been the most signicant variables which have been neglected in
empirical studies (for example, Cole et al., 1997; Selden and Song,
1994) on the EKC, but that have started to be used by the environmental
justice literature to investigate the possible causal relationship between
income inequalities and pollution levels.5
Over the last two decades, environmental justice literature has
grown very rapidly. Numerous studies provide evidence of inequities
in the spatial distribution of environmental quality (e.g., Aradhyula
et al., 2006; Bullard, 1983; Bullard and Wright, 1987, 1989; Bullard
et al., 2010; Goldman, 1991; Hamilton, 1993, 1995; Nieves and Nieves,
1992). However, some others nd limited or no support for the existence of environmental inequities (e.g., Anderton et al., 1994; Been
and Gupta, 1997 and more recently Cory and Rahman, 2009).
Anderton et al. (1994), using the 1980 US census data and employing
multivariate regression techniques to investigate environmental equity
in the demographics of dumping, nd that education and occupation,
but not race, are signicant indicators of waste facilities in a census
tract. Been and Gupta (1997) using 1990 US census data, obtain mixed
evidence on environmental inequities: while waste disposal sites
proved to be correlated with race and income, neither the percentage
5
On the causal relationship between poverty/inequality and pollution there is a broad
ongoing debate. Poor people, in fact, are assumed to have a lower willingness to pay for
a clean environment, thus showing a higher willingness to live in damaged environmental
and social areas. The underlying question posed by several authors (e.g. Thurow, 1980) is
Can we really expect people who suffer severe problems of poverty to care about urban
air pollution? which is not any easy question to answer. As observed by MartinezAlier and Guah (1997), in fact, also the poor are environmentalists, though it is not always
this very term that is being used to describe their concern for the environment. Their environmentalism is to be distinguished from that of rich people as it is usually associated
with ecological distribution conicts; accordingly, these two varieties of environmentalism have been characterized by the authors as the environmentalism of the afuence or
of the enhanced quality of life, on the one hand, and as the environmentalism of survival
or livelihood, on the other.

A.R. Germani et al. / Ecological Economics 106 (2014) 6982

of poor nor the percentage of African Americans were signicant factors


in deciding the siting of waste sites.
Mohai and Bryant (1992) reviewed fteen various environmental
inequity studies conducted between 1971 and 1992 and concluded
that nearly all the studies showed evidence of inequities, based on
income and race, in the distribution of environmental hazards. They
also suggest that factors such as housing discrimination and the location
of jobs may have led poor and racial minorities to move closer to
hazardous facilities due to the cheapest available housing and potential
job opportunities. In a study on the continental United States, Downey
and Hawkins (2008a) nd that family composition is an important
predictor of neighborhood environmental hazard levels since femaleheaded families and single-father families with young children are overrepresented in neighborhoods with high toxic concentration values.
However, the authors underline that there is still the need to understand why these typologies of families are overrepresented in environmentally hazardous neighborhoods and why certain segments of these
populations are more overrepresented than others; future research
should try to determine why minority, poor, or working-class singleparent families are more likely than other families to live in environmentally hazardous neighborhoods, considering if potential aspects
such as lifecycle factors, public housing and welfare policies could
improve the actual understanding of environmental inequalities.
In another study, Downey and Hawkins (2008b) using spatial analysis and multilevel data, nd that black, white, and Hispanic households
with similar incomes live in neighborhoods of dissimilar environmental
quality, and that the association among neighborhood, income levels
and hazard levels varies according to neighborhood and household
racial composition. Their ndings demonstrate that in the United
States, blacks are more highly represented in census tracts with high
toxic concentration levels than are any other major racial/ethnic group
in the country.
Similarly, Ash and Fetter (2004) combining 1990 USA Census block
group data for urbanized areas with 1998 data on toxicity-adjusted
exposure to air pollution, found that communities with higher concentrations of lower-income people and people of color experience
disproportionate exposure to environmental hazards. Additionally,
ndings highlighted the importance of controlling for interregional
heterogeneity, a fact which we shall take in due consideration
while conducting our empirical investigation.
The environmental justice debate is only beginning to develop at
the European Union level6 (for a comprehensive review on European
environmental justice issues, see Eloi, 2011). However, especially in
the UK, the environmental justice debate has started to expand by integrating environmental issues and social justice perspective. McLeod
et al. (2000) found that higher social classes were more likely to be
exposed to greater air pollution. However, other studies (Brainard
et al., 2002; Naess et al., 2007; Namdeo and Stringer, 2008; Wheeler
and Ben-Shlomo, 2005) found that air quality is poorer among households of low social class. Environmental inequalities were explored
also in Helsinki by Rotko et al. (2000) and Rotko et al. (2001): levels
of NO2 decreased with a higher level of education. In Sweden, two studies showed evidence of social inequalities related to NO2. Stroh et al.
(2005) found that the strength of the association between the socioeconomic status and NO2 concentrations varied considerably between
cities. In another study, Chaix et al. (2006) found that children from
areas with low neighborhood socio-economic status were more exposed
to NO2 both at home and at school.

6
The institutional beginnings of the environmental justice approach can be dated from
the drafting of the UNECE Convention on Access to Information, Public Participation in
Decision-making and Access to Justice in Environmental Matters, adopted at the Fourth
Ministerial Conference in the Environment for Europe process in UNECE (1998). In its Article 1, the Convention states as an objective to guarantee the rights of access to information, public participation in decision-making, and access to justice in environmental
matters in accordance with the provisions of this Convention.

71

Four other European studies explored social inequalities related to


air pollution. In Netherlands, according to Kruize et al. (2007), lower
income groups live in areas with higher levels of NO2 than greater
income groups. In Germany, Schikowski et al. (2008) revealed the existence of a social gradient with higher PM10 exposures among subjects
with less than 10 years of schooling than among those with higher
education. Havard et al. (2009), using a French deprivation index
(Havard et al., 2008), found that in Strasbourg the mid-level deprivation
areas were the most exposed to NO2, PM10 and CO. By contrast, in Rome,
Forastiere et al. (2007) found that the higher social classes appear to
reside in areas with high trafc emissions. This disparity is even stronger when socio-economic status rather than income is considered.
From this review, it is clear that in Europe the empirical literature on
environmental pollution and socio-economic status is a relatively novel
topic compared to the US. Moreover, the few available studies generate
mixed ndings regarding exposure disparities. As already mentioned,
Italy is one of the less investigated countries. In what follows, we aim
to cover this gap by trying to establish the existence (or not) of an
inverted U-shaped relationship between pollution and income, as well
as environmental injustices among Italian provinces.
As mentioned in the Introduction, however, our analysis will be
complemented taking into account also another set of variables that,
along with the EKC approach and the environmental justice approach,
complete the picture of the different driving forces linking economic
features with environmental issues. We refer to the ImpactPopulation
AfuenceTechnology (IPAT) realms (Andreoni and Levinson, 2001;
York et al., 2003). The key drivers scrutinized under this framework,
which we will consider are: institutional quality (Dasgupta et al., 2006),
and technological capabilities (Costantini and Martini, 2010). Finally, we
shall tackle possible transboundary pollution effects considering spatial
correlation problems which might in turn generate polluting clusters of
neighboring provinces.
3. Motivation and Methods
3.1. Motivation
To motivate this empirical analysis, we adopt the theoretical framework developed by Hamilton (1995) in which he puts forward three
alternative explanations to account for the pollution patterns: i) pure
discrimination related to race/gender, ii) the Coase Theorem, and
iii) the theory of collective action (Olson, 1965).
Under the race/gender discrimination hypothesis, facility operators
are assumed to look at the racial composition of communities surrounding polluting facilities and decide to locate or increase releases in areas
with higher concentrations of minority or low-income groups. Hence,
race is perceived to be a factor behind such decisions, that prevails compared to other economic factors (i.e., costs, efciency) that would be of
greater importance to a rational prot-maximizing rm. Mohai and
Bryant (1992) have identied a number a possible relationships
between race/gender composition and facilities' siting decisions, such
as (a) lower costs of doing business (due to the availability of lower
land values and lower incomes in minority communities); (b) lack of
conict in poor areas due to weak political power or insufcient community resources; and (c) limited mobility of minorities due to poverty
and housing discrimination.
The second explanation applies the standard version of the Coase
theorem, suggesting that, in a world without transaction costs, a polluting rm will locate (or increase pollution) in areas in which the releases
will cause the least damage. Looking for the lowest damage can be
translated, from the polluting rm's perspective, in locating in areas
where potential compensation demands (i.e., for adverse health impacts and property loss caused by exposures to pollution) and liability
costs are expected to be lower.
Finally, under the last explanation, rms may decide to increase
releases in minority and poor community areas because they face less

72

A.R. Germani et al. / Ecological Economics 106 (2014) 6982

(political) collective actions. In an ideal Coasean world, the victim


would be able to negotiate compensation directly with the polluter. However, the compensation demands, in reality, are typically negotiated at
community level through the political process. This could lead to results
that appear similar to the pure race/gender discrimination hypothesis:
rms will decide to locate or increase releases in areas where they face
the lowest political opposition to their actions. To the extent that minority communities are less likely to be politically active, then these communities will be more likely to experience higher levels of pollution.
These alternative theories predict that certain variables should
explain pollution levels. The race/gender discrimination hypothesis
tests whether factors such as the race and the gender composition of
the population predict releases. The Coase theorem hypothesis tests
whether economic factors, such as income levels, explain releases.
Factors such as age, education and the number of households with children are expected to inuence the incentives to undertake political
actions (Filer et al., 1993). Along these variables, we shall consider in
our empirical investigation also the role of judicial inefciency, which,
as showed by various authors (see, for instance, Diarra and Marchand,
2011; Germani and Scaramozzino, 2007) should positively impact on
air pollution. We shall include in our regression model also some
other variables which might affect the environmental performance at
provincial level. First, in the vein of IPAT approach (ImpactPollution
AfuenceTechnology) we include measure of the level of technology.
Second, we shall consider regional expenses on environmental issues
capturing differences across regions. Finally, we account for the
existence of spatial correlation problems including, in our empirical
estimates, geographic environmental spillovers from other provinces
(on this point see: Costantini et al., 2013; Maddison, 2006).

