MSc Thesis International Economics and Business
Faculty of Economics and Business
The Nexus Between Economic Growth, Inequality and
Environmental Degradation: An Empirical Analysis
Supervisor: Dr. P. Rao Sahib
Co-assessor: Dr. D.H.M. Akkermans
Denitsa Georgieva S3075052
d.georgieva@student.rug.nl
Abstract
I investigate the theoretically ambiguous relationship between social inequalities and pollution using a panel dataset, with a substantially larger number of observations than most existing literature, including 161 developed, emerging, and developing economies for the 1990-2013 period. I examine specifically the effects of inequality on various pollution measures, namely carbon dioxide emissions, access to safe water, water pollution, and deforestation, by integrating different inequality measures into the formulation of the so-called environmental Kuznets curve. The results indicate that in most cases there is a clear trade-off between reduction of income inequality and reduction of environmental degradation. The study also provides evidence that there are considerable differences between emerging economies and other countries when it comes to the effects of inequality on pollution.
1
1. Introduction
The concept of sustainable development, presented by the Brundtland Commission report (WCED, 1987), suggests that economic, social and ecological factors should interact in such a way that long-term growth brings benefits for both people and the planet. Exploring the connections between the three pillars of sustainability has been an issue of research interest ever since the publication of this report, starting with early work by Tobey (1989), Bernstam (1990), and Beckerman (1992).
However, economic growth and improved wellbeing over the last few decades have come at the cost of high environmental pressure and led to a human-induced climate change (Schandl et al., 2016). Since the 1950s, average GDP per capita has increased over 20 times (World Bank, World Development Indicators Dataset) and this economic growth has been crucial for fighting poverty, improving living standards and promoting shared prosperity (Dollar et al., 2016). On the other hand, for the same period, the greenhouse gases have raised substantially: the global carbon emissions from fossil fuels alone have increased by over 400 percent (Boden et al., 2010). With the climate change accelerating (Karl et al., 2015), it is vital for economic growth to be sustainable and, thus, for people to explore the mechanisms that assure this. The question of who and how to govern the relationship between economic systems and the environment is yet to be answered. However, a line of thought supported by some empirical evidence has introduced the idea that “economic growth and respect of ecological constraints are compatible in the long run” (Clement and Meunie, 2010).
The World Bank’s World Development Report 1992, which was focused on environment, claimed that some environmental problems such as “most forms of air and water pollution… initially worsen but then improve as incomes rise”. This relationship is called the environmental Kuznets curve (EKC) and hypothesizes that the environmental degradation is an inverted U-shaped function of income per capita (Stern, 2004). This implies that eventually, growth reduces the negative impacts of economic activities on the environment.
2 mechanisms underlying the relationships between income, its distribution, and the environment may differentiate between countries at different stages of development (Grunewald et al., 2017; Drabo, 2010; Torras and Boyce, 1998). Older studies examined those interconnections for a limited number of countries and years due to data availability, while more recent ones tend to cover more observations and to differentiate between developed (high-income) and developing (low-income) countries.
The purpose of this study is to re-examine the growth-inequality-environment nexus for the recent period of 1990-2013, drawing on the work of Torras and Boyce (1998) by using expanded and improved data on inequality (Solt, 2016), that allows for the inclusion of a larger number of countries and observations than the existing literature. I will also contribute to the literature by paying particular attention to emerging economies, rather than distinguishing between developed and developing countries only. While emerging economies (EEs) are playing a growing role in the world economy, since their share in the world GDP has increased more than two-fold for the period of study (World Bank, World Development Indicators Dataset), there is little knowledge of the effects of inequality on environmental degradation in those countries. The EEs are characterized by rapid economic growth, which is generally expected to put higher pressure on the environment (Pao and Tsai, 2010). Therefore, the strength and even the direction of the relationships between per capita GDP, inequality, and environmental degradation in emerging economies may differ
from those in developed and developing countries. While this study focuses mainly on CO2
emissions as an indicator of environmental degradation, other variables (water pollution, access to safe water and deforestation) are also considered. Furthermore, following Torras and Boyce (1998), I use proxies for power inequality that contribute to capturing other dimensions of social inequality, rather than income distribution only. To summarize, this study aims to (i) empirically test the existence of EKC for various environmental quality measures for the recent period of 1990-2013; (ii) introduce inequality into the relationship using an expanded dataset on income inequality as well as additional social inequality measures; (iii) investigate if the results for emerging economies differ from those for developed and developing countries.
3 environmental degradation differs substantially between emerging economies and other countries.
The rest of the paper is organized as follows. Section 2 provides an overview of the theoretical framework on the growth-pollution and inequality-pollution links. Section 3 presents a literature review of the main empirical findings on the environmental Kuznets curve and the environmental impact of inequality. The relevance of examining those relationships in the context of the emerging economies is discussed in Section 4. Section 5 describes the data used, while the econometric analysis and the results are presented in Section 6. In Section 7 the policy implications of the results are discussed before concluding remarks are provided in Section 8.
2. Theoretical framework
2.1. An introduction to the environmental Kuznets curve (EKC)
The environmental Kuznets curve derives its name from the similarity with the inverted U-shaped relationship between inequality and economic growth observed by Kuznets (1955). It hypothesizes that environmental degradation will increase at early stages of economic development but will slow down beyond some turning point of per capita income. The effect of growth on the environment can be explained through three main mechanisms: ‘scale effects’, ‘composition effects’, and ‘technique effects’ (Grossman and Krueger, 1991; Dinda, 2004). According to the ‘scale effects’, the expansion of a country’s economy results in a greater consumption of common resources and higher environmental pressure (Dasgupta et al., 2002). The ‘composition effects’ refer to structural changes in the economy. While in the early stages of development the transition from agricultural to industrial society causes increasing environmental harm owing to a higher energy use, the following shift towards services would have the opposite effect. The negative impact of the ‘scale effects’ is also partly offset by the ‘technique effects’ since growing incomes generally lead to technological progress and therefore, to the substitution of ‘dirty’ technologies and production methods with more efficient ones.
4 Both Arrow et al. (1995) and Stern et al. (1996) stress that, even if the EKC hypothesis holds in some countries, this might be due to the effects of free trade on the distribution of polluting activities. As explained by Stern (1996, 2004), the Heckscher–Ohlin trade theory proposes that each country specializes in the production of goods requiring an abundance of its own factor endowments. For instance, developed countries would specialize in capital-intensive activities (such as manufacturing), requiring high-skilled labor, whereas developing countries would specialize in labor-intensive activities (such as natural resource extraction), requiring low-skilled labor. Then, decreasing environmental degradation in developed countries may result in increasing environmental degradation in developing ones (the so-called ‘race to the bottom’). Therefore, strengthening environmental regulations in developed countries might further stimulate the outsourcing of polluting industries towards poorer countries (Lucas et al., 1992). If the poorer countries themselves start implementing similar regulations later on, they will face a tough challenge since there might be nowhere else to outsource polluting activities (Stern et al., 1996).
Dasgupta et al. (2002) provide a critical review of EKC discussing two additional scenarios to the conventional EKC and the ‘race to the bottom’ scenario. Firstly, the ‘new toxic’ scenario claims that while some pollutants are getting reduced with income, industrial societies create new ones and ultimately, there is no overall improvement in environmental quality. Secondly, they also present an optimistic review of the conventional curve arguing that pollution levels may start falling at lower per capita income than before due to technological change. So, developing countries’ environmental degradation peaks could be lower than the ones reached earlier in developed countries.
