Abstract
In this applied study, we use US county data to examine patterns in fine particulates (also called fine particulate matter or PM2.5) ambient concentrations as a measure of air pollution within the framework of the environmental Kuznets’ curve (EKC). We pay particular attention to the role of social capital and notions of ruralness. Consistent with expectations, we find that peak of the EKC ranges between US$24,000 and US$25,500 for PM2.5 concentrations depending on the estimator used. Also consistent with expectations, higher levels of social capital places downward pressure on PM2.5 concentrations, but that effect is weaker in more rural areas. The implication is that the promotion of economic growth may harm the environment at lower levels of income but will improve the environment as income continues to grow.
Keywords
Introduction
The trade-off between economic well-being and environmental quality has long been the focal point of academic and policy discussions. Rising populations and levels of economic well-being evoke concern about the state of our environment and the sustainable of growth. The notion that economic growth may be linked to environmental improvements is a considerable shift from the assumption that we must choose between growth and protecting the environment. A growing body of empirical evidence (e.g., Holtz-Eakin and Selden 1992; Selden and Song 1994; Lopez 1993; Berrens et al. 1997; Suri and Chapman 1998; List and Gallet 1999; Pfaff, Chaudhuri, and Nye 2004; Rupasingha et al. 2004; Paudel, Zapata, and Susanto 2005; Paudel and Schafer 2009; Paudel et al. 2011; Dietz, Rosa, and York 2012) suggests that the relationship between environmental quality and economic growth and well-being is more complex than the traditional view of pollution as an inevitable and continuous by-product of economic growth.
In their early work on the potential environmental impacts of the North American Free Trade Agreement, Grossman and Krueger (1991, 1995, 1996) discovered an inverted-U–shaped relationship between gross domestic product (GDP) and two measures of urban air pollution among a group of developed and developing countries. Grossman and Krueger’s results suggest a possible economic path in which initial growth is linked to increasing levels of pollution, and subsequent growth is linked to decreasing levels of pollution. Grossman and Krueger (1991, 35) concluded that “…economic growth tends to alleviate pollution problems once a country’s per capita income reaches [an estimated income threshold].” They referred to this inverted-U relationship as the environmental Kuznets’ curve (EKC), after the original income-inequality Kuznets curve of similar form (Kuznets 1955; Williamson 1965) and paved the way for a lengthy debate over the theoretical and empirical robustness of the income and environmental quality relationship (Arrow et al. 1995; Stern 1998, 2004a, 2004b; Dasgupta et al. 2002; Dinda 2004).
Hypotheses about why or how this inverted-U relationship exists mostly fall into two camps: (1) at some level of income firms adapt to market or regulation changes by employing abatement technologies or production efficiencies, known as a technique effect and (2) at some level of income residents begin to object to the presence of polluting firms, known as a composition effect. The first scenario reflects changes in both consumption and production patterns and results in an overall reduction in pollution due to the lower pollution levels per unit of production (Selden and Song 1994; Pfaff, Chaudhuri, and Nye 2004). Some explanations incorporate internalization of pollution externalities (Jones and Manuelli 1995) or increasing returns to scales with regard to pollution abatement technologies (Andreoni and Levinson 2000). The second scenario reflects residential demand for higher environmental quality. This can take the form of “not in my backyard” or NIMBY where pollution levels drop locally, but overall levels remain constant or potentially increase, as pollution-generating processes are pushed to lower-income areas (Arrow et al. 1995; Suri and Chapman 1998). Despite growing interest in the relationship between income and pollution relationship, the EKC has been extensively criticized (Stern 1998, 2004b; Dasgupta et al. 2002), particularly with regard to the disconnect between the limited theoretical work and the more extensive empirical analyses.