3.2. Methods
As broadly discussed, the main objective of this work is to test whether air releases generated by the industrial sector could be explained using
economic (income level) as well as socio-demographic variables. However, a methodological issue that arises in the literature is that regressing
pollution levels on income can introduce endogeneity problems.
In fact, when investigating if pollution amounts may be determined
by income levels, it must be considered that rms may be attracted
to locate in low-income areas (Hamilton, 1993, 1995), but also lowincome groups may choose to live in or nearby polluted areas for economic reasons (e.g. cheaper rents). Hence, there is a problem of reverse causality,7 due to the fact that (a) income might affects pollution levels through
growing levels of production, or (b) the amount of pollution might affect
the level of income due to the fact that low-income groups can decide to
live in cheaper, but more polluted, areas.8 In order to minimize possible
7
Reverse causality is one of the main sources of endogeneity problems. Been (1994) also points out this endogeneity problem and resolves it by using pre-siting demographic
data (i.e., data from before the industrial plants were built). Ringquist (1997) uses a control variable approach by controlling for housing prices; Gray and Shadbegian (2004) use
instrumental variables.
8
Note that a similar endogeneity problem might arise from the use of sociodemographic variables. However, the values of the explanatory variables (both economic
and socio-demographic) are observed at their 2001 Census values: thus, the 2001 economic and socio-demographic characteristics are used to explain air releases in 2005
(see Arora and Cason, 1999, on the use of lagged explanatory variables to avoid
endogeneity bias). Hence, in our estimation model, we assume that the economic and
socio-demographic conditions (predetermined economic, social and demographic provincial data observed at time t) take some time (a four-year time lag) to exhibit their effects
on the levels of air pollution (observed at time t + 1). Causality problems might still occur
in our regressions if pollution is persistent over time. However, as pointed out by an anonymous referee, the fact that we use province-level data makes the issue of locating on the
basis of cheap rent less of a problem. In fact, data support the idea that in Italy internal migration is generally low when compared with other countries, such as the USA (Faini et al.,
1997). In a recent study, Napolitano and Bonasia (2010) showed for the period 19952006
that wage differential was found to be an important determinant of internal migration. On
the other hand, housing price differentials seemed to have little or no power to explain migration from one region to another.

bias due to this endogeneity problem, we propose an instrumental variable model to control for the endogeneity issue.
In the Results section, we present estimates from an ordered probit
model and an ordered probit model with instrumental variables. We
estimate two versions of our model: (1) in the rst version, we consider
as dependent variable a measure of air pollution obtained aggregating
twelve local pollutants (see Section 4.1 for details); in the second version, the dependent variable is the main global pollutant i.e. CO2.
Moreover, we also estimated our model under different econometric
specications, to further examine its robustness. Specically, we estimate a simple OLS model and a two-stage least square (2SLS) with
instrumental variables9 (results referring to the OLS and 2SLS estimates
will be reported in Appendix A). The main advantage of the ordered
probit model over either the OLS model or the 2SLS model in our
study is that the marginal effects allow us to determine the impact of
each explanatory variable (e.g., ethnicity, income and minorities) on
the probability of each level of air pollution emissions. Even though
this ranking approach to measure the amount of pollution has not
been widely used in previous environmental justice studies, there are
some precedents for using an ordered probit analysis.10
The estimated model takes the following form:
 
 

2
yi 0 1 ln Y pc 2 ln Y pc 3 lpcunivdegree 4 lpcasia
5 lpcafr 6 lpcfemhead 7 lchildren 8 lelders 9 jud inef
10 jud ef 11 D 12 emp RD 13 exp RD 14 reg exp e
1
The dependent variable measures air pollution emissions, which are
estimated against a set of independent variables. To measure income
levels, we use the variable ln(Y pc ), the logarithm of the income
values; we also considered the quadratic specication of the same
variable ln2(Ypc) which, as broadly discussed, allows us to capture
the presence (or not) of an inverse U-shaped relationship between
income and pollution. In the instrumental variable model we instrument income and income square using a set of variables which will
be discussed in Section 4.3.
Eq. (1) includes also a variable capturing high level of education
(lpcunivdegree), which measures the share of people with an undergraduate university degree. Regarding ethnic group characteristics, the
percentage of African population (pcafr) and the percentage of Asian population (pcasia) are used. The percentages of children, elders and of families with a female as the head of the household are considered to be
groups that could suffer from possible environmental discrimination.
As discussed in Section 3.1 above, we account in our regression
model also for law enforcement. Specically, we include two variables:
jud_inef accounts for the average length (expressed in years per 10,000
population) of completed proceedings this is to be considered as a
measure of the judicial inefciency; jud_ef considers the per capita
number of overall proceedings (pending and completed) in the courts
located in the Italian provinces this is to be considered as a measure
9
A multitude of different statistical approaches are employed in the environmental justice analysis, depending on the nature of the dependent variable. Multivariate analysis (for
example, Been and Gupta, 1997; Pastor et al., 2001), as well as logit (for example, Brooks
and Sethi, 1997; Cory and Rahman, 2009; Hamilton, 1995, where the dependent variable,
i.e. the level of exposure to pollution levels, assumes the value of 1 if there was an increase
of exposure in the zip code and 0 otherwise) and probit models (Aradhyula et al., 2006;
Ringquist, 1997) are used.
10
Ordinal probit or logit regressions have in fact been used with ordered (from low to
high) dependent variables (Agresti, 1990). Sadd et al. (1999) estimate an ordered logit
model on Los Angeles neighborhoods, building a dependent variable ordered according
to the level of assumed health hazard, which takes a value of 0 if the tract has no air release, a value of 1 if it has an air release that does not contain carcinogen compounds,
and a value of 2 if it contains carcinogen air release. Forastiere et al. (2007), in investigating
the relationship between exposure to trafc emissions and socio-economic conditions in
Rome, grouped air pollution (i.e., particulate matter emissions, PM10) into four categories:
low, mid-low, mid-high, and high emissions, using the 20th, 50th, and 80th percentiles as
cut-off points.

A.R. Germani et al. / Ecological Economics 106 (2014) 6982

73

Table 1
Descriptive statistics of the air pollutants that compose the dependent variable in the ordered probit regression.
Pollutants descriptive statistics (N = 103 provinces)
Pollutants
Local
Sulphuric dioxide*
Nitrogen oxides*
Carbon monoxide**
Ammonia**
Arsenic****
Chromium**
Copper**
Mercury**
Nickel*
Lead*
Selenium*
Benzene*
Global
Carbon dioxide***

Mean
8557.90
8980.24
16,963.46
3847.59
363.38
508.68
405.63
69.95
1332.13
1943.42
109.11
153.64

3,493,031

Median
1990.71
3946.46
9025.16
2224.30
107.35
185.51
126.35
34.18
564.34
695.85
30.93
114.53

1,848,533

Standard deviation
17,423.36
11,918.65
38,034.08
4817.7
894.59
810.91
1524.76
138.26
1978.39
4622.00
215.34
140.91

4,561,720

Minimum value
0.303
14.60
1502.72
14.25
0.56
0.65
3.53
1.58
3.64
0.83
0.36
16.82

Maximum value
96,385.52
54,407.96
376,509.9
28,817.56
8076.338
4481.793
15,151.66
1076.05
13,554.87
42,187.82
1665.048
917.8066

129,135.4

2.85e + 07

Notes: *substance measured in micrograms; **substance measured in milligrams; ***substance measured in megagrams; ****substance measured in nanograms. In the ordered probit
regression analysis, all the different measurement units were converted into megagrams.

of the efciency of law enforcement. Arguably, pollution will be lower in


provinces with efcient courts (able to handle a high number of proceedings) and efcient enforcement, since long trials are likely to postpone the timing of punishment (Becker, 1968) and this could be an
important factor inducing rms to undertake illegal activities.
As mentioned above, we include in our estimation the emissions produced in the neighboring provinces (D) as these may capture the role of
agglomeration phenomena. As observed by Costantini et al. (2013) if
the coefcient has a positive sign, it reveals the existence of agglomerative forces producing a concentration of dirty activities in circumscribed
geographical areas. Otherwise, if internal emission intensity is negatively
inuenced by external one, some centripetal forces are predominant, as
industries located in a dirty area try to become cleaner in order to capture some comparative advantages, for example related to become rst
comers in the green industry (2013: 103). Further, we consider a
technological measure including public and private expenditure in R&D
measured at provincial level (exp_RD) as well as a provincial index of
researchers, technicians and other personnel engaged in R&D activities
both in the public and private sector (emp_RD). Finally, we included
a measure of regional expenses on environmental issues (reg_exp).