The complicated and sometimes contradictory theoretical foundations of the EKC have resulted in concerns about the omission of key variables that could modify the growth-pollution link. This line of thought provoked interest in the third pillar of sustainability - social factors and, in particular, inequality.
2.2. Mechanisms linking inequality and environmental degradation
Higher income inequalities are generally considered undesirable and are found to have harmful secondary impacts on various factors, such as health and economic growth (Wilkinson and Pickett, 2010). The increasing attention on environmental worsening has brought about the question if environmental quality could be yet another phenomenon impacted negatively by greater income differences within countries. As today both inequality and the environment are experiencing crises, it might be that those effects are reinforcing each other (Berthe and Elie, 2015). Figure 1 summarizes schematically the main
5
Figure 1. Channels linking inequality and environmental degradation
The backbone of these theories is a paper published by Boyce (1994) that develops a hypothesis called the ‘power-weighted social decision rule’ (i, Figure 1). He explains that choices concerning the environment are based on a balance of power between those benefiting from a particular environmental degradation activity (‘winners’) and those who bear the net costs (‘losers’). If the ‘winners’ prevail in the long run, the overall environmental effects would diverge from the social optimum. Boyce suggests that higher income generally grants more power since economics and politics within a country tend to favor the preferences of the wealthy people. Therefore, income inequality would negatively impact the environment through power inequality. Further, severe inequalities also influence the valuations of the costs and benefits of environmental destruction actions through the higher purchasing power of the ‘winners’ that translates into a greater ability to alter market values, shape others’ preferences and direct the future path of technological advancements. The last dimension that Boyce (1994) describes is environmental time preference. He hypothesizes that greater inequality leads to both rich and poor generating short-run benefits and being less concerned about long-term environmental costs. According to him, in countries with a high degree of economic inequality, poor people are more careless about the future because they focus their lives on survival, while rich people have a high savings rate and tend to extract more resources in a short term since they fear reallocation of power and wealth.
6 greater willingness to pay for environmental protection (iv, Figure 1). Thus, equality does not systematically minimize environmental harm and a more unequal society could even generate less environmental degradation. The debate between Boyce and Scruggs underlines the complexity of mechanisms linking inequality and the environment.
Another channel through which inequality may determine environmental quality is social cooperation (ii, Figure 1). As suggested by Borghesi (2000), it is more difficult to achieve public policy solutions to environmental problems in unequal societies because of less trust and cooperation. If the economic growth in a country disproportionately benefits wealthy people, the median voter’s preferences would not be switched towards ‘pro-environment public expenditure’ (Magnani, 2000). Studies suggest that social trust between individuals stimulates collective action towards protecting the environment such as recycling (Sonderskov, 2011) and using public transportation (Van Lange et al., 1998), and that income inequality within a country may hamper the development and diffusion of new environmental technologies (Vona and Patriarca, 2011).
Further, the consumption patterns in a country can also be linked to the level of inequality. Veblen (1899) argues that in more unequal societies individuals consume more as they tend to compare themselves with the wealthier social classes (iii, Figure 1). Poorer members of the society will try to meet the living standards they desire by earning and spending more, and overall the consumption and pollution levels will be higher than those in an egalitarian society (Cushing et al., 2015). This line of reasoning fits, for example, with the recent developments in the United States: simultaneous increase in income inequality, household debts, and working hours (Frank, 2012).
7
3. Empirical findings of the literature
3.1. Empirical studies on the environmental Kuznets curve
Studies on the environmental Kuznets curve emerged in the early 1990s with Grossman and Krueger's (1991) paper on the potential alleviation of pollution problems in Mexico due to NAFTA, and Shafik and Bandayopadhyay's (1992) background study for the World Development Report 1992. Most studies employ the following reduced form equation to test for the presence of the EKC:
𝑷𝑶𝑳𝒊𝒕 = 𝜷𝟎+ 𝜷𝟏𝑮𝑫𝑷𝒊𝒕+ 𝜷𝟐𝑮𝑫𝑷𝒊𝒕𝟐 + 𝜷𝟑𝑿𝒊𝒕+ 𝜺𝒊𝒕, (1)
where POL is the specific environmental indicator; GDP is GDP per capita; X relates to control variables of influence on environmental degradation; the subscript 𝑖 denotes a country; 𝑡
refers to time; and 𝜺𝒊𝒕 is the error term. Therefore, the inverted U-shaped relationship will
exist if 𝜷𝟏>0 and 𝜷𝟐<0. Some studies also include a cubic term of GDP per capita in order to
detect a possible second turning point.
Grossman and Krueger (1991) estimate the EKC for sulfur dioxide (SO2), smoke and
suspended particles using the Global Environment Monitoring System (GEMS) dataset that monitors air quality in urban areas throughout the world. Their regressions include a cubic functional form of per capita GDP (in PPP), various geographical variables, a time trend, and a trade intensity variable. They find that the levels of sulfur dioxide and smoke reach a turning point at around a GDP per capita of US$4,000-5,000 in 1985 dollars (but then perhaps rises again at income levels over US$10,000–15,000) and that the mass of suspended particles monotonically decreases with increasing GDP per capita. Shafik and Bandayopadhyay (1992) estimate the EKC for ten pollutants, using three functional forms: log-linear, quadratic, and cubic. They find that the EKC is confirmed for local air pollutants, since they start decreasing when countries approach middle-income levels, but the results are mixed for other indicators. For instance, CO2 emission rates and river quality seem to
worsen unambiguously with growing incomes.
Most of the subsequent studies also confirmed empirically the existence of the EKC for a number of air and water pollution indicators. Grossman and Krueger (1995) provide
evidence that air concentration of SO2 and 11 indicators of water quality improve together
with economic growth after a critical level of income per capita is reached. They reveal the turning point to be at GDP per capita of less than US$8,000 (1985 dollars). De Bruyn et al. (1998) find that economic growth has a direct positive effect on the levels of emissions of carbon dioxide (CO2), nitrogen oxides (NOx) and sulfur dioxide (SO2) in the Netherlands,
8 improvements. A recent city-level study of Japan conducted by Kasuga and Takaya (2016), also supports the EKC hypothesis. They conclude that environmental indicators on air and river quality improve as incomes increases which is consistent with previous literature as income levels of Japan are above the turning point.
With respect to less common environmental measures, some of which are also examined in this study, the EKC is most often not supported. Indicators with direct effects on human health, such as access to urban sanitation and access to safe drinking water, tend to improve monotonously with growing GDP per capita (Dinda, 2004), while the empirical evidence in the case of deforestation is quite controversial (Bhattarai and Hammig, 2001; Koop and Tole, 2001).
Notwithstanding the abundant EKC literature, it has failed to provide conclusive results and economic mechanism reasoning on the interconnections between economic growth and environmental degradation (Stern, 2004). Even among the studies supporting the EKC, there is a lack of agreement on the turning points beyond which different pollutant levels start improving. In accordance with this, Copeland and Taylor (2004) state that “Our review of both the theoretical and empirical work on the EKC leads us to be skeptical about the existence of a simple and predictable relationship between pollution and per capita income’’(p. 8).
3.2. Empirical studies on the nexus between inequality and environmental degradation A sizable empirical literature on the effects of inequality on environmental degradation has emerged in the recent period. Table 1 gives an overview of the main contributions to this topic that provide valuable insight to the present study. Generally, the empirical analysis is based on the following function of pollution indicators:
𝑷𝑶𝑳 = 𝒇 (𝑮𝑫𝑷, 𝑮𝑰𝑵𝑰, 𝝅, 𝑿), (2)
where POL is the pollution indicator; GDP refers to the GDP per capita; GINI is the Gini coefficient; π represents political variables/power inequality measures; and X is vector of control variables.