Short of closing the gap between theory and empirics, significant strides have been made in how researchers analyze the income–pollution relationship. A wide variety of environmental degradation measures have been used to represent environmental degradation—air pollution, water pollution, hazardous waste, toxic releases, among others. Additionally, the range of socioeconomic and demographic explanatory variables has grown, increasing model accuracy by removing omitted variable bias and shedding light on factors affecting the shape and height of EKCs. For example, Berrens et al. (1997) explore how racial diversity, educational attainment, household size, homeownership, and the proportion of rural population affect the relationship between income and hazardous waste among US counties. Rupasingha et al. (2004) explicitly argue that a similar portfolio of socioeconomic variables should be considered in EKC estimation to mitigate as much omitted variable bias as possible.
Of particular interest is the work by Paudel et al. (2011), Grafton and Knowles (2004), and Rupasingha et al. (2004), who suggest that local networks, social norms, levels of trust, and the ability to cooperate and undertake collective action, traits that are collectively termed social capital, are vital to the explanation of local and regional pollution levels. Communities with high levels of civic engagement and a well-defined network between community members, businesses, and policy makers are more likely to have widespread buy-in from community members on decisions about pollution issues. Furthermore, engagement between community members has an impact on individual behavior—social norms. If I am better connected to my neighbors, local business owners, or city officials, I am likely to care more about overall community performance and my role, or perceived role, in that performance.
Canonical works of the latter, such as Hardin’s (1968) Tragedy of the Commons and Ostrom’s (1990) Governing the Commons: The Evolution of Institutions for Collective Action, have led to a more nuanced analysis of social capital and environmental degradation (or its inverse, environmental quality; Pennington and Rydin 2000; Rydin and Pennington 2000; Pretty and Ward 2001; Grafton and Knowles 2004). In a qualitative study on the links between social capital and environmental quality, Pretty and Ward (2001, 112) ask, “[t]o what extent are social and human capital prerequisites for long-term improvements in natural capital?” In this context, natural capital is synonymous with environmental quality. They examine the role of nongovernment groups in the management of natural resource stocks, and focus on group problem solving and enforcement as the mechanisms by which social capital increases environmental quality.
This study builds on the work of Paudel et al. (2011, 1) who argue that “a country with stronger civic solidarity may more aggressively implement pollution control measures due to an overall, collective vigilant approach to pollution-emitting industries.” Using data for US counties, we model the relationship between income and environmental quality measured by fine particulate (also called fine particulate matter or PM2.5) ambient concentrations as a indicator of air pollution within the framework of the EKC. We explicitly model how social capital influences the EKC relationship, hypothesizing that higher levels of social capital put downward pressure on the EKC and shift the income threshold defining the peak of the EKC to a lower level.
The study is composed of four sections beyond these introductory comments. Next, we review the two deductive stylized theories that have been suggested as foundation for what has traditionally been an inductive empirical theory. We the present our empirical model after summarizing some of the empirical literature. Here we outline the basic empirical model, our metric of social capital along with a discussion of our estimation methods including a spatial Durbin estimator that explicitly allows for us to model spatial spillovers. Our results are then presented, followed by a summary of the key findings of the study and areas of future work.
A Simple Theoretical Framework
One of the central criticisms of environmental income trade-off captured in the EKC is that it is an inductive theory based on empirical observation and is not deduced from a stylized theoretical framework. But there have been a handful of attempts to provide a more solid theoretical foundation. As noted by Li, Grijalva, and Berrens (2007), Dinda (2004), and Stern (2004a, 2004b), there have been several theoretical justifications for the observed empirical relationship such as scale, technological and composition of production effects (e.g., Grossman and Krueger 1991), international trade, globalization and the displacement hypothesis (e.g., Copeland and Taylor 1995; Ekins 1997; Wheeler 2000), direct foreign investment and diffusion of technology (e.g., Reppelin-Hill 1999), and the pollution haven hypothesis (e.g., Antweiler, Copeland, and Taylor 2001; Liddle 2001) or race to the bottom hypothesis (Mani and Wheeler 1998). Most of these, however, only make sense within the international development literature and on their face value has limited application to more developed economies such as the United States.