4. Data
4.1. Dependent Variables
To build the dependent variable used in the ordered probit
regression, we dene a province-level index of air pollution. Following
Brooks and Sethi (1997),11 we use threshold limit values (TLVs) in
order to adjust for toxicity. As put by the authors a threshold limit
value is the amount of airborne concentration in mg/m3 of a substance
to which a worker may be repeatedly exposed for a normal 8-hour
workday and a 40-hour workweek without adverse health effect
(Brooks and Sethi, 1997: 236).
We employ the Italian threshold limit values established by law for
ambient pollution using the GESTIS-Substance Database which contains
information for the safe handling of hazardous substances and other
chemical substances at work.12
11

Brooks and Sethi (1997) created a weighted toxicity index using threshold limit values
(TLVs), combined with a distance function, to develop an exposure measure for each U.S.
zip code.
12
For an in-depth discussion on threshold limit values and how these can be used to
built the province-level index of air pollution, see Germani et al. (2012).

For the purpose of aggregating pollutants, our air pollution index was
constructed in the following way. Let Eij dene the emission of pollutant j
from province i, and let Tj denote the threshold limit value associated
with substance j. Then, the toxicity-weighted aggregated level of air pollution in province i is dened as: AP i

Eij

. This procedure provides us


Tj
with a measure of emission for each province, which represents our dependent variable in the ordered probit model. So, for every province in
our data set, the sum of all hazardous substances considered, weighted
by their associated threshold limit values, was calculated.
We compute four different specications of the dependent variable
aggregating twelve local pollutants (see data section below and the descriptive statistics provided in Table 1), that is: (i) E: raw data on emission levels; (ii) NE: normalized raw data on emission levels; (iii) IWE:
data on emission levels divided by the Italian threshold limit values;
(iv) NIWE: normalized data on emission levels divided by the Italian
threshold limit values. The four specications yield very similar results
both in terms of their statistical signicance and of the signs of coefcients. For the sake of clarity, we shall report only ndings associated
to IWE specication of the dependent variable.13
As mentioned earlier, we shall consider a version of our regression
model where the dependent variable is the main global pollutant as
CO2. The analysis of carbon dioxide emissions takes particular relevance
for three orders of reasons. First, this pollutant represents (together
with CH4, N2O and CFCs whose commercialization for internal use is
forbidden in Italy) the main gas responsible for the global warming
among the entire greenhouse gases (UNEP, 1999). As reported by Roca
et al. (2001), in fact, the ow of this gas represents one of the key indicators (if not the key indicator) adopted by Eurostat to monitor the
European climate change. Second, having Italy ratied the Kyoto Protocol
(1997) which establishes to limit the emissions of the main greenhouse
gases carbon dioxide emissions are expected to have decreased in the
recent years. Finally, this efuent is used in the majority of studies on
the EKC hypothesis and, therefore, makes the results from the present
study comparable with other investigations.
We use the information on air pollution provided by the Italian
Institute for Environmental Protection and Research (ISPRA)14 which is
j

13
Findings relative to the other three specications are available from the authors upon
request.
14
ISPRA is the Institute for Environmental Protection and Research established by Italian
Law 133/2008. The Institute performs the functions of three former institutions: APAT
(Agency for Environmental Protection and Technical Services), ICRAM (Central Institute
for Applied Marine Research), and INFS (National Institute for Wildlife).

74

A.R. Germani et al. / Ecological Economics 106 (2014) 6982

Fig. 1. Air pollution emissions for the rst twenty most polluted Italian provinces (in million megagrams).

responsible for the National Emission Inventory. The ISPRA dataset includes data on air emissions in all the Italian provinces (103 provinces
distributed over 20 regions).15 This is a comprehensive database that collects all emission estimates of the major pollutants including greenhouse
gases, ozone precursors, benzene, particulate matters, heavy metal and
polycyclic aromatic hydrocarbon.16 The national inventory is reported
to the European Commission at the national aggregated level, but it is calculated at the regional level and then disaggregated at the provincial
level. In the Disaggregation of the National Inventory 2005 Report, data
related to the disaggregation of the emissions of the national inventory
at the provincial level are available, divided by activity according to the
SNAP (Selected Nomenclature for Air Pollution) classication.17 The
SNAP classication consists of 11 macro-sectors. We use data relative to
macro-sector 1 (combustion in energy and transformation industry),
macro-sector 3 (combustion in manufacturing industry) and macrosector 4 (production processes), since we want to base our analysis on
air pollution emissions released by the industrial sector and not also
from agriculture and road, air, or sea transportation. In other words, we
restricted our analysis on industrial emissions since our objective is to
test whether industrial plants siting might affect the demographic composition of population and the socio-economic characteristics of the different provinces. Air pollution emissions are expressed in megagrams:
the average level of air releases is 3,532,766 Mg with a minimum
value of 131,997.8 Mg (Prato) and a maximum value of 2.86e
+ 07 Mg (Rome). The average per-capita air emission levels, instead,
is 7.14 Mg with a minimum value of 0.57 megagrams (Prato) and a
15
We are aware of the fact that the use of too broad a scale or unit of analysis has been
discouraged (Anderton et al., 1994), but the most disaggregated available Italian data are
only at provincial level.
16
One of the main limitations of the ISPRA dataset is that it is not possible to obtain information on the compliance trends and on the enforcement activities in each province. It
would be desirable to have access to more detailed data sets on the number of inspections
conducted by enforcement authorities, on compliance levels and on the implications of the
different penalty means. The lack of accurate and incomplete information does not allow
policymakers to understand how the Italian system of enforcing environmental laws work
and what reforms may be needed.
17
This classication includes all activities which are considered relevant for atmospheric
emissions. The ISPRA database is characterized by three different typologies of emissions:
area, point and linear sources. For area emissions (emissions from sources distributed on
the territory) a direct measurement is not feasible, and it is necessary, therefore, to estimate them from statistical data and specic emission factors. The approach that ISPRA
has applied is based on a linear relation between source activity and emission, following
this relation: Ei = A FEi, where: Ei = emission of the pollutant i (g year1); A = activity
indicator (i.e. produced amount, fuel consumption, etc.); FEi = emission factor for the pollutant i (i.e., g t1 of product, kg/kg of solvent, g inhabitant1).

Fig. 2. Per-capita air emissions levels for the rst twenty most polluted provinces
(per capita megagrams).

maximum value of 41.04 Mg (Taranto). Figs. 1 and 2 illustrate,


respectively, air emission levels and per-capita emission levels for the
rst twenty most polluted provinces. Note that the dependent variable
is specied as emission per unit area; in fact, since provinces vary
in size, this ratio provides a better measure of emissions intensity as
compared to per capita emissions.
4.2. Independent Variables
The primary source for the demographic data used in this analysis
is the Italian Census 2005 Population and Housing (Italian Statistical
Agency ISTAT). The selected variables and their summary statistics
are provided in Table 2.
The independent variables were chosen according to the theoretical
framework described in Section 3 and are among the most commonly
used in EKC and environmental justice studies. Additional variables,
such as the entrepreneurial spirit, the infrastructural endowment and
the efciency of the judicial system constitute an improvement upon
previous studies. Other demographic variables like children (percentage
of population less than six years old) and elders (percentage of population more than 65 years old) are also examined as they are assumed to
be inherently more susceptible to air pollution (Chaix et al., 2006;
Greenberg, 1993). Moreover, the percentage of family households with
a female as the head of the household is considered to be a group that
could suffer from possible environmental discrimination (Arora and
Cason, 1999). Regarding racial characteristics, the percentage of African
residents and the percentage of Asian residents are used.
4.2.1 . High Education
Educational levels are considered as a relevant variable for polluters
to select where to place their polluting plants and for sorting based on
existing pollution levels. Hence, we include this variable among our
explanatory variable of air pollution.18
4.2.2 . Income
The Tagliacarne Institute and the Union of Italian Chambers of
Commerce provide the data related to real household disposable income
per capita in each province.
4.2.3 . Spatial Variable
In the ordered probit regression a spatial regressor is also included.
Following Costantini et al. (2013) we employ a binary contiguity concept,
18

We are grateful to an anonymous referee for pointing this out.