9
Table 1. Summary of empirical findings from studies examining the effect of inequality on environmental degradation
Authors Period of
study
Data Dependent variable Inequality measure
Effect of higher inequality on the dependent variable (I=improves; D=degrades; NS= not significant)
Other variables Estimator Linear/ Log model
Grunewald et al., 2017 1980-2008 158 countries Carbon dioxide per capita GINI I (LI)/ D (HI)
*LI=low income; HI=high income
GDP, GDP*GINI, Agriculture VA, Manufacturing VA, Services VA, Urban Population, Polity
OLS, FE, Group FE
Log
Clement and Meunie, 2010
1988-2003 83 developing and transition countries
Sulfur dioxide GINI NS GDP, GDP2, GDP3, Political rights FE, GMM Linear
Organic water pollution (BOD) I FE
D GMM
Drabo, 2010 1970-2000 90 countries Sulfur dioxide Carbon dioxide
Organic water pollution (BOD)
GINI D (developing countries only) D
D
GDP, GDP2,Population density, Primary school enrollment, Fertilizer use, Institution quality, FDI, Trade openness
2SLS FE Log
Heerink et al., 2001 1985 64 countries, including 16 countries in Sub-Saharan Africa No safe water No sanitation Deforestation Sulfur dioxide Particulate matter Carbon dioxide per capita Nitrogen soil depletion Phosphorus soil depletion
GINI D D D NS NS I I I GDP, GDP2 OLS Log
Koop and Tole, 2001 1961- 1992 48 tropical developing countries
Deforestation GINI GINI for land
D GDP, GDP growth, Population density, Change in population
FE Linear
Borghesi, 2000 1988-1995 37 countries Carbon dioxide per capita GINI I GDP, GDP2, GDP3, Population Density, Industry VA
OLS Linear and Log
NS FE
Ravallion et al., 2000 1975-1992 42 countries Carbon dioxide per capita GINI I GDP, GDP2, GDP3, Population, Time trend
OLS, FE Log Scruggs, 1998 1979-1990 25-29 countries Sulfur dioxide
Particulates Fecal coliform Dissolved oxygen GINI NS I NS D
GDP, Democracy, Industrial site, Period dummies
OLS Linear
Torras and Boyce, 1998 1977-1991 287 stations/ 58 countries Sulfur dioxide Smoke Heavy particles Dissolved oxygen Fecal coliform Safe water (%) Sanitation (%)
GINI D (LI) / I (HI) D (LI) /NS (HI) I (LI) /NS (HI) I (LI) /NS (HI) NS (LI)/ I (HI) D (LI) /NS (HI) NS (LI) /NS (HI) GDP, GDP2, GDP3, Political rights and civil liberties, Literacy rates
10
greater inequality is found to be associated with more SO2 emissions and smoke and less
access to safe water in income countries, but not in high-income ones. Also, in low-income countries is where literacy and political freedom are found to be strong predictors of environmental quality. Clement and Meunie (2010) also use political rights (based on the Freedom House index) as a measure of power inequality and find that they are associated
with lower SO2 emissions across 83 developing and transition countries, even though the
Gini coefficient has no significant effect on SO2 emissions.
Scruggs (1998) performs cross-national empirical analysis for the period 1979-1990 using a pooled model. In his main regression, he obtains conflicting results since income inequality is only significant for two out of four pollutants examined. It is found to cause a more deteriorated environment when dissolved oxygen, that measures water pollution, is used as an indicator but less environmental degradation in the case of particulates that measure air pollution. Those findings seem to confirm Scruggs’ view that inequality is not systematically linked to lower environmental quality.
Examining whether environmental degradation is a channel through which inequality affects health, Drabo (2010) performs an empirical analysis on 90 countries for the 1970-2000 period. Even though he shows that income inequality has a significant negative effect
on air pollution indicators (CO2 and SO2) and on water pollution (BOD), he stresses that this
effect varies between developed and developing countries and is mitigated by good institutions. A less extensively covered environmental degradation indicator, deforestation, that is also considered in this study, is investigated by Koop and Tole (2001). Based on a panel of 48 tropical developing countries in the period 1961-1992, their empirical results show that deforestation rates, measured by the percentage annual decrease in forest area, are lower in more egalitarian countries.
Turning to studies that focus on CO2 emissions, which is the pollutant of main interest
in this study, Ravallion et al. (2000) demonstrate that higher inequality both between and within countries decreases carbon emissions for a dataset of 42 countries. Even though they estimate a simple pooled OLS model and a fixed effects model, they refer to the pooled model as their preferred specification. Questioning this choice, Borghesi (2000) argues that the results strongly depend on the model specification and that a fixed effects model better depicts the reality in this context. Basing his analysis on 37 countries for the period 1988-1995 and estimating the models in both levels and logs, he finds that more income inequality
leads to lower CO2 emissions according to the pooled OLS model but this impact turns out to
11 income. So, they reach similar results as Heerink et al. (2001) for low and middle-income countries, but find out that the opposite is true for upper middle-income and high-income
countries: higher income inequality is associated with more CO2 emissions.
Clearly, despite the growing amount of research contributions, the literature on the relationship between income inequality and the environment has yet to reach a theoretical or empirical consensus. The main reason for the conflicting outcomes is the particular environmental degradation measure examined (air quality, water or soil pollution, deforestation, etc.) as different indicators may have different dynamics. Further, the group of countries, the control variables included, and the specific econometric model applied may also lead to big differences in the results.
4. The case of emerging economies (EEs)
The emerging economies (EEs) are of particular interest in this study because of their growing importance in the world economy and lack of previous literature on the topic focused on them. Most of the EEs are undergoing a process of rapid industrialization which means that according to the ‘scale’ and the ‘composition effects’ of the EKC, a high environmental pressure shall be observed. Therefore, a sustained growth of EEs would lead to vast increments of global consumption that may follow the unsustainable consumption patterns of developed countries. Indeed, as shown in Figure 2, the growth in GDP per capita
has come at the expense of higher CO2 per capita emissions. That means that EEs pose large
Figure 2. Average CO2-GDP per capita values in emerging countries. Data Source: World
12 new challenges to environmental sustainability. Such concerns are also supported by the
growing contribution of EEs to global CO2 emissions. It has increased substantially to 52% in
2013 from under 40% in the 1990s, at an average rate of 1% per year for the last 10 years of the study period (World Development Indicators). The acceleration of emissions, mainly due to increased energy consumption in BRICS countries, has made the extent to which economic development relies on energy an issue of concern for policy makers and energy analysts around the globe (Dai et al., 2016). Therefore, a continued growth of EEs could be an engine for world economy growth and could provide opportunities for a large share of the world’s population, especially in major EEs such as China and India, but this may impose high environmental damage. So, it is particularly important for those countries to explore mechanisms for environmental control and to follow patterns leading to sustainable growth.
While the strong economic growth in EEs have contributed to increasing incomes, income disparities have remained high, and have even shown an upward trend in most EEs (Babu et al., 2016). The few exceptions where inequality has decreased include mainly Latin American countries, such as Brazil and Peru, which have marked some progress but as of 2013, still have extremely high Gini coefficients of around 45. With major increases in income inequality in post-communist countries in Central and Eastern Europe and China, on average EEs are more unequal today than they were in 1990 (Solt, 2016). While inequality is a problem in both developed and emerging countries, the forces that create it vary substantially. The main reasons for inequality in EEs are large informal sectors, urban-rural regional divides, gaps in access to education, and limited employment opportunities for women (OECD General Economics and Future Studies, 2011). As both, environmental pressure and income inequality currently have high levels in EEs, this study provides an examination of how those developments interact and what the corresponding implications are.