We suggest that the most likely driving factor is changes in consumer demand for environmental quality as income grows as suggested by Beckerman (1992), Carson, Jeon, and McCubbin (1997), McConnell (1997), and Chaudhuri and Pfaff (1998). At low-income levels, consumers are willing to make the trade-off of degraded environmental quality in the name of income growth. But at some level of income, the peak of the EKC, tastes and preferences change, and the demand for improved environmental quality grows. Andreoni and Levinson (2000, 264) assert that “high income individuals demand more consumption and less pollution…when abatement is possible with increasing returns, high-income [firms] can more easily achieve both goals.” In other words, as consumers demand higher quality environmental conditions, firms in richer countries are better position to adapt to these changing demands. Unless consumers demand shifts and markets are created, there is little reason for firms to shift production processes.
Pfaff, Chaudhuri, and Nye (2004) articulate a framework in which pollution is a by-product of household consumption and the income–pollution relationship results from a shift in the composition of consumption as household income increases. Consumption decisions about the amount of goods purchased and the type of goods purchased determine marginal pollution levels. Higher-income households, they argue, indirectly “buy” increased environmental amenities by substituting more expensive, cleaner goods for cheaper, dirtier goods. The single household, two-good model presented by Pfaff et al. specifies utility maximization via consumption of two nonmarket goods—non-environmental services and environmental amenities. These nonmarket goods are derived from the actual consumption of marketed commodities, which are categorized into dirty goods qd
and clean goods. In this context, environmental amenity is the difference between a theoretical pristine environment and pollution
The household decision rule is to maximize utility through consumption of market goods, given a budget constraint. This simple model illustrates how a household can demonstrate its relative value of environmental amenities by choosing (1) the amount of services it consumes and (2) how those services are produced. For example, consider a household that currently purchases 90 percent of its monthly electricity needs from the coal-dependent local utility and obtains the remainder from rooftop solar panels. If the household’s electricity use increases and it does not install additional solar panels, its consumption of electricity will further degrade the environment (scale effect). Alternatively, if the household’s electricity use remains constant, but it installs an additional solar panel, it will require a lesser amount of coal-based electricity and environmental amenities, the marginal impact to the initial endowment, will improve (technique effect).
Given these two decision points, scale and technique, the utility function can be expressed in terms of non-environmental services, environmental amenities, and the fraction of goods that are clean goods. It is plausible, but not necessary, that a demand for services would be greater than a demand for environmental amenities at lower levels of income because the initial endowment of amenities has not yet been degraded by high levels of consumption (such as the endowment for higher-income levels), so there is less concern for conservation. Additionally, marginal price represents a larger share of the household budget for lower-income households, suggesting price sensitivity would be greater at the lower end of the income continuum. It is also plausible, but not necessary, that the opposite is true for higher-income households in both instances. Such conditions would result in a U-shaped income–amenities relationship or an inverted-U income–pollution relationship.
The key point made by Pfaff, Chaudhuri, and Nye (2004, 191) is that “the ability to substitute between [types of] marketed goods [those with dirty production process and those with clean], allows a separation of two decisions: how much service to consume, and how to produce that service. The fact that these two decisions may move independently with respect to income allows for their combined effect to be non-monotonic.” It then becomes an empirical question: do higher-income households spend a greater portion of their overall expenditures on clean goods compared to lower-income households? If this is indeed the case, Pfaff, Chaudhuri, and Nye show that a high-income household that consumes more than a low-income household may contribute less to overall pollution levels because it is buying services with relatively clean production processes.
When using a Cobb–Douglas specification for household preferences, Pfaff, Chaudhuri, and Nye (2004, 191) found that, “the greater the weight on services and the greater the [initial amenities] endowment, the more the household will consume the dirty good for services while degrading the endowment before substituting to more expensive by less degrading goods.” In other words, low-income households in a rural setting that has not yet experienced significant pollution prioritize consumption over maintaining a clean environment. In an effort to buy goods and services as cheaply as possible, low-income households tend to purchase less expensive goods, which Pfaff et al.’s model assumes to carry with them dirtier production processes.