A.R. Germani et al. / Ecological Economics 106 (2014) 6982

75

Table 2
Instrumental and independent variables descriptive statistics (N = 103 provinces).
Variable
Instrumental variables
infrastructure
females
age15to34
age35to49
pcunivdegree
Independent variables
pcincome
pcasia
pcafr
pcfemhead
children
elders
judicial inefciency
judicial efciency
D
emp_RD
exp_RD
reg_exp

Mean

Standard deviation

Minimum value

Maximum value

556.902
247,242
90,399.67
93,498.53
0.105

81.927
271,537.8
101,921.6
106,050.4
0.022

443
39,886
12,641
14,172
0.067

1000
1,696,080
678,961
714,986
0.178

14,675.68
0.0027
0.0064
0.117
5.27
19.84
2.962
0.086
7010.65
2.58
0.52
99.62

3024.42
0.0031
0.0048
0.023
0.75
3.09
1.024
0.059
6982.57
1.17
0.27
121.51

9104.12
0.00015
0.0007
0.077
3.69
12.52
1.336
0.005
494.60
0 .7
0.2
21.7

20,613.52
0.020
0.021
0.201
7.35
25.91
5.968
0.356
37,089.05
5.4
1.5
749.4

Note: The variables used in logs in the regression are presented in their original levels.

assuming that emissions of nearby provinces (dened as direct neighbors


sharing a common border) might affect the level of pollution of a province. We consider the rst-order contiguity with direct neighbors, giving
weight wrs = 1 to each s-th province neighboring province r and wrs = 0
to all other provinces. Consequently, the spatial regressor is dened as the
sum of emissions available in directly neighboring provinces expressed
as: D

Es wrs , where Es is the emission level per unit area (as in

s1; sr

the case for the dependent variable) of the s-th province, N is the total
number of provinces and wrs = 1 if s is neighboring r. Note that when dening our spatial regressor, we do not consider neighboring emissions
of local pollutants, but rather we concentrate on the main global pollutant (i.e. CO2) as this is most likely to generate geographical spillover effects. In fact, as observed by Maddison (2006), it is when
considering transboundary pollution effects that the existence of
spatial correlation problems should be recognized and tackled.19
4.2.4 . Technological Variables
As discussed above, we also consider intramural R&D expenditures as
well as a provincial index of researchers, technicians and other personnel
engaged in R&D activities. Both variables refer to public and private sector.
4.2.5 . Regulatory Factors
In order to control for regulatory factors we include in the ordered
probit estimation a measure of environmental expenditure made by
regional authorities which capture the regional effort to preserve a clean
environment. As observed by Costantini et al. regional environmental expenditures capture a specic feature of environmental regulations, and
are associated with a citizen's willingness to pay for abating emissions
(2013: 108). Specically, we consider the total regional expenditure per
inhabitant.20
These demographic and economic data are implemented with two
variables capturing judicial efciency and judicial inefciency. These
two variables are: (i) the number of proceedings (completed as well as
pending) which is a measure of the efciency of law enforcement and is

19
Given the geographical dimension of our case study, we consider emissions coming
from other provinces, while no transboundary pollution at international level is accounted
for (see also Costantini et al., 2013).
20
Note that data cover all 20 Italian regions and refer to the year 2004 since earlier data
were not available.

dened in terms of per capita number of trials in each province;21


(ii) the average length of proceedings, which is a measure of judicial inefciency, and is dened as the per capita length of trials in each province.22
By merging the above described environmental, demographic and
economic data we produced a database that, as we believe, can contribute to the extremely exiguous literature on EKC and environmental
justice studies in Italy (see Tables A1, A2 and A3 for a summary of
these variables).

4.3. Instrumental Variables


As mentioned above, we instrumented income and income square
using a set of variables which include age, gender and geographical
dummies. More specically, age (grouped into ranges of 5 years each) is
categorized into two groups of age, namely (i) age range from 15 to
34 years and (ii) age range from 35 to 49 years. Another independent
variable employed is female (percentage of females in the population)
to examine female-based variations in the distribution of households
income. We introduced two geographical dummy variables to reect
the territorial sub-division of Italy (Bagnasco, 1977): North Italy (comprehensive of North-eastern and North-western Italian provinces) and
Central Italy. The dummy South Italy is left out as reference. Finally,
we considered the level of infrastructure present in each province, measured as an indicator of the transportation infrastructure endowment
(Guiso et al., 2004).23 These instruments are reasonably credible: while
they are correlated with income, they do not have an effect on
environmental quality, except through their effect on income itself.

21
The data on the number of proceedings by the Il Sole 24 Ore is an elaboration of the data released by the Italian Ministry of Justice.
22
Note that trial and appeal delays are one of the major problems associated with the inefciency of justice in Italy.
23
Data were drawn from the yearly report of data and social indicators on quality of life
performed by the leading Italian nancial newspaper, Il Sole 24 Ore that publishes this annual report on quality of life every year since 1989. The 103 Italian provinces are ranked
according to a summary indicator of their quality of life constructed collecting ofcial statistical data. The nal quality of life indicator is based on 36 social indicators related to six
main areas: consumption and wealth, labor and business, environment and services, justice efciency and criminality, population, leisure. Even though the statistical robustness
of these rankings is often criticized (Lun et al., 2006; Vitali and Merlini, 1999), the results
of the Il Sole 24 Ore report it constitutes a very regular collection and analysis of data on
quality of life.

76

A.R. Germani et al. / Ecological Economics 106 (2014) 6982

Table 3
Ordered probit estimation, IV ordered probit estimation and marginal effects (dependent variable: air pollution emissions data weighted by the Italian threshold limit values).
Dependent variable:
IWE_area

Ordered probit estimations

IV ordered probit estimations

Marginal effects (IV ordered probit)

Coefcients (P N |z|)

Coefcients (P N |z|)

Y=1

Y=2

Y=3

Y=4

ln(Ypc)

131.183a
(0.001)
6.851a
(0.001)
1.567b
(0.032)
0.055
(0.773)
0.193
(0.520)
6.116a
(0.000)
7.234a
(0.000)
1.184b
(0.044)
11.355a
(0.000)
.3754
(0.378)
0.493a
(0.000)
0.705a
(0.004)
1.306b
(0.041)
0.000
(0.786)
669.989
670.943
671.942
103
0.21

129.996a
(0.000)
6.745a
(0.000)
1.648
(0.051)
0.084
(0.662)
0.120
(0.628)
6.061a
(0.002)
7.862a
(0.001)
1.259
(0.059)
11.152a
(0.000)
0.361
(0.255)
0.494a
(0.002)
0.706b
(0.016)
1.306
(0.111)
0.000
(0.884)
670.072
671.025
672.029
103

26.131a
(0.000)
1.356a
(0.000)
0.331b
(0.042)
0.016
(0.663)
0.024
(0.627)
1.218a
(0.001)
1.580a
(0.000)
0.253
(0.064)
2.241a
(0.000)
0.072
(0.250)
0.099a
(0.001)
0.141b
(0.014)
0.262
(0.110)
0.000
(0.884)

10.534a
(0.000)
0.546a
(0.000)
0.133
(0.076)
0.006
(0.661)
0.009
(0.632)
0.491b
(0.012)
0.637a
(0.004)
0.102
(0.057)
0.903a
(0.001)
0.029
(0.282)
0.040a
(0.009)
0.057b
(0.038)
0.105
(0.133)
0.000
(0.885)

6.234b
(0.027)
0.323b
(0.027)
0.079
(0.084)
0.004
(0.667)
0.005
(0.637)
0.290b
(0.039)
0.377b
(0.031)
0.060
(0.144)
0.534
(0.054)
0.017
(0.311)
0.023b
(0.047)
0.033
(0.093)
0.062
(0.184)
7.310
(0.884)

30.431a
(0.000)
1.579a
(0.000)
0.385
(0.053)
0.019
(0.662)
0.028
(0.627)
1.418a
(0.002)
1.840a
(0.001)
0.294
(0.053)
2.610a
(0.000)
0.084
(0.254)
0.115a
(0.001)
0.165b
(0.015)
0.305
(0.110)
0.000
(0.884)

ln2(Ypc)
lpcunidegree
lpcasia
lpcafr
lpcfemhead
lchildren
lelders
jud_inef
jud_eff
D
empl_RD
exp_RD
reg_exp
cut1
cut2
cut3
Number of observations
pseudo R2
Wald chi2
a
b

1.83

Signicant at the 1% level.


Signicant at the 5% level.

5. Results
Tables 3 and 4 report the results of the ordered probit and IV ordered
probit models. For the sake of clarity, we rst discuss results associated
with IWE as the dependent variable (see Table 3); subsequently, we
provide account of results obtained when considering as our dependent
variable CO2 emissions (see Table 4).
First, it should be noted that when the dependent variable is ordered,
estimated parameters do not reect a unit change of an independent
variable on probability; thus, the estimated coefcients in an ordered
probit have no direct interpretation. For this reason, we also calculate
the associated marginal effects (for a discussion of calculating marginal
effects see Greene, 2003: 738). These can be interpreted as the change
in the probability of attaining different levels of air pollution emissions
as a result of a unit change in each explanatory variable. Notice that
the sum of the marginal effects equals zero. The signs on the marginal
effects of the signicant variables do not remain constant: more specifically, in the rst and the second air pollution categories Pr(Y = 1:
low emissions) and Pr(Y = 2: medium-low emissions) the statistically
signicant variables have opposite signs compared to the third and the
fourth air pollution categories Pr(Y = 3: medium-high emissions) and
Pr(Y = 4: high emissions).
The income variables (in their logarithmic linear and quadratic
specications) are both statistically signicant (both in the ordered probit
and IV ordered probit specication), showing that an increase in income
translates to an increase in the probability of attaining high levels of air
pollution emissions (see the last column of Table 3). Moreover, the

statistical signicance of the squared term for income and its negative relationship with the dependent variable, allow us to identify an inverse Ushaped relationship between income and air releases with an estimated
turning point occurring just above 14,000.00 per capita. Following
Millimet et al. (2003), we use the delta method24 to derive standard
error and 95% condence interval around the turning point estimate25
and obtained a range roughly between 13,000 and 16,000.26 This interval falls well within the range of the income data (which as showed in
Table 2 varies between 9104 and 20,614), a fact that strengths the evidence in favor of the existence of the EKC. Further, we plotted a graph
depicting the Environmental Kuznets Curve, where we allow only income
to vary; the graph (see Fig. 3) describes the quadratic relationship linking
pollutions levels and income along with the 95% condence intervals, and
clearly shows the existence of a turning point in the relationship.
The traditional interpretation of this environmental Kuznets (1955)
curve is that an increase in economic activity leads to a higher probability of attaining high levels of air pollution, but beyond the turning point,

24
This method consists in linearly approximating the nonlinear function of the turning
point in the regression coefcients.
25
A symmetric condence interval is formed based on the assumption that per capita income is normally distributed.
26
As a robustness check we also employed a parametric bootstrap method through
Monte Carlo simulation as well as a nonparametric bootstrap based on resampling observations. In both cases results obtained with the delta method were conrmed.