5. Data
A list of all variables, their definition, the data source and examples of previous studies that use the same data source in their analyses is provided in Appendix I. In this section I will further discuss the choice of the variables used and their source, the expected directions of the relationships, and the final sample.
5.1. Dependent Variables
CO2 emissions
13
environmental quality is carbon dioxide (CO2) emissions in metric tons per capita. CO2 is the
largest contributor to global warming and the consequences of its emissions are global which differentiates it from other pollutants with primarily home country effects (for example, water pollution). As discussed by Stern (2004), the spatial dimension of pollution may be an important characteristic since environmental policies are more likely to be applied to local environmental problems, rather than global ones.
The increase of CO2 emissions in the past decade reflects mainly the steady increase in
fossil energy consumption in emerging economies (NEAA, 2015). Thus, CO2 emissions are
currently the most discussed environmental issue as their reduction requires international cooperation. The data is taken from World Development Indicators (2017) that use the dataset of US Oak Ridge National Laboratory (ORNL) at the Carbon Dioxide Information Analysis Center. ORNL estimates refer to carbon dioxide emissions from the burning of fossil fuels and the manufacture of cement. Despite the fact that the ORNL dataset is recognized
as one of the richest and most harmonized sources on CO2 emissions, it faces two major
limitations (Borghesi, 2000). Firstly, estimations may differ from actual values as annual emissions for each fuel group are calculated by multiplying the average carbon content for this group by the annual fuel consumption (Marland and Rotty, 1984). Secondly, emissions from other sources such as land-use, livestock and deforestation are not taken into account, which may lead to underestimation of total emissions for countries with large agricultural sectors.
Other indicators of environmental degradation
14
5.2. Independent variables
GDP per capita
To measure income per capita, I use World Bank’s World Development Indicators 2017. More specifically, I apply GDP per capita in PPP constant 2011 international dollars so that the values are relevant for cross-country analysis. GDP PPP adjusts GDP in such a way that each dollar can buy the same amount of goods or services irrespective of the country. Since the theory predicts that environmental degradation could first increase at low income levels but then decrease at high income levels, I include the square of GDP PPP per capita to allow detection of a non-linear relationship. I also use the cubic term of per capita GDP that allows for a second turning point, as the decrease of pollution beyond a certain income level may also have a temporary effect (Dinda, 2004).
Income inequality
My main measure of inequality, the Gini coefficient, is taken from the newly available improved SWIID dataset (Solt, 2016). It is based on the World Income Inequality Database (WIID) (UNU-WIDER, 2015) and standardizes this heterogeneous and unbalanced dataset by using several other data sources and multiplying imputing values in order to make the data comparable over time and between countries (Scholl and Klasen, 2016; Grunewald et al., 2017). As a result, SWIID offers coverage double that of the next largest income inequality dataset and is considered a substantial improvement in terms of data quality and country coverage. It currently consists of comparable Gini indices of net and market income inequality for 176 countries from 1960 to 2015. While the market Gini stands for gross income inequality (before taxes and transfers), net income inequality represents the post-government re-distribution efforts in the form of taxes and subsidies. Following Solt (2016), who notes that “most researchers will find the net-income inequality series to be the best suited to their needs conceptually”, and Babu et al. (2016) who demonstrate that net inequality levels in emerging countries closely follow the gross inequality levels, I will use net inequality for my research. According to Boyce’s theory, inequality is expected to have a negative impact on environmental quality but as stated by Clement and Meunie (2010) and found in empirical literature, that may heavily depend on both the country and the specific pollutant analyzed.
Power inequality
15 of pollution. Barrett and Graddy (2000) state that the inclusion of political freedom variables is crucial to explaining the environmental quality in a country, as it would “depend also on citizens being able to acquire information about the quality of their environment, to assemble and organize, and to give voice to their preferences for environmental quality; and on governments having an incentive to satisfy these preferences by changing policy, perhaps the most powerful incentive being the desire to get elected or re-elected” (p. 434). In this study, I consider two proxies for power inequality introduced by Torras and Boyce (1998): political rights and civil liberties indices developed by the Freedom House; and literacy rates taken from World Development Indicators Database (2017). Since 1972, Freedom House publishes an annual report called Freedom in the World, on the degree of democratic freedoms by assessing the current state of political and civil rights in 195 countries which are
widely used in the literature. The political rights index reflects whether the government
came to power through free and fair elections, and whether political pluralism exists and the opposition has a realistic opportunity to take power through elections. The civil liberties index reflects the freedom of expression and belief, whether the rule of law applies to everybody, and the constraints on the rights of the individuals to debate, demonstrate, and
to form organizations. For the purposes of this study, I compose an aggregate of the two
variables, each of which is measured on a 1-7 scale with one meaning the most freedom, and seven the least. Following Torras and Boyce (1998), I subtract the sum from 14 and obtain a 0-12 scale in which lower values representing less freedom and vice versa. Data on literacy rates is based on national and international surveys and reflects the percentage of people aged 15 and above who can both read and write. If a country has a higher rating for political rights and civil liberties, or in other words is more democratic, equality of power would be expected to be more widespread. With power highly related to information access, a higher literacy rate would mean that a larger share a country’s population is empowered and, thus, the distribution of power is more equitable (Torras, 2005). Clearly, income and power inequality are expected to be correlated to some extent but their effects on environmental quality are expected to differ depending on the pollutant and the country group under examination.
5.3. Control variables
16 country’s GDP. Population density and urbanization are generally expected to worsen environmental quality, since more concentration of people in a certain area would put more pressure on the resources available in this area. However, it might be that they “facilitate some environmental improvements, for example, through economies of scale in the provision of sanitation facilities” (Torras and Boyce, 1998, p. 153). It could also be that higher population density is associated with greater concern for environmental damage and, thus, promotes taking actions to reduce it (Scruggs, 1998). If industry value added represents a large share of GDP, that would suggest that a country is specialized in more pollution-intensive activities like manufacturing and mining, and hence, environmental quality would
be lower, especially when it comes to CO2 emissions. Data on the three variables is drawn
from the World Bank World Development Indicators (2017). 5.4. Sample size
The initial sample based on GDP per capita and Gini index data availability contains 161 countries of which 36 are developed, 23 are emerging, and 102 are developing. In order to allocate a country to a particular group, I follow the IMF (2015) that classifies 39 economies as advanced (developed) and 23 countries, including BRICS, Turkey, and Mexico as emerging. The final sample size varies depending on the data available for each pollutant and set of explanatory variables. First, the relationship between GDP and the environment is examined
using a panel data of 158, 88, 158, and 159 countries for CO2 emissions, organic water
pollution, no access to safe water, and deforestation respectively for the period 1990-2013 (1990-2007 for water pollution). Then, introducing different measures of inequality as additional explanatory variables reduces slightly the samples of countries to respectively 142, 82, 143, and 144. The number of observations for each estimation is presented in the regression results tables.
6. Econometric analysis 6.1. Descriptive statistics
Table 2 presents the descriptive statistics of all variables in the regression model. GDP per capita ranges from US$351 to 95,578 (in PPP), with a mean of US$13,083, while the mean value of the Gini coefficient is 37.90. Detailed descriptive statistics per country and country group (developed, emerging, developing) are presented in Appendix II.