This theoretical explanation for the potential non-monotonicity of the income–pollution relationship within the context of the household utility function is intuitive, but finding the necessary household consumption data for empirical testing would be a challenge. Comparing purchases of “green” products to traditional products would be the obvious strategy. There are, however, a few criticisms of the stylized model presented by Pfaff, Chaudhuri, and Nye (2004) worth noting. One, they treat environmental amenities as a normal good, while other explanations consider it to be a luxury good. Two, Pfaff et al. ignore the role of geography and localized amenities (or localized pollution), often reflected by the NIMBY mentality. Suri and Chapman (1998), among others, make the argument that rather than pay more for cleanly produced goods, high-income areas shift dirty production processes to poorer areas and can then import the manufactured goods without the production-associated pollution. Finally, the framework put forth by Pfaff et al. hinges on consumers’ awareness of the environmental footprint of various production processes. It seems relatively rare that the average consumer takes the initiative to collect this information; however, even without such information, it is likely that a combination of price signals and social pressure (think Prius vs. Jeep) serve as a reasonable proxy.
Beyond the inductive nature of the preponderance of the literature, there are several theoretical criticisms of the EKC model. As argued by Stern (2004b), a key criticism that has been advanced in the literature is that there is no feedback from environmental damage and economic performance (i.e., income). Because income is assumed to be exogenous it is not possible for changes in the environment to impact income. For rural US counties, environmental quality has been demonstrated to be a major determent of growth (e.g., Deller et al. 2001). Another fundamental criticism is that the available theories fail to predict the “tipping point.” The theories provide foundations for one or the other side of the EKC but fail to adequately predict the level of income defining the peak of the curve. But most of the criticisms are structured within a trade framework such as the Heckscher–Ohlin trade theory. Given that the literature tends to be dominated by studies that use multicountry data, as opposed to within country data, this focus of critiques makes sense. But given our interest in within US patterns, it is not clear if most of these trade-based critiques apply.
Empirical Framework 1
One of the most consistent critiques of the EKC is that it lacks a solid theoretical foundation but is built on a large and growing empirical literature. This empirical literature expands upon Grossman and Krueger’s initial research in four key ways: (1) use of a variety of pollution measures to test sensitivity of results, (2) use of a variety of study areas again to test sensitivity of results and indirectly explore the role of institutions, (3) inclusion of additional explanatory variables (Grossman and Krueger account only for population density and a handful of dummy variables accounting for variation among data collection methods across countries), and (4) adoption of more sophisticated econometric techniques. Given the reduced-form nature of empirical EKC models, quality data and accurate estimation methods are vital.
The early EKC studies tended to err on the side of preventing simultaneity bias and generally had a sparse list of control variables. Holtz-Eakin and Selden (1992, 5) states, “[p]erhaps the most fundamental (aspect of the reduced-form specification) is the exclusion of explanatory variables other than per capita GDP.” At the risk of introducing factors that are endogenous to income, Holtz-Eakin and Selden (1992) argue for the use of two-way fixed effects estimation methods to account for unit-level correlation and time trends. Selden and Song (1994) and List and Gallet (1999) are also conservative in their inclusion of additional explanatory variables, with the exception of population density, which limits the information about underlying mechanisms of the EKC relationship that can be extracted.
Suri and Chapman (1998) estimate seven models, one with no additional explanatory variables and six with varying measures of international trade, concluding that the turning point of the EKC curve is substantially raised with the introduction of trade variables. They interpret their findings as evidence that the exportation of dirty manufacturing processes from rich countries to poor countries is a significant part of the story underlying the EKC concept. Given the inductive nature of the EKC theory, empirical tests often suffer from omitted variable bias. Suri and Chapman’s dramatic shift in income turning points illustrate the need for caution when interpreting empirical results from an isolated EKC study.