A.R. Germani et al. / Ecological Economics 106 (2014) 6982

77

Table 4
Ordered probit estimation, IV ordered probit estimation and marginal effects (dependent variable: CO2 emissions).
Dependent variable:
CO2_area

Ordered probit estimations

IV Ordered probit estimations

Marginal effects (IV ordered probit)

Coefcients (P N |z|)

Coefcients (P N |z|)

Y=1

Y=2

Y=3

Y=4

ln(Ypc)

56.274
(0.129)
2.877
(0.142)
1.262
(0.057)
0.076
(0.613)
0.154
(0.538)
4.863a
(0.001)
6.425a
(0.000)
0.592
(0.210)
8.202a
(0.006)
0245
(0.508)
0.407b
(0.033)
0.467
(0.144)
0.895
(0.270)
0.000
(0.779)
310.6
311.453
312.359
103
0.14

55.531a
(0.000)
2.807a
(0.000)
1.371
(0.085)
0.102
(0.595)
0.217
(0.363)
4.713b
(0.010)
6.658a
(0.003)
0.658
(0.309)
7.942a
(0.002)
0.235
(0.442)
0.408a
(0.005)
0.439
(0.116)
0.784
(0.319)
0.000
(0.897)
0.000
0.000
.
103

13.024a
(0.000)
0.658a
(0.000)
0.321
(0.078)
0.023
(0.596)
0.050
(0.366)
1.105a
(0.007)
1.561a
(0.002)
0.154
(0.310)
1.862a
(0.002)
0.055
(0.439)
0.095a
(0.004)
0.103
(0.110)
0.183
(0.316)
0.000
(0.897)

4.342a
(0.000)
0.219a
(0.001)
0.107
(0.106)
0.007
(0.596)
0.016
(0.352)
0.368b
(0.034)
0.520b
(0.012)
0.051
(0.310)
0.621a
(0.008)
0.018
(0.453)
0.031b
(0.022)
0.034
(0.147)
0.061
(0.335)
0.000
(0.897)

2.935b
(0.033)
0.148b
(0.036)
0.072
(0.119)
0.005
(0.604)
0.011
(0.381)
0.249
(0.065)
0.351b
(0.043)
0.034
(0.350)
0.419
(0.062)
0.012
(0.459)
0.021
(0.053)
0.023
(0.175)
0.041
(0.346)
7.030
(0.897)

14.431a
(0.000)
0.729a
(0.000)
0.356
(0.087)
0.026
(0.595)
0.056
(0.362)
1.224b
(0.010)
1.730a
(0.003)
0.171
(0.306)
2.064a
(0.001)
0.061
(0.442)
0.106a(0.005)

ln2(Ypc)
lpcunidegree
lpcasia
lpcafr
lpcfemhead
lchildren
lelders
jud_inef
jud_eff
D
empl_RD
exp_RD
reg_exp
cut1
cut2
cut3
Number of observations
pseudo R2
Wald chi2
a
b

0.114
(0.115)
0.203
(0.320)
0.000
(0.897)

4,335,741.29

Signicant at the 1% level.


Signicant at the 5% level.

as income increases further, the demand for a cleaner environment reduces the level of pollution. This outcome might have very different origins. One possible interpretation (coherent with a post-materialist/
ecological modernization type of framework) would suggest that in the
richest Italian provinces industrial rms are more likely to invest in technology and innovation and to control air pollution. An alternative

Fig. 3. EKC and 95% condence interval (dependent variable: IWE_area).

interpretation, however, would suggest that polluting facilities might


migrate from wealthier areas to developing areas; this, in turn, should
be interpreted as a form of environmental load displacement and environmental cost shifting, rather than environmental awareness reached
by producers as income levels rise. These considerations should suggest
a cautious approach when interpreting EKC nding.
Results associated with High Education indicate that a 1%
increase in the share of people with a university degree translates
into a marginally signicant negative effect (ranging just below
0.4 percentage points) in the probability of attaining high levels
of air pollution emissions. So, these estimates suggest that high
education is a relevant variable for polluters to select where (not)
to place their polluting plants and for sorting based on existing
pollution levels.
Further, we can notice that the percentages of Asian and African
foreigners are never, in the four scenarios, statistically signicant.
Hence, the results provide no support for the contention that ethnicity could be associated with a disparate-impact discrimination for
environmental harm. When interpreting this nding, however, one
should bear in mind that in Italian provinces, Asian and African minorities account on average for only 0.27% and 0.64%, respectively,
of the population. Indeed, this might explain why our estimated
coefcients are never statistically signicant.
The results, further, indicate that a 1% increase in the number of
family households with a female head translates into an increase
just below 1.4 percentage points in the probability of attaining
high levels of air pollution emissions: so, these estimates suggest
that air releases are greater, on average, in provinces with greater

78

A.R. Germani et al. / Ecological Economics 106 (2014) 6982

proportion of female-headed households. The results for children


are somewhat similar: a 1% increase in the percentage of resident
children translates into an increase by around 1.8 percentage
points in the probability of attaining high levels of air pollution
emissions. Those are key results of our investigation suggesting
that greater concentrations of females as household heads and of
children are likely to be associated with increased levels of air
pollution.
These results are novel in the European context, but in line with
recent nding obtained for the US society. Downey and Hawkins
(2008a) found in a recent paper that in the continental United States,
female-headed families are overrepresented in neighborhoods with
high toxic concentration values, as are single-parent families with children less than six years old. The explanation put forward by the authors
is that these typologies of families are more likely than other families to
live in low-income and minority neighborhoods (Franklin et al., 1995;
Wilson, 1987) which, in turn, are disproportionately likely to house
environmental hazards (Brulle and Pellow, 2006). Their ndings are
robust and persist after controlling for race, income, and other neighborhood characteristics commonly associated with environmental hazard presence.
Also in our case, results are obtained after controlling for various economic and socio-demographic characteristics. Provided that our ndings
are based on cross-section data, caution must be taken in interpreting
the ndings reported here. However, these results appear to corroborate
the view (coherent with an environmental justice approach) that femaleheaded families as well as families with a larger number of children, result
in less political power to inuence siting of polluting facilities.
When looking at judicial inefciency measures, we can observe that
the sign of the coefcient of jud_inef is positive and highly signicant,
and the sign of the coefcient of jud_ef is negative but not statistically
signicant, implying that an increase in judicial inefciency (captured
by the average length of completed proceedings) is associated with an
increase in the probability of having high releases. In other words, provinces with high judicial inefciency and inadequate enforcement, measured by judicial delay, are more likely to experience more releases
than provinces with lower judicial efciency.
The spatial variable (i.e. release of neighboring provinces) is also
signicant, suggesting that the above-mentioned clustering effect is
occurring.27 As suggested by Costantini et al. (2013), the positive coefcient for environmental spillovers may be interpreted as an effect
of the technology adopted. The aggregation process of polluters in
relation to contiguous geographical areas is plausibly followed by
common choices in the adoption of cleaner or dirtier technologies.
This nding is in line with the technological effort variable, emp_RD,
which has a negative sign and is statistically signicant, suggesting a
negative correlation between technological effort and the level of emissions. Note that R&D expenditures carry an opposite sign, suggesting
that an extra nancial effort in R&D produces an increase in the emissions' level. This result could be due to the fact that R&D expenditures
are reported to provincial GDP values and also to the fact that data
refer to a general effort in R&D and not just environmentally related
R&D efforts (since this more specic data was not available at provincial
level). However, it is worth mentioning that this variable loses signicance when looking at marginal effects. Finally, regional environment
expenditure is never signicant.28

27
Recall that we considered CO2 release of neighboring provinces; however, as a robustness check, we replicate the regression model considering neighboring emissions of local
pollutants (i.e. IWE) obtaining highly comparable results. Results are available from the
authors upon request.
28
As an additional robustness check we run a restricted version of our regression model
excluding spatial, technological and regulatory variables. We observe that results are very
similar to those reported in Table 4 but the pseudo R2 is lower, suggesting that these variables improve the t of the estimation model. Data are available from the authors upon
request.