17
Table 2. Descriptive statistics
Variable
Number of
observations Mean
Standard
deviation Skewness Kurtosis Min Max
CO2 emissions per capita 3738 3.91 4.58 2.02 9.49 0.01 36.82
Organic water pollution (BOD) 887 0.18 0.06 1.97 8.41 0.09 0.45
% of population with no access to safe water 3738 16.73 18.52 1.18 3.50 0 86.80
Deforestation (% annual reduction) 3704 0.06 1.37 1.67 49.33 -7.96 26.92
GDP per capita 3741 13083.13 14097.19 1.76 6.67 350.97 95577.89 Gini 2969 37.90 8.98 0.26 2.53 14.06 67.21 Political rights 3796 7.45 3.71 -0.34 1.84 0 12 Literacy rate 3208 82.43 21.15 -1.30 3.78 10.89 99.99 Population density 3844 200.73 702.50 8.13 71.76 1.41 7636.72 Industry share 3524 28.95 11.39 1.45 8.00 3.33 96.74 Urban population 3864 53.43 23.22 0.03 2.06 5.42 100
Table 3. Correlation matrix
Variable (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11)
CO2 emissions per capita (1) /
Organic water pollution (BOD) (2) -0.38 /
% of population with no access to safe water (3) -0.58 0.34 /
18
far as environmental quality indicators are concerned, CO2 emissions seem to be an outlier
as suggested. While the other three environmental degradation indicators are positively correlated with higher inequality and negatively correlated with per capita income, exactly
the opposite holds for CO2 emissions. Those relationships are based on simple correlations
and do not take into account the possible impacts of other variables. The following econometric sections provide a solution to this.
In the first instance of the econometric analysis, I examine the relationship between environmental quality and per capita GDP to check for the existence of the EKC. Second, I introduce potentially omitted variables in the relationship, namely income and power inequality. Lastly, I examine whether there is a differential impact of inequality on the environment depending on the group of countries under study.
6.2. Analysis of the Environmental Kuznets Curve (EKC)
The studies on the EKC differ in three main aspects: (i) the choice of quadratic or cubic functional form for GDP per capita; (ii) the choice between linear and log-linear models; and (iii) the specification used: pooled ordinary least squares, fixed effects or random effects (Borghesi, 2000).
Most previous studies choose a quadratic functional form for their estimations as it describes the conventional inverted U-shape of the EKC. However, there are several authors that have found that for some environmental quality measures a cubic functional form can better describe the relationship between income and environmental quality (Grossman and Krueger, 1995; Torras and Boyce, 1998; Borghesi, 2000). Therefore, I include the cube of GDP PPP to make my study comparable with others and to allow for the possibility of detecting a second turning point (N-shaped curve).
Turning to the second issue, generally both linear and log-linear models fit the data well among the EKC literature (Borghesi, 2000). One of the main drawbacks of the log-linear model is that it restricts the variables to be positive. As far as EKC is concerned, it may be reasonable to consider the possibility that emissions or some other measures of environmental quality can become negative for high-income levels (Galeotti and Lanza, 2005). Therefore, generally a linear specification will be preferred as this requirement is relaxed and its coefficients provide an immediate interpretability of the shape of the relationship. However, a log transformation is a useful tool in case that the dependent variable does not follow a normal distribution needed for obtaining reliable results. A normal quantile plot and a Shapiro-Wilk W test for normality reject the hypothesis for a normal distribution for all indicators of environmental degradation (see Appendix III). While log
19 pollution and improves the goodness of fit for those models, this does not hold for access to
safe water and deforestation. Therefore, I apply a log-linear model when CO2 emissions and
water pollution are taken as dependent variables but a linear one for access to safe water and deforestation.
As far as the model specification is concerned, I will estimate a pooled ordinary least squares model (OLS) and a fixed effects model (FE). All OLS regressions are estimated with robust standard errors as White test indicates the presence of heteroscedasticity (see Appendix III). This method captures both cross-country and inter-temporal differences within countries in examining the effect of income and inequality on the environment. On the other side, the fixed effects model allows controlling for unobserved heterogeneity at the country and year level. The country dummies control for time-invariant omitted variables bias, whereas the year dummies control for global effects such as shocks that otherwise might be captured by the explanatory variables. In this specification, only the variance within countries over time is used. The F-test performed confirms that the use of fixed effects is more relevant since the hypothesis that there are no fixed cross-country differences is rejected at 1 percent level of significance for all measures of environmental degradation (see Appendix III).
Turning to the choice between fixed effects and random effects, only a few studies include a random effects specification (Selden and Song, 1994; Magnani, 2000). Even though a fixed effects model has as a disadvantage the loss of information for the variation between countries, I will argue that it is more adequate for investigating the growth-inequality-environment relationships, since I include a sample of almost all the countries in the world and the explanatory variables are time-varying. An important feature of the random effects is that they assume that the country effects are not correlated with the explanatory variables. For example, that would mean that the efficiency of production or environmental policies in a country are not correlated with the country’s GDP which is not a reasonable assumption. This reasoning is further confirmed by the Hausman test (reported in all estimation tables) that rejects the null hypothesis that the coefficients estimated by the random effects estimator are the same as the ones estimated by the fixed effects estimator for all regressions (except when deforestation is taken as a dependent variable but this model has a very low adjusted R-squared for both FE and RE). Therefore, the fixed effects model is my preferred specification for all regressions.
20
𝑷𝑶𝑳𝒊𝒕= 𝜷𝟎+ 𝜷𝟏𝑮𝑫𝑷𝒊𝒕+ 𝜷𝟐𝑮𝑫𝑷𝒊𝒕𝟐 + 𝜷𝟑𝑮𝑫𝑷𝟑𝒊𝒕 + 𝜷𝟒𝑫𝑬𝑵𝑺𝒊𝒕+ 𝜷𝟓𝑰𝑵𝑫𝒊𝒕+ 𝜷𝟔𝑼𝑹𝑩𝒊𝒕+ 𝜺𝒊𝒕, (3)
𝑷𝑶𝑳𝒊𝒕= 𝜷𝟎+ 𝜷𝟏𝑮𝑫𝑷𝒊𝒕+ 𝜷𝟐𝑮𝑫𝑷𝒊𝒕𝟐 + 𝜷𝟑𝑮𝑫𝑷𝒊𝒕 𝟑 + 𝜷𝟒𝑫𝑬𝑵𝑺𝒊𝒕+ 𝜷𝟓𝑰𝑵𝑫𝒊𝒕+ 𝜷𝟔𝑼𝑹𝑩𝒊𝒕+ 𝝁𝒊𝒕+ 𝝀𝒊𝒕+ 𝜺𝒊𝒕, (4)
where POL represents the level of pollution or environmental degradation (CO2
emissions/organic water pollution/no access to safe water/deforestation); GDP is GDP per capita; DENS is population density; IND is industry share in GDP; URB is the share of urban
population; 𝝁𝒊𝒕 measures the country fixed effects (𝑖 = 1, … , 𝑁); 𝝀𝒊𝒕 measures the year fixed
effects (𝑡 = 1, … , 𝑇); and 𝜺𝒊𝒕 is the error term. Table 4 presents the regression results
obtained from equation (3) and equation (4) for each of the pollution variables described in the data section.