Berrens et al. (1997) claim to include “a wide variety of independent variables” but are limited to only metrics of racial composition in their analysis. They include percentage of whites and percentage of whites squared and found an inverted-U relationship between racial composition and hazardous waste, similar to that for income and hazardous waste. Rupasingha et al. (2004) include a number of socioeconomic variables: population density, educational attainment, ethnicity, inequality, urban/rural dummy variables, and size of the manufacturing sector (in terms of employment) in their analysis of the relationship between income and toxic releases among US counties. They found evidence of an EKC with per capita income and toxic releases. Furthermore, their research reveals a negative correlation between educational attainment and pollution, a positive correlation between ethnic fragmentation and pollution and a tendency for higher pollution levels in urban areas and lower levels in rural areas. Their key finding, however, is the “existence of an inverse Kuznets-type relationship between income inequality and environment conditions, holding constant the level of income,” and posit that in communities with extreme levels of inequality, “those at the top of the distribution start to look out for the interests of the entire community, rather than only their own interests” (Rupasingha et al. 2004, 421).
Begun and Eicher (2008) suggest that the lack of a solid theoretical foundation upon which to base the selection of control variables, coupled with the vast heterogeneity in the set of control variables used within the EKC literature and the effects associated with those control variables, results in a high degree of model uncertainty. Indeed, Begun and Eicher (2008, 796) argue that “EKC is thus a case study of extreme model uncertainty where the true model is unknown and several competing approaches exist that hypothesize about the exact relationship between environmental quality and income.” Raftery (1995) and Raftery, Madigan, and Hoeting (1997) and others (e.g., Durlauf and Quah 1999; Brock and Durlauf 2001; Brock, Durlauf, and West 2007) argue that in the presence of model uncertainty inferences based on a single regression model overstates the precision of coefficient estimates. One approach is to use meta-analysis, such as those done for the EKC literature by the Cavlovic et al. (2000), Li et al. (2007), and Goldman (2012), to draw stronger inferences out of the larger empirical literature. Another approach, as suggested by Begun and Eicher (2008), is to introduce alternative methods that base inferences on all competing models. Within a Bayesian setting, one can compare these competing models by weighting each by the posterior probability that the model is indeed the true model. This method, referred to Bayesian model averaging, is gaining wider acceptance as a means to determine the relevant set of control variables. In a study of sulfur dioxide (SO2) concentrations across countries and time, Begun and Eicher (2008) find that only a fraction (about one-third) of seventeen possible control variables are relevant to the EKC. In addition, they find only weak evidence of the classic inverted-U associated with the EKC.
One area of research within the broader environmental economics literature explores the linkages between civic engagement, collective action, and management of environmental resources (Rydin and Pennington 2000; Pennington and Rydin 2000; Pretty and Ward 2001; Grafton and Knowles 2004; Paudel et al. 2011). The examination of the interplay between notions of social capital and environmental resources is advancing within the EKC framework. In a qualitative study on the links between social capital and environmental quality, Pretty and Ward (2001, 112) ask, “[t]o what extent are social and human capital prerequisites for long-term improvements in natural capital?” In this context, natural capital is synonymous with environmental quality. They examine the role of nongovernment groups in the management of natural resource stocks and focus on group problem solving and enforcement as the mechanisms by which social capital increases environmental quality.
Paudel and his colleagues (2011, 1) maintain that “a country with stronger civic solidarity may more aggressively implement pollution control measures due to an overall, collective vigilant approach to pollution-emitting industries.” Social capital, in this instance, is measured using an index of responses from the World Values Survey, a common source for country-level data pertaining to social capital (Grafton and Knowles 2004; Paudel et al. 2011). The selected survey questions focus on trust and voluntary involvement in five types of associations: trade unions, political parties, environmental groups, youth-focused groups, and professional organizations.
Paudel et al. (2011) find mixed results with regard to income effects and the impact of social capital. The significance associated with the income variables vary considerably among pollution measures and among model specifications for a single measure. Even in the instances where there is a significant relationship between income and pollution, the peak of the EKC exhibits nonsensical results like negative or complex income values. Paudel et al. (2011, 10) conjecture that highly agricultural societies may display both high levels of social capital and high levels of particulate matter, but conclude that “[t]he results of this study do not support the hypothesis that social capital as measured here is a significant factor in reducing pollution. It may be that a better proxy for social capital or more spatially specific data would lead to different conclusions.”