As already mentioned, we employed two different specications


of the dependent variable. Our results are largely conrmed in
terms of statistical signicance and coefcient signs, when using
CO2 as dependent variable (see Table 4). Specically, the statistical
signicance of the income variables and the signs are conrmed in
the IV ordered probit estimates, validating the existence of an inverse
U-shaped relationship between income and air releases. The same caveats
discussed above apply to the interpretation of this nding. Again ethnic
groups and elders are not statistically signicant but female-headed
household and children are highly signicant and bear positive signs.
Also judicial efciency/inefciency measures have the same sign and signicance as above. Most of the differences between the two estimations
concentrate in the technological and in the spatial regressors, which in
the case of CO2 estimates are never signicant. In conclusions, estimates
obtained using CO2 provide overall a similar picture to that obtained
using the IWE index. This, in turn, provides an indirect robustness check
to our ndings.
As a further robustness check we replicate our estimates using OLS
and 2SLS regression. Also in this case estimates (which are reported in
Table A4) largely conrm our ndings.

6. Conclusions
This paper presents a reduced-form statistical analysis on the relationship between air pollution and economic, institutional, technological and socio-demographic characteristics across the Italian provinces.
Our approach uses the level of air pollution emissions (twelve types of
local pollutant substances from the industrial sector), released by industrial plants in 2005 as the measure of environmental quality, merged
with 2001 data on institutional, technological and socio-demographic
characteristics at provincial level. The main objective is to ascertain
whether income, ethnicity and gender composition of the population can help explain releases and whether environmental injustice
arguments can be identied in Italy. Even though our analysis does
not allow us to determine why the gender composition of the population and income interact to produce environmental inequities, it
does allow us to increase our understanding of environmental inequality in Italy.
The estimates obtained by the ordered probit models indicate that
an increase in income is expected to increase the probability of higher
levels of air pollution releases, that is releases increase with income.
Our estimates are also consistent with an inverse U-shaped environmental Kuznets curve: once income exceeds a turning point, air pollution decreases with increasing income.
Our search for environmental injustice nds evidence that releases
tend to be higher in provinces with high concentration of females as
households' head and with high concentration of children. Our ndings
do not allow identifying any environmental discrimination based on
ethnicity suggesting that environmental justice issues are not likely in
Italy to be perceived in racial and ethnic terms but in terms of social categories and gender households' composition.
We nd also that greater judicial inefciency (or lenient law
enforcement) is associated with higher levels of pollution. This result
suggests that a better implementation, all through the country, of the
local enforcement of environmental laws can play an important role
in creating the conditions for better relationships between rms and
judicial institutions improving, thus, the overall environmental
quality.
Finally, we nd evidence of a clustering effect at the provincial
level suggesting the existence of agglomerative forces producing a
concentration of dirty activities in specic geographical areas, as
well as a technological effect (when technology capability is proxied
by R&D employees) suggesting that those provinces with higher
number of R&D employees are more likely to reduce their polluting
emissions.

A.R. Germani et al. / Ecological Economics 106 (2014) 6982

79

Appendix A. Data Sources, Variables Description and Robustness Check

Table A1
Pollutant description and threshold limit values.
In the following table the types of pollutants and their respective US and Italian threshold limit values are reported.
Pollutants description and threshold limit values
Code

Pollutant

US TLVs 8 h mg/m3

Italian laws on air


pollution 8 h

001
002
005
008
M01
M03
M04
M06
M07
M08
P11

Local
SO2 sulphur dioxide
NOX nitrogen monoxide
CO carbon monoxide
NH3 ammonia
As arsenic
Cr chromium
Cu copper
Ni nickel
Pb lead
Se selenium
Benz benzene (NIOSH)

13
30
55
35
0,2
1
1
1
0.05
0.2
0.32

80 g/m3
40 g/m3
10 mg/m3
10 mg/Nm3
6 ng/m3
0.05 mg/m3
1 mg/m3
0.1 g/m3
0.5 g/m3
0.2 g/m3
5 g/m3

006

Global
CO2 carbon dioxide

9000

100,000 t/year

Notes: *substance measured in micrograms; **substance measured in milligrams; ***substance measured in megagrams; ****substance measured in nanograms. In the ordered probit
regression analysis, all the different measurement units were converted into megagrams.

Table A2
Variable description and data sources.
Variable

Description

Source

air pollution emissions

Data on air emissions in all the Italian provinces expressed as ratio of emissions to area

per capita income

Natural logarithm of per capita income at provincial level, year 2001

infrastructure

Transportation infrastructural index

females
age15to34
age35to49
pcuniversitydegree

Number of female component the population


Number of people aged between 15 and 34 years old
Number of people aged between 35 and 49 years old
Percentage of the population which has an undergraduate university degree

pcfemalehead

Percentage of family households with a female as the head of the household

pcasia

Percentage of Asian residents

pcafr

Percentage of African residents

children

Percentage of children b6 years old

judicial efciency

Number of civil proceedings pending and concluded at courts located in a province divided by the
population
Trial length (expressed in years) divided by the population

ISPRA Air Emissions Provincial


Inventory year 2005
Elaborated from Institute
Tagliacarne our calculation
Il Sole 24 Ore Quality of Life
Report data-year 2001
ISTAT data Census data, 2001
ISTAT data Census data 2001
ISTAT data Census data 2001
Elaborated from ISTAT data
Census data 2001
Elaborated from ISTAT data
Census data 2001
Elaborated from ISTAT data
Census data 2001
Elaborated from ISTAT data
Census data 2001
Elaborated from ISTAT data
Census data 2001
Elaborated from ISTAT-year
2001
Elaborated from ISTAT-year
2001
Elaborated from ISTAT-year
2001
Elaborated from ISTAT-year
2001
Elaborated from ISTAT-year
2001
ISTAT

judicial inefciency
employees in R&D
expenditure in R&D
regional environmental
expenditure
territorial dummies: northern
provinces, central provinces

Number of researchers, technicians and other personnel engaged in R&D in the public administration
sector, universities and public and private enterprises (by provinces, per 1000 inhabitants)
Provincial expenses for R&D of public and private enterprises to GDP
Regional expenditure for environmental protection to GDP
The geographical distinction in the three macro-areas has been done following the denition of ISTAT.
North-west and North-East regions comprehend: Liguria, Lombardia, Piemonte, Valle d'Aosta, Friuli
Venezia Giulia, Emilia R., Trentino, Veneto.
Central regions: Toscana, Marche, Umbria, Lazio.
Southern regions: Abruzzo, Basilicata, Calabria, Campania, Molise, Puglia, Sardegna, Sicilia

80

A.R. Germani et al. / Ecological Economics 106 (2014) 6982

Table A3
Correlation matrices.

IWE
pcincome
pcunidegree
pcasia
pcafr
elders
children
pcfemhead
judicial_ineff
judicial_eff
D
emp_RD
exp_RD
reg_exp

CO2
pcincome
pcunidegree
pcasia
pcafr
elders
children
pcfemhead
judicial_ineff
judicial_eff
D
emp_RD
exp_RD
reg_exp

IWE

pcincome

pcunidegree

pcasia

pcafr

elders

children

pcfemhead

judicial_ineff

judicial_eff

emp_RD

exp_RD

reg_exp

1
0.116
0.209
0.084
0.050
0.038
0.133
0.029
0.207
0.030
0.070
0.057
0.031
0.048

1
0.127
0.461
0.527
0.251
0.600
0.626
0.355
0.045
0.008
0.071
0.040
0.060

1
0.073
0.374
0.294
0.139
0.091
0.069
0.031
0.079
0.053
0.028
0.033

1
0.536
0.188
0.097
0.095
0.339
0.192
0.098
0.134
0.169
0.116

1
0.107
0.069
0.150
0.370
0.070
0.013
0.154
0.156
0.025

1
0.875
0.227
0.106
0.021
0.009
0.114
0.115
0.023

1
0.150
0.096
0.001
0.056
0.140
0.151
0.062

1
0.023
0.026
0.078
0.145
0.072
0.027

1
0.012
0.075
0.027
0.042
0.035

1
0.325
0.286
0.169
0.020

1
0.1651
0.116
0.197

1
0.919
0.361

1
0.389

CO2

pcincome

pcunidegree

pcasia

pcafr

elders

children

pcfemhead

judicial_ineff

judicial_eff

emp_RD

exp_RD

reg_exp

1
0.171
0.317
0.190
0.006
0.122
0.058
0.062
0.247
0.043
0.086
0.113
0.057
0.095

1
0.127
0.461
0.527
0.251
0.600
0.626
0.355
0.045
0.008
0.071
0.040
0.060

0.073
0.374
0.294
0.139
0.091
0.069
0.031
0.079
0.053
0.028
0.03

1
0.536
0.188
0.097
0.095
0.339
0.192
0.098
0.134
0.169
0.116

1
0.107
0.069
0.150
0.370
0.070
0.013
0.154
0.156
0.025

1
0.875
0.227
0.106
0.021
0.009
0.114
0.115
0.023

1
0.150
0.096
0.001
0.056
0.140
0.151
0.062

1
0.023
0.026
0.078
0.145
0.072
0.027

1
0.012
0.075
0.027
0.042
0.035

1
0.325
0.286
0.169
0.020

1
0.165
0.116
0.197

1
0.919
0.361

1
0.389

Table A4
Linear regressions OLS estimation and 2SLS estimation.


^
ln Ypc
ln


^
Ypc

lpcunidegree
lpcasia
lpcafr
lpcfemhead
lchildren
lelders
jud_inef
lpcproceedings
lsigma_co2
addetti_rs
capacit_innov
environ_expenses
constant
Number of observations
R2
a
b

Signicant at the 5% level.


Signicant at the 1% level.