Table 4. Regression for the determinants of environmental quality (Model 1), pooled OLS and fixed effects specifications
Log CO2 emissions
Log Organic water
pollution (BOD)
No access to safe
water Deforestation
Variable OLS FE OLS FE OLS FE OLS FE
GDP (per capita) 0.30*** 0.10*** -0.03*** -0.005 -3.36*** 0.21** -0.07*** -0.03 (47.45) (17.11) (-5.80) (-1.14) (-40.08) (2.49) (-10.67) (-1.12) GDP Squared -0.76*** -0.24*** 0.06*** -0.02* 8.99*** 1.38*** 0.18*** 0.13** (-34.47) (-15.64) (4.61) (-1.69) (30.62) (5.87) (8.69) (2.07) GDP Cubed 0.54*** 0.16*** -0.04*** 0.02*** -6.43*** -1.08*** -0.12*** -0.08 (25.64) (11.97) (-3.93) (2.80) (-23.23) (-5.38) (-7.10) (-1.52) Population Density 0.01*** 0.00 -0.01*** 0.00 -0.26*** -0.61*** 0.00 -0.08*** (4.60) (0.44) (-16.76) (0.09) (-8.11) (-6.07) (1.39) (-2.76) Industry Share 0.03*** 0.01*** -0.01*** -0.00 0.09*** -0.10*** -0.01*** -0.01** (15.76) (6.38) (-6.02) (-0.41) (4.43) (-7.16) (-5.56) (-1.99) Urban Population 0.01*** 0.02*** 0.002** -0.01*** -0.15*** -0.28*** -0.01*** -0.01** (9.55) (10.36) (2.42) (-3.34) (-12.30) (-10.53) (-4.27) (-1.99) Constant -2.81*** -1.41*** -1.32*** -1.29*** 42.71*** 35.68*** 1.22*** 1.04*** (-50.55) (-15.03) (-25.84) (-10.06) (51.12) (24.98) (12.77) (2.68) Adj 𝑅2/Within 𝑅2 0.80 0.29 0.33 0.11 0.63 0.55 0.11 0.01 Observations 3422 3422 812 812 3420 3420 3383 3383 Countries 158 158 88 88 158 158 159 159 Hausman test 229.65 142.54 489.02 11.68 (0.00) (0.00) (0.00) (0.99)
Turning point (in PPP US$)
Min 29 371 27 571
Max 63 739 73 696
Notes: Figures in parentheses are t-ratios for regression coefficient. *** statistically significant at 1 percent level; **
statistically significant at 5 percent level; * statistically significant at 10 percent level. To obtain similar magnitudes, the following original series have been scaled prior to estimation: GDP is divided by103; GDP Squared is divided by108; GDP
21
CO2 emissions
In the case of CO2, I obtain the inverted U-shaped Kuznets curve as emissions initially
rise with increasing per capita income and then decline. That is valid for both pooled OLS and FE models. Those results are consistent with the recent findings of Grunewald et al. (2017). However, it is important to note that the cubic term is highly significant and negative which indicates an N-shaped curve, meaning that emissions eventually start increasing again for very high GDP per capita levels. For comparison purposes, I also estimated the model
with only income and I found that the turning points1 are quite robust to the exclusion of the
control variables. The first turning point, situated between approximately US$27,000 and US$30,000 (according to the FE and OLS model respectively), is relatively high and roughly corresponds to the per capita income levels in Portugal, Czech Republic, and Slovenia. The estimations show that emissions decrease through an income level ranging between US$64,000 and US$74,000. Therefore, the final increasing part has a limited significance since only two countries (Singapore and Luxembourg) have such a high per capita GDP as of 2013. Whether this is a true pattern on an individual country level is a matter of further study but there is a possibility that technology has diminishing returns when it comes to pollution-reducing solutions (Dinda et al., 2000). Then, for very high-income levels the ‘scale effect’ may once again prevail over the ‘composition’ and ‘technique effects’.
The results for the control variables are consistent with my expectations. Industry
share and urban population are associated with more CO2 emissions since the coefficients
are positive and statistically significant for both OLS and FE models. Turning to population
density, it is only found to increase CO2 emissions in the pooled model but this relationship is
insignificant in my preferred specification (FE).
Other pollutants
As expected, the regression results for other pollutants lead to conclusions quite
different from those for CO2 emissions. The relationship between environmental
degradation and income goes against the EKC predictions in all three cases. However, the pooled OLS and the FE models achieve different or even opposite results regarding the direction and the strengths of the links under study. Borghesi (2000) explains that the econometric model adopted is crucial in identifying the particular relationships between pollution, income, and inequality. As mentioned, simply pooling the data disregards the heterogeneity of the countries, and therefore does not properly capture reality. So, when conflicting results are obtained, the FE model is preferred to the OLS model. Organic water
22 pollution only starts decreasing at high income levels and deteriorating again at per capita income of around US$60,000 (in PPP). The percentage of the population with no access to safe water seems to firstly increase but then decrease with income. However, a plot of the observations shows a monotonically diminishing curve with some flattening in the middle range values (the subsampling analysis in the following section confirms this). Turning to deforestation, only the squared term of per capita GDP is statistically significant, indicating that beyond a certain point, economic growth causes more environmental harm. However, it is important to note that the explanatory power of the model for deforestation measured by the adjusted R-squared is quite low.
The three control variables, population density, industry share and urban population,
are highly significant in most cases, but unlike the direction of the relationship for CO2
emissions, their increase is generally associated with better environmental quality. Since water pollution, access to safe water, and deforestation have implications mainly for the home country, the alternative explanation for facilitation of environmental improvements due to higher population density and urbanization is plausible. As far as industry share is
concerned, it could be that its deterioration effect only impacts CO2 emissions as other
environmental measures, for instance deforestation, are rather linked to the agricultural sector.
6.3. Analysis of the impact of inequality on the environment across country groups A growing amount of literature suggests that the relationship between pollution and per capita GDP cannot be understood without considering the role of inequality as a key social factor. Therefore, I introduce inequality in the model (referred to as Model 2) in order to examine its effects on pollution and on the relationship between pollution and income. As discussed, while my main inequality variable is income inequality measured by the Gini coefficient, I also include political and civil rights and literacy rates as regressors to account for power inequality. The FE and OLS specifications that I estimate for Model 2 are written respectively as: 𝑷𝑶𝑳𝒊𝒕 = 𝜷𝟎+ 𝜷𝟏𝑮𝑫𝑷𝒊𝒕+ 𝜷𝟐𝑮𝑫𝑷𝒊𝒕𝟐 + 𝜷𝟑𝑮𝑫𝑷𝒊𝒕𝟑 + 𝜷𝟒𝑮𝑰𝑵𝑰 + 𝜷𝟓𝑹𝑰𝑮𝑯𝑻𝑺𝒊𝒕 + 𝜷𝟔𝑳𝑰𝑻𝒊𝒕+ 𝜷𝟕𝑫𝑬𝑵𝑺𝒊𝒕+ 𝜷𝟖𝑰𝑵𝑫𝒊𝒕+ 𝜷𝟗𝑼𝑹𝑩𝒊𝒕 + 𝜺𝒊𝒕, (5) 𝑷𝑶𝑳𝒊𝒕 = 𝜷𝟎+ 𝜷𝟏𝑮𝑫𝑷𝒊𝒕+ 𝜷𝟐𝑮𝑫𝑷𝒊𝒕𝟐 + 𝜷𝟑𝑮𝑫𝑷𝒊𝒕𝟑 + 𝜷𝟒𝑮𝑰𝑵𝑰 + 𝜷𝟓𝑹𝑰𝑮𝑯𝑻𝑺𝒊𝒕 + 𝜷𝟔𝑳𝑰𝑻𝒊𝒕+ 𝜷𝟕𝑫𝑬𝑵𝑺𝒊𝒕+ 𝜷𝟖𝑰𝑵𝑫𝒊𝒕+ 𝜷𝟗𝑼𝑹𝑩𝒊𝒕+ 𝝁𝒊𝒕+ 𝝀𝒊𝒕+ 𝜺𝒊𝒕, (6)
23 The direction and strength of the effect of inequality on environmental degradation may differ depending on the group of countries examined. In this study, emerging economies (EEs) are of particular interest because of their increasing role in the world economy and in terms of global pollution levels. The Chow test reported in Tables 5 and 6 shows that there are differences in the estimated parameters for developed, emerging, and developing countries. One way to account for those differential impacts is the methodology applied by Torras and Boyce (1998). In their study, they create dummy variables for high- and low-income countries and interact those with the inequality variables so that separate coefficients are estimated for all inequality measures and for each group of countries. While this model is more compact, it is not fully interacted and has as a major drawback the assumption that all other coefficients are the same for the different groups. I argue that this is not necessarily a good description of reality as it assumes that the effects of GDP growth on pollution are stable across country groups. An alternative approach is adopted by Borghesi (2000) and Clement and Meunie (2010) who perform a subsample analysis that allows for all coefficients to differ across different groups of countries. Therefore, to verify whether inequality has a distinct effect on environmental quality in emerging economies, I repeat the analysis of Model 2 for the sets of developed, emerging, and developing countries. This results in more homogeneous groups of countries owing to less variation in the levels of income inequality and per capita GDP within each group. Even though the analysis suffers from a loss of degrees of freedom, the number of observations per each sample is still sufficient, since the p-value for the F-test of overall significance has a value of zero for all subsamples. Further, the explanatory power of the subsamples as measured by adjusted R-squared are generally higher than the value for the entire sample. The estimations of the OLS and FE specification for Model 2, including the entire sample and the subsample analysis, are presented respectively in Tables 5 and 6.