Basic Empirical Model
The empirical model used in this cross-sectional analysis starts with a simple equation that is linear in the parameters and includes a squared income term, allowing for an inverted-U shaped function:
where Pi is represents a specific pollutant for the ith county, yi is per capita income, SCi is the social capital index, and RuralSCi reflects potential variation in social capital impacts depending on the degree of ruralness in the county. Zi is a vector of control variables accounting for socioeconomic characteristics of the county, as well as composition of the local economy, and Ds is a vector of state dummy variables that reflect unobserved state-level policies that encourage or discourage certain industries significantly impact on pollution levels. Evidence of an EKC-type relationship would be given by income parameters that are significantly different from zero, where β1 is positive and β2 is negative.
Pollution comes in many forms, varying in geographical scope and impact to ecosystems and communities. Some impacts are mostly local (i.e., they do not travel very far), such as forest clear-cutting, lake contamination, or litter. Most pollution-generating processes, however, have far-reaching effects into other areas of local ecosystems or into other geographical regions. Air emissions are an obvious example. Pollutants released into the air travel with the wind, decreasing breathability and visibility, causing acid rain, and in the case of CO2, enhance greenhouse effects worldwide. Ultimately, the major determinant of which pollution measure(s) is(are) used in EKC studies rests on the availability of quantitative data that is comparable across observations. This is a significant constraint, even within highly developed areas with significant funding for data monitoring and collection. For this study, we focus on the concentrations of fine particulates (also called fine particulate matter or PM2.5).
Fine particulate matter, defined as particles with an aerodynamic diameter of 2.5 microns or less, are emitted directly from combustion sources and formed secondarily from concentrations of SO2, nitrogen oxides, or organic compounds. Processes reliant on the combustion of coal, oil, diesel, gasoline, or wood, such as manufacturing, electricity production, and transportation, as well as high-temperature processes such as smelting and steel production, are the principle sources of fine particulates. Human exposure to particulate matter is linked to a range of serious health issues. According to the Environmental Protection Agency (EPA 2011), the effects of prolonged exposure include “premature mortality, aggravation of respiratory and cardiovascular disease (as indicated by increased hospital admissions and emergency room visits, school absences, work loss days, and restricted activity days), aggravated asthma, acute respiratory symptoms, chronic bronchitis, decreased lung function…[and] may result in tens of thousands of excess deaths per year, and many more cases of illness among the U.S. population”.
Concentrations data are from the Centers for Disease Control and Prevention (CDC) and reflect the estimated annual average ambient concentrations of PM2.5 in micrograms per cubic meter. CDC data on concentrations are used as an alternative to “raw” EPA data because the EPA monitors PM2.5 in only 20 percent of US counties. The EPA has 4,000 monitoring stations nationwide; however, most stations are located in urban counties. Concentrations are modeled for the remainder of counties by the CDC, in conjunction with the EPA, using a hierarchical Bayesian statistical model. The impetus for the modeling effort is to make “predictions available for environmental public health tracking purposes in areas of the country that do have monitors and to fill in the time gaps when monitors may not be recording data” (CDC 2011). Given our focus on rural areas, the estimated concentrations present a limitation to the study: we are looking for patterns in statistically derived data. Unfortunately, ecological studies of rural areas within the United States are increasingly dependent on statistically derived data. For example, the movement of the US Census Bureau to the American Community Survey has fundamentally altered the nature of the data researchers rely on for their work. The EKC, if it is present in rural areas, is based on people’s perceptions of pollution or environmental quality and these perceptions are based on the best available data. For ambient concentrations of particulate matter, the CDC data are the best available and would be the basis for perceptions and policy actions. 2
A wide range of socioeconomic variables are included in the empirical model. As noted above, one of the weaknesses of many empirical EKC studies is the scarcity of variables, beyond income, that help explain the variation in pollution levels. Omission of key variables, however, can lead to bias and inconsistency in parameter estimates. Expanding on the data set proposed by Rupasingha et al. (2004), we include, in addition to per capita income, variables reflecting the following hypothesized factors of pollution (a table of descriptive statistics is provided in Table 1):
Descriptive Statistics (n = 2,995).