Dependent variable:
IWE_area

Dependent variable:
IWE_area

Dependent variable:
CO2_area

Dependent variable:
IWE_area

OLS coefcients

2SLS coefcients

OLS coefcients

2SLS coefcients

68.566
(0.383)
3.499
(0.396)
1.265
(0.077)
0.085
(0.567)
0.012
(0.965)
3.509b
(0.001)
5.362b
(0.000)
0.811
(0.089)
8.486b
(0.000)
0.305
(0.314)
0.294b
(0.000)
0.513a
(0.011)
0.856
(0.088)
0.000
(0.776)
362.721
(0.330)
103
0.41

39.913
(0.242)
2.030
(0.261)
0.869
(0.141)
0.084
(0.524)
0.134
(0.563)
3.463b
(0.008)
4.701b
(0.001)
0.556
(0.211)
6.861a
(0.021)
0.261
(0.444)
0.286
(0.072)
0.383
(0.150)
0.626
(0.354)
0.000
(0.950)
220.457
(0.176)
103
0.33

133.655
(0.154)
6.943
(0.157)
0.491
(0.434)
0.129
(0.364)
0.099
(0.706)
3.517b
(0.001)
5.097b
(0.001)
0.623
(0.146)
7.247a
(0.013)
0.282
(0.378)
0.265a
(0.047)
0.479
(0.085)
1.007
(0.175)
0.000
(0.718)
667.354
(0.138)
103
0.30

74.503
(0.032)
3.887a
(0.033)
1.042
(0.087)
0.023
(0.878)
0.106
(0.673)
3.865b
(0.003)
4.679b
(0.000)
0.925
(0.084)
7.709b
(0.002)
0.318
(0.366)
0.293b
(0.005)
0.490a
(0.020)
0.823
(0.126)
0.000
(0.744)
382.469a
(0.021)
103
0.42

A.R. Germani et al. / Ecological Economics 106 (2014) 6982

References
Agresti, A., 1990. Categorical Data Analysis. Wiley, New York.
Anderton, D.L., Anderson, A.B., Oakes, J.M., Fraser, M.R., 1994. Environmental equity: the
demographics of dumping. Demography 31, 229248.
Andreoni, J., Levinson, A., 2001. The simple analytics of the environmental Kuznets curve.
J. Public Econ. 80, 269286.
Aradhyula, S., Burns, M., Cory, D.C., 2006. On notions of fairness in environmental justice.
Cardon Research Papers in Agricultural and Resource Economics, n. 4. University of
Arizona.
Arora, S., Cason, T.N., 1999. Do community characteristics inuence environmental
outcomes? Evidence from the Toxics Release Inventory. South. Econ. J. 65, 691716.
Ash, M., Fetter, R.T., 2004. Who lives on the wrong side of the environmental tracks?
Evidence from the EPA's risk screening environmental indicators model. Soc. Sci. Q.
85 (2).
Bagnasco, A., 1977. Tre Italie. La problematica territoriale dello sviluppo italiano. Il Mulino,
Bologna.
Becker, G.S., 1968. Crime and punishment: an economic approach. J. Polit. Econ. 76,
169217.
Beckerman, W., 1992. Economic growth and the environment: whose growth? Whose
environment? World Dev. 20, 481496.
Been, V., Gupta, F., 1997. Coming to the nuisance or going to the barrios: a longitudinal
analysis of environmental justice claims. Ecol. Law Q. 24 (1), 155.
Been, V., 1994. Locally undesirable land uses in minority neighborhoods: disproportionate
siting or market dynamics. The Yale Law Journal 103, 13831421.
Bhagawati, J., 1993. The case for free trade. Sci. Am. 4249.
Bousquet, A.,Favard, P., 2000. Does S. Kuznets' Belief Question the Environmental Kuznets
Curves? IDEI Working Papers 107. Institut d'conomie Industrielle (IDEI), Toulouse.
Brainard, J., Jones, A., Bateman, I., Lovett, A., Fallon, P., 2002. Modelling environmental
equity: access to air quality in Birmingham, England. Environ. Plan. A 34, 695716.
Brooks, N., Sethi, R., 1997. The distribution of pollution: community characteristics and
exposure to air toxics. J. Environ. Econ. Manag. 32 (233), 250.
Brown, P., 1995. Race, class, and environmental health: a review and systematization of
the literature. Environ. Res. 69, 1530.
Brulle, R.J., Pellow, D.N., 2006. Environmental justice: human health and environmental
inequalities. Annu. Rev. Public Health 27, 3.13.22.
Bruvoll, A., Medin, H., 2003. Factors behind the environmental Kuznets curve a
decomposition of the changes in air pollution. Environ. Resour. Econ. 24, 2748.
Bullard, R.D., 1983. Solid waste sites and the black Houston community. Sociol. Inq. 53,
273288.
Bullard, R.D., Wright, B.H., 1987. Environmentalism and the politics of equity: emergent
trends in the black community. Mid Am. Rev. Sociol. 12, 2138.
Bullard, R.D.,Wright, B.H., 1989. Toxic waste and the African-American community. Urban
League Rev. 13, 6775.
Bullard, R.D., Mohai, P., Saha, R., Wright, B., 2010. Toxic wastes and race at twenty: why
race sill matters after all of these years. Environ. Law 38, 371.
Chaix, B., Gustafsson, S., Jerrett, M., Kristersson, H., Lithman, T., Boalt, A., Merlo, J.,
2006. Children's exposure to nitrogen dioxide in Sweden: investigating environmental injustice in an egalitarian country. J. Epidemiol. Community Health
60, 234241.
Cole, M.A., Neumayer, E., 2005. Environmental policy and the environmental Kuznets
curve: can developing countries escape the detrimental consequences of economic
growth? In: Dauvergne, P. (Ed.), International Handbook of Environmental Politics.
Edward Elgar, Cheltenham and Northampton, pp. 298318.
Cole, M.A., Rayner, A.J., Bates, J.M., 1997. The environmental Kuznets curve: an empirical
analysis. Environ. Dev. Econ. 2, 401416.
Coondoo, D., Dinda, S., 2002. Causality between income and emissions: a country
group-specic econometric analysis. Ecol. Econ. 40, 351.
Copeland, B.R., Taylor, M.S., 2004. Trade, growth and the environment. J. Econ. Lit. 42,
771.
Cory, D.C., Rahman, T., 2009. Environmental justice and enforcement of the Safe Drinking
Water Act: the Arizona arsenic experience. Ecol. Econ. 68, 18251837.
Costantini, V., Martini, C., 2010. A modied environmental Kuznets curve for sustainable
development assessment using panel data. Int. J. Global Environ. Issues 10 (1/2),
84122.
Costantini, V., Mazzanti, M., Montini, A., 2013. Environmental performance, innovation
and spillovers. Evidence from a regional NAMEA. Ecol. Econ. 89, 101114.
Dasgupta, S., Hamilton, K., Pandey, K., Wheeler, D., 2006. Environment during growth:
accounting for governance and vulnerability. World Dev. 34 (9), 15971611.
de Bruyn, S.M., van den Bergh, J., Opschoor, J.B., 1998. Economic growth and emissions:
reconsidering the empirical basis of environmental Kuznets curve. Ecol. Econ. 25,
161175.
Diarra, G.,Marchand, S., 2011. Environmental compliance, corruption and governance: theory and evidence on forest stock in developing countries. Etudes et DocumentsCERDI
(E2011.01).
Dinda, S., 2001. A note on global EKC in case of CO2 emission, Economic Research Unit,
Indian Statistical Institute, Kolkata mimeo.
Dinda, S., 2004. Environmental Kuznets curve hypothesis: a survey. Ecol. Econ. 49, 431455.
Downey, L.,Hawkins, B., 2008a. Single_mother families and air pollution: a national study.
Soc. Sci. Q. 89 (2), 523536.
Downey, L.,Hawkins, B., 2008b. Race, income and environmental inequality in the United
States. Sociol. Perspect. 51 (4), 759781.
Eloi, L., 2011. Issues in environmental justice within the European Union. Ecol. Econ. 70,
18461853.
Evans, G.W., Kantrowitz, E., 2002. Socioeconomic status and health: the potential role of
environmental risk exposure. Annu. Rev. Public Health 23, 303331.