CO2 emissions
Including inequality increases the explanatory power of the model for all pollutants
and all specifications. The nature of the relationship between CO2 and GDP per capita is
24
Table 5. Regression for the determinants of environmental quality including inequality and subsample analysis (Model 2), pooled OLS estimation
Log CO2 emissions
Log Organic water pollution (BOD)
No access to safe water Deforestation Variable Whole sample Developed countries Emerging countries Developing countries Whole sample Developed countries Emerging countries Developing countries Whole sample Developed countries Emerging countries Developing countries Whole sample Developed countries Emerging countries Developing countries GDP (per capita) 0.21*** 0.12*** 0.14** 0.72*** -0.03*** -0.05*** -0.11*** 0.09** -2.21*** -0.44*** -2.63*** -5.26*** -0.04*** -0.06*** 0.52*** 0.03 (31.72) (9.15) (2.45) (18.01) (-5.64) (-8.44) (-2.99) (2.30) (-20.57) (-4.48) (-4.92) (-9.75) (-3.34) (-3.71) (6.82) (0.40) GDP Squared -0.49*** -0.25*** 0.09 -5.51*** 0.07*** 0.12*** 0.66** -0.71* 5.86*** 0.88*** 12.55*** 29.76*** 0.12*** 0.15*** -4.20*** -1.57* (-25.67) (-8.39) (0.20) (-11.79) (5.14) (7.36) (2.31) (-1.89) (20.07) (4.30) (3.03) (6.06) (3.72) (3.86) (-7.31) (-1.72) GDP Cubed 0.35*** 0.17*** -1.35 13.98*** -0.05*** -0.08*** -1.34* 1.52 -4.16*** -0.52*** -22.52** -57.27*** -0.08*** -0.09*** 9.59*** 6.26** (21.27) (8.39) (-1.17) (8.95) (-4.63) (-6.74) (-1.96) (1.35) (-17.64) (-4.04) (-2.30) (-4.37) (-3.34) (-3.61) (7.17) (2.20) Gini -0.01*** 0.02*** -0.01** -0.02*** -0.01*** 0.003 0.002* 0.003 -0.22*** 0.12*** 0.16*** -0.00 0.001 -0.01 -0.03*** 0.01* (-4.23) (4.39) (-2.26) (-7.87) (-15.82) (1.28) (1.96) (1.58) (-8.38) (11.53) (4.29) (-0.02) (0.81) (-1.29) (-7.15) (1.95) Political Rights -0.02*** -0.05*** -0.02* -0.03*** -0.00*** 0.05*** 0.02*** 0.03*** 0.01 0.32*** -0.12 0.01 -0.01*** -0.00 -0.02 0.03* (-3.95) (-3.64) (-1.74) (-4.19) (-2.79) (3.95) (4.68) (4.09) (0.29) (4.56) (-1.37) (0.10) (-2.70) (-0.09) (-1.46) (1.81) Literacy Rates 0.02*** 0.02*** 0.02*** 0.02*** 0.001* -0.002 -0.003 0.00 -0.11*** -0.06*** -0.06** -0.16*** -0.01*** 0.01 -0.06*** -0.004* (19.76) (4.85) (6.76) (14.08) (1.65) (-0.51) (-1.32) (0.30) (-8.33) (-3.43) (-2.34) (-8.02) (-4.16) (0.90) (-13.61) (-1.65) Population Density 0.003*** -0.01*** -0.13*** -0.04*** 0.01*** -0.01*** -0.06*** -0.10*** 0.03 -0.01 -0.09 -2.67*** 0.01*** 0.01*** -0.04** -0.06 (2.62) (-6.32) (-9.15) (-3.25) (4.96) (-3.64) (-3.21) (-6.76) (1.09) (-1.03) (-0.59) (-7.62) (3.18) (5.11) (-2.14) (-1.62) Industry share 0.03*** 0.02*** 0.01** 0.02*** 0.01*** -0.01*** -0.00 -0.01** -0.22** 0.07*** 0.27*** 0.05 0.001 -0.02*** 0.01** 0.00 (19.42) (9.36) (2.44) (8.47) (2.75) (-2.75) (-0.79) (-2.42) (-2.37) (17.67) (9.18) (1.23) (0.13) (-3.77) (2.27) (0.70) Urban population 0.01*** 0.01*** -0.01*** 0.01*** -0.001 0.01*** 0.01*** -0.01*** -0.23*** -0.00 -0.04** -0.21*** -0.01*** -0.02*** 0.03*** -0.02*** (7.62) (8.96) (-4.87) (3.17) (-0.63) (5.76) (6.16) (-4.74) (-12.91) (-0.58) (-2.47) (-8.89) (-3.58) (-3.27) (7.77) (-5.77) Constant -3.53*** -2.66*** -0.63*** -3.42*** -1.65*** -1.92*** -1.55*** -1.56*** 55.48*** 4.22* 18.16*** 63.21*** 1.22*** 1.19 2.58*** 0.93*** (-33.00) (-6.08) (-3.18) (-28.72) (-19.39) (-3.51) (-6.26) (-15.06) (29.76) (1.68) (7.02) (27.07) (6.39) (0.94) (7.44) (3.40) Adj 𝑅2/Within 𝑅2 0.86 0.37 0.74 0.85 0.36 0.50 0.59 0.21 0.70 0.49 0.67 0.65 0.13 0.15 0.36 0.15 Observations 2439 666 512 1261 743 340 159 244 2445 666 513 1266 2416 673 496 1247 Countries 142 34 23 85 82 28 18 36 143 34 23 86 144 35 23 86 Chow test 60.91 12.74 68.58 23.07 (0.00) (0.00) (0.00) (0.00)
Notes: Figures in parentheses are t-ratios for regression coefficient. *** statistically significant at 1 percent level; ** statistically significant at 5 percent level; * statistically significant at 10 percent level. To obtain similar
25
Table 6. Regression for the determinants of environmental quality including inequality and subsample analysis (Model 2), fixed effects estimation
Log CO2 emissions Log Organic water pollution (BOD) No access to safe water Deforestation
Variable Whole sample Developed countries Emerging countries Developing countries Whole sample Developed countries Emerging countries Developing countries Whole sample Developed countries Emerging countries Developing countries Whole sample Developed countries Emerging countries Developing countries GDP (per capita) 0.08*** 0.05*** 0.18*** 0.35*** 0.002 0.00 -0.01 0.02 0.22*** -0.30*** -1.48*** -2.28*** -0.02 -0.01 0.27*** 0.10 (17.07) (8.19) (9.66) (16.44) (0.37) (0.05) (-0.26) (0.56) (2.74) (-13.86) (-4.27) (-6.59) (-0.83) (-0.43) (2.64) (1.10) GDP Squared -0.18*** -0.08*** -0.82*** -2.31*** -0.03*** -0.02 0.03 -0.11 1.24*** 0.55*** 5.76** 33.23*** 0.10** 0.12* -1.53** -0.75 (-14.78) (-6.36) (-6.21) (-13.16) (-2.72) (-1.54) (0.13) (-0.50) (6.12) (11.30) (2.41) (11.69) (2.04) (1.88) (-2.22) (-0.98) GDP Cubed 0.12*** 0.05*** 1.33*** 5.22*** 