81

Faini, R., Galli, G., Gennari, P., Rossi, F., 1997. An empirical puzzle: falling migration and
growing unemployment differentials among Italian regions. Eur. Econ. Rev. 41,
571579.
Filer, J.E., Kenny, L.W., Morton, R.B., 1993. Redistribution, income and voting. Am. J. Polit.
Sci. 37, 6387.
Forastiere, F.,Stafoggia, M.,Tasco, C.,Picciotto, S.,Agabiti, N.,Cesaroni, G.,Perucci, C.A., 2007.
Socioeconomic status, particulate air pollution, and daily mortality: differential
exposure or differential susceptibility. Am. J. Ind. Med. 50, 208216.
Franklin, D.L.,Smith, S.E.,McMiller, W.E.P., 1995. Correlates of marital status among African
American mothers in Chicago neighborhoods of concentrated poverty. J. Marriage
Fam. 57, 141152.
Germani, A.R., Scaramozzino, P., 2007. Modeling discretion through randomized strategies. SOASCeFiMS working paper n. 83 (http://www.cems.ac.uk/documents/
research-76.pdf).
Germani, A.R.,Morone, P.,Testa, G., 2012. Enforcement and air pollution: an environmental
justice case study. MPRA working paper n. 38656.
Goldman, B.A., 1991. The Truth About Where You Live: an Atlas for Action on Toxins and
Mortality. Random House, New York.
Gray, W.,Shadbegian, R., 2004. Optimal pollution abatement - whose benets matter, and
how much? Journal of Environmental Economics and Management 47, 510534.
Greenberg, M.R., 1993. Proving environmental inequity in siting locally unwanted land
uses. Risk Issues Health Saf. 4, 235252.
Greene, W.H., 2003. Econometric Analysis, 5th edition. University Prentice Hall, New York.
Grossman, G., Krueger, A., 1991. Environmental impact of a North American free trade
agreement. NBER working paper, 3914.
Grossman, G.M., Krueger, A.B., 1993. Environmental impacts of a North American free
trade agreement. In: Garber, P. (Ed.), The U.S.Mexico Free Trade Agreement. MIT
Press, Cambridge, MA.
Grossman, G.,Krueger, A.B., 1995. Economic growth and the environment. Q. J. Econ. 110,
353377.
Guiso, L.,Sapienza, P.,Zingales, L., 2004. The role of social capital in nancial development.
Am. Econ. Rev. 94 (3).
Hamilton, J.T., 1993. Politics and social costs: estimating the impact of collective action on
hazardous waste facilities. RAND J. Econ. 24, 101125.
Hamilton, J.T., 1995. Testing for environmental racism: prejudice, prots, and political
power? J. Policy Anal. Manag. 14 (1), 107132.
Harbaugh, W., Levinson, A., Wilson, D.M., 2002. Re-examining the empirical evidence for
an environmental Kuznets curve. Rev. Econ. Stat. 84 (3), 541551.
Havard, S., Deguen, S., Bodin, J., Louis, K., Laurent, O., Bard, D., 2008. A small-area index
of socioeconomic deprivation to capture health inequalities in France. Soc. Sci. Med.
67, 20072016.
Havard, S., Deguen, S., Zmirou-Navier, D., Schillinger, C., Bard, D., 2009. Trafc-related
air pollution and socioeconomic status: a spatial autocorrelation study to assess
environmental equity on a small-area scale. Epidemiology 20, 223230.
Kijima, M., Nishide, K., Ohyama, A., 2010. Economic models for environmental Kuznets
curve: a survey. J. Econ. Dyn. Control. 34, 11871201.
Koop, G., 1998. Carbon dioxide emissions and economic growth: a structural
approach. J. Appl. Stat. 25, 489515.
Kruize, H.,Driessen, P.P.,Glasbergen, P.,van Egmond, K.N., 2007. Environmental equity and the
role of public policy: experiences in the Rijnmond region. Environ. Manag. 40, 578595.
Kuznets, S., 1955. Economic growth and income inequality. Am. Econ. Rev. 65, 128.
List, J.A., Gallet, C.A., 1999. The environmental Kuznets curve: does one size t all? Ecol.
Econ. 31, 409423.
Lun, G.,Holzer, D.,Tappeiner, G.,Tappeiner, U., 2006. The stability of rankings derived from
composite indicators: analysis of the Il Sole 24 Ore quality of life report. Soc. Indic.
Res. 77, 307331.
Maddison, D., 2006. Environmental Kuznets curves: a spatial econometric approach.
J. Environ. Econ. Manag. 51 (2), 218230.
Martinez-Alier, J., Guah, R., 1997. Varieties of Environmentalism: Essays North and South.
Earthscan, London.
Mazzanti, M., Montini, A., Nicolli, F., 2009. The dynamics of landll diversion: economic
drivers, policy factors and spatial issues. Resour. Conserv. Recycl. 54 (1), 5361.
McLeod, H., Langford, I., Jones, A., Stedmand, J., Day, R., Lorenzoni, I., Bateman, I., 2000. The
relationship between socioeconomic indicators and air pollution in England and
Wales: implications for environmental justice. Reg. Environ. Chang. 1, 1885.
Millimet, D.L., List, J.A., Stengos, T., 2003. The environmental Kuznets curve: real progress
or misspecied models? Rev. Econ. Stat. 85 (4), 10381047.
Mohai, P., Bryant, B., 1992. Environmental racism: reviewing the evidence. In: Bryant,
Bunyan, Mohai, Paul (Eds.), Race and the Incidence of Environmental Hazards.
Westview Press, Boulder, Colorado, pp. 163176.
Naess, O., Piro, F.N., Nafstad, P., Smith, G.D., Leyland, A.H., 2007. Air pollution, social
deprivation, and mortality: a multilevel cohort study. Epidemiology 18,
686694.
Namdeo, A.,Stringer, C., 2008. Investigating the relationship between air pollution, health
and social deprivation in Leeds, UK. Environ. Int. 34, 585591.
Napolitano, O.,Bonasia, M., 2010. Determinants of different internal migration trends: the
Italian experience. MPRA paper n. 21734.
Nieves, L.A., Nieves, A.L., 1992. Regional differences in the potential exposure of U.S.
minority populations to hazardous facilities. Paper presented at the annual meeting
of the Regional Science Association in Chicago, November.
Olson, M., 1965. The Logic of Collective Action. Harvard University Press, Cambridge, MA.
Panayotou, T., 1993. Empirical tests and policy analysis of environmental degradation at
different stages of economic development, Geneva, Switzerland. International Labour
Ofce, working paper, WP238.
Panayotou, T., 2000. Economic growth and the environment, Harvard University. CID
working paper, p. 56.

82

A.R. Germani et al. / Ecological Economics 106 (2014) 6982

Pastor, M.,Sadd, J.,Hipp, J., 2001. Which came rst? Toxic facilities, minority move-in, and
environmental justice. J. Urban Aff. 23, 121.
Ringquist, E., 1997. Equity and the distribution of environmental risk: the case of TRI
facilities. Soc. Sci. Q. 78 (4), 811829.
Roca, J.E.,Padilla, E., Farre, M.,Galletto, V., 2001. Economic growth and atmospheric pollution in Spain: discussing the environmental Kuznets hypothesis. Ecological Economics 39, 8599.
Rotko, T., Koistinen, K., Hanninen, O., Jantunen, M., 2000. Sociodemographic descriptors of
personal exposure to ne particles (PM2.5) in EXPOLIS Helsinki. J. Expo. Anal. Environ.
Epidemiol. 10, 385.
Rotko, T.,Kousa, A., Alm, S.,Jantunen, M., 2001. Exposures to nitrogen dioxide in EXPOLISHelsinki: microenvironment, behavioral and sociodemographic factors. J. Expo. Anal.
Environ. Epidemiol. 11, 216223.
Sadd, J.L.,Pastor, M.,Boer, J.T.,Snyder, L.D., 1999. Every breath you take: the demographics
of toxic air releases in Southern California. Econ. Dev. Q. 13 (2), 107123.
Schikowski, T., Sugiri, D., Reimann, V.,Pesch, B., Ranft, U.,Krmer, U., 2008. Contribution of
smoking and air pollution exposure in urban areas to social differences in respiratory
health. BioMed Central Public Health 8, 179.
Selden, T., Song, D., 1994. Environmental quality and development: is there a Kuznets
curve for air pollution emissions? J. Environ. Econ. Manag. 27, 147162.
Shak, N., Bandyopadhyay, S., 1992. Economic growth and environmental quality: timeseries and cross-country evidence. World Bank Policy Research working paper, p. 904.
Soytas, U., Sari, R.,Ewing, T., 2007. Energy consumption, income, and carbon emissions in
the United States. Ecol. Econ. 62, 482489.

Stern, D., 2004. The rise and fall of the environmental Kuznets curve. World Dev. 32,
14191439.
Stroh, E.,Oudin, A.,Gustafsson, S.,Pilesj, P.,Harrie, L., Strmberg, U., Jakobsson, K., 2005. Are
associations between socioeconomic characteristics and exposure to air pollution a
question of study area size? An example from Scania, Sweden. Int. J. Health Geogr. 4, 30.
Thurow, L., 1980. The Zero-Sum Society: Distribution and the Possibilities for Change.
Basic Books, New York.
UNECE (United Nations Economic Commission for Europe), 1998. Environment for Europe, Fourth Ministerial Conference. Aarhus, Denmark. http://www.unece.org/
leadmin/DAM/env/documents/1998/ece/ece.cep.41.e.pdf.
UNEP (United Nations Environment Programme), 1999. Climate Change Information Kit.
UNEP Information Unit for Conventions.
Vitali, O.,Merlini, A., 1999. La qualit della vita: metodi e veriche. Riv. Ital. Econ. Demogr.
Stat. 53, 593.
Wheeler, B.W., Ben-Shlomo, Y., 2005. Environmental equity, air quality, socioeconomic
status, and respiratory health: a linkage analysis of routine data from the Health
Survey for England. J. Epidemiol. Community Health 59, 948954.
Wilson, W.J., 1987. The Truly Disadvantaged: the Inner City, the Underclass, and Public
Policy. University of Chicago Press, Chicago, IL.
York, R.,Rosa, E.A.,Dietz, T., 2003. STIRPAT, IPAT and ImPACT: analytic tools for unpacking
the driving forces of environmental impacts. Ecol. Econ. 46 (3), 351365.

Vous aimerez peut-être aussi