0.03*** 0.03** -0.20 0.32 -1.01*** -0.31*** -8.60 -89.22*** -0.08* -0.10** 2.65* 1.77 (11.60) (5.78) (4.34) (11.18) (3.56) (2.39) (-0.36) (0.59) (-6.00) (-9.06) (-1.55) (-11.79) (-1.87) (-2.30) (1.67) (0.87) Gini -0.00 -0.00 -0.02*** 0.00 0.003 -0.01*** 0.01** -0.01*** -0.49*** 0.03*** -0.24*** 0.18*** -0.05** -0.04*** -0.02** -0.02** (-0.97) (-0.36) (-8.59) (0.13) (0.91) (-3.94) (2.49) (-4.35) (-6.01) (2.85) (-6.46) (6.16) (-2.46) (-2.71) (-2.15) (-2.11) Political Rights 0.01*** -0.03*** 0.00 0.01*** -0.00 0.004 0.001 -0.02*** -0.11*** 0.06 -0.37*** -0.30*** -0.01*** -0.15*** -0.07*** -0.04** (2.65) (-2.76) (0.32) (3.11) (-1.57) (0.39) (0.19) (-3.65) (-7.06) (1.53) (-6.88) (-4.10) (-3.46) (-2.79) (-4.39) (-1.99) Literacy Rates 0.01*** 0.01 0.01*** 0.01*** 0.002 0.02 -0.01** -0.00 -0.18*** -0.21*** -0.33*** -0.03 -0.03*** 0.18*** 0.00 0.00 (9.93) (1.19) (3.94) (5.24) (0.79) (1.36) (-2.16) (-0.77) (-6.39) (-6.80) (-6.42) (-1.06) (-4.16) (4.58) (0.04) (0.01) Population Density -0.01*** -0.03*** 0.20*** -0.07 -0.01*** -0.00 -0.04 -0.20* 0.09*** 0.01 -0.27 -0.29 -0.01*** -0.08*** -0.13 -0.36* (-2.82) (-6.63) (7.95) (-1.59) (-5.13) (-0.78) (-0.40) (-1.87) (4.18) (0.74) (-0.59) (-0.41) (-2.62) (-3.45) (-0.95) (-1.90) Industry share 0.01*** 0.01*** 0.01*** 0.00* -0.01* -0.01*** -0.00 -0.00 -0.40*** -0.03*** -0.08*** -0.08*** -0.05*** 0.03*** 0.01 -0.03*** (7.79) (4.95) (3.59) (1.78) (-1.75) (-3.75) (-1.49) (-1.22) (-7.93) (-3.20) (-3.07) (-3.71) (-3.97) (3.03) (0.86) (-4.67) Urban population 0.02*** 0.01*** 0.03*** 0.02*** -0.01** 0.01* -0.00 0.01*** -0.13*** -0.04*** -0.01 -0.16*** 0.01 -0.04*** 0.00 -0.05*** (12.72) (3.66) (12.09) (6.99) (-2.29) (1.88) (-0.38) (2.61) (-6.37) (-4.02) (-0.27) (-3.62) (1.01) (-3.00) (0.07) (-3.85) Constant -2.35*** -0.17 -2.22*** -2.89*** -1.23*** -3.41*** -0.37 -1.51*** 38.71*** 27.48*** 67.09*** 40.11*** 2.31*** -13.50*** 0.09 3.42*** (-17.86) (-0.20) (-10.75) (-17.52) (-5.91) (-2.75) (-0.64) (-5.92) (17.44) (8.52) (18.01) (14.93) (4.07) (-3.27) (0.08) (4.60) Adj 𝑅2/Within 𝑅2 0.40 0.43 0.79 0.44 0.13 0.22 0.45 0.41 0.61 0.41 0.83 0.71 0.05 0.23 0.10 0.08 Observations 2439 666 512 1261 743 340 159 244 2445 666 513 1266 2416 673 496 1247 Countries 142 34 23 85 82 28 18 36 143 34 23 86 144 35 23 86 Hausman test 153.25 72.19 214.08 20.15 (0.00) (0.00) (0.00) (0.93) Chow test 26.58 7.27 34.93 3.49 (0.00) (0.00) (0.00) (0.00)
Notes: Figures in parentheses are t-ratios for regression coefficient. *** statistically significant at 1 percent level; ** statistically significant at 5 percent level; * statistically significant at 10 percent level. To obtain similar
26 of developed, emerging, and developing countries, the EKC is confirmed in all three cases but the coefficients differ. That is, the initial increase in per capita income leads to higher emissions in all subsamples but the effects in emerging and developing countries are much stronger than in developed ones.
The impact of the inequality variables on per capita CO2 emissions heavily depends on
the specific group of countries under consideration. This is an important alert for aggregation problems in previous studies that draw general conclusions based on their results. For the entire sample, the coefficient of income inequality is negative but only significant for the OLS model which is consistent with the findings of Borghesi (2000). The coefficients for the subsamples of developed and developing countries are significant and respectively positive and negative for the OLS regression results but insignificant for the FE specification. However, as far as emerging economies are concerned, an increase in income
disparity is found to decrease CO2 emissions for both OLS and FE models, which goes against
Boyce’s main hypothesis.
The two measures of inequality of power have a statistically significant effect at 1
percent level on CO2 emissions for both the pooled OLS and the FE models for the whole
sample. Therefore, their inclusion is crucial to understanding the effects of social inequality on environmental degradation. A higher index of political rights and civil liberties leads to fewer emissions for all samples according to the OLS model. However, according to the FE model, this only holds for developed countries, while an increase in political rights is associated with higher emissions in the entire sample and in developing countries and has no significant effect in emerging economies. The effect of literacy rates on emissions is found to be more stable across country groups and different specifications. Generally, an increase in literacy rates has an upward influence on emissions, except for developed countries. One possible explanation would be that the average value of literacy rates in developed countries has very little variation since the mean value is already almost 99 percent for the time period under study.
Other pollutants
Comparable to the CO2 emissions results, the relationship between per capita GDP and