Note: M = mean; PM2.5 = fine particulate matter; SD = standard deviation.
Social capital;
Ruralness;
Age;
Education;
Racial fragmentation;
Poverty rate;
Income inequality;
Commute time;
Homeownership;
Change in population;
Change in income;
Change in unemployment;
Size of farming sector;
Size of manufacturing sector.
Traditionally, within the EKC literature, income is measured using per capita income. But Stern (2004b) argues that median income measures are more appropriate because per capita figures are right skewed toward a minority of high earners, with more people falling below the mean than above it. This critique, however, is more applicable to the cross-country studies that make up the bulk of the EKC literature, where the difference between per capita and median income can be quite large. Among US counties, however, the two measures are highly correlated (rpci,medinc = 0.8472 with p = .0001), and we believe within the US context per capita income is the appropriate metric of income. 3
To capture the concern of Stern (2004a, 2004b), we follow the lead of Rupasingha et al. (2004) and include the gini coefficient of income inequality; a ratio of the cumulative share of income earned and the cumulative share of earners, ordered from lowest to highest (Burkey 2006). Perfect equality is represented by a value of zero and perfect inequality is represented by one. In addition to capturing the limitations of using per capita income observed by Stern (2004a, 2004b), Torras and Boyce (1998, 158) show that “more equitable distributions of power tend, ceteris paribus, to result in better environmental quality.” Thus, we expect higher levels of income equality to be associated with higher levels of environmental quality. We add an additional element of the income measure by including the poverty rate. Poverty reflects the percentage of county residents living below the poverty line. Although this variable is highly correlated with income, it measures variation along the low end of the income continuum and is thus expected to be positively related with pollution. The greater the portion of a county living in poverty, the greater the population that is willing to sacrifice environmental quality for production and income increases.
EKC studies tend to include some measure of population density or the spatial pattern of economic activity to account for the correlation between human activities and pollution generation (Grossman and Krueger 1995; Kaufmann et al. 1998). Selden and Song (1994) and Rupasingha et al. (2004) also control for population density, but alternatively hypothesize that it is inversely correlated with pollution emission because “thinly populated areas are likely to be less concerned about pollution activity than densely populated areas” (Rupasingha et al. 2004, 413).
In this study, Ruralness denotes the percentage of the county population that lives in rural areas (multiplied by 100), as designated by the US Census Bureau. Ruralness offers a better description of population dispersion within a county compared to population density because it reflects the frequency of a given lifestyle (urban vs. rural), not simply the aggregate population divided by the county’s physical area. For example, a county encompassing one large city and otherwise rural residents could have the same population density as a similarly sized county that is even populated. It is hypothesized that more rural counties have lower PM2.5 concentrations, given more sparse pollution production processes (factories, power plants, etc.). Furthermore, because rural areas are expected to have lower concentrations of particulate matter, they likely worry less about environmental protection and have fewer or more lax regulations (Selden and Song 1994). From this perspective, we expect that rural counties tend to correlate with higher emission levels.
Age reflects the median age of county residents and is thought to relate positively to pollution levels for a few reasons. Younger people are often more proactive with regard to environmental conservation efforts and renewable energy technologies. They are more likely to have young children and thus worry about their children’s future with regard to the status of the environment they will eventually inherit. Finally, younger populations tend to be more willing to adapt, and adapt more quickly, to new technologies and/or processes that may decrease pollution.
A racial diversity variable (RacialFrag) is included, following a number of US state-level and county-level studies (Gawande et al. 2000; Rupasingha et al. 2004). The relationship between racial diversity and pollution shares some connection to that between social capital and pollution. Racially heterogeneous communities tend to have more variation among economic preferences and social ties, making collective action efforts to enact stricter environmental regulation more challenging (Rupasingha et al. 2004). Racial diversity is quantified by a racial fragmentation index, following Alesina and La Ferrara (2000). The index reflects the probability that two randomly drawn people from a given county will belong to different racial groups. Probability is calculated using the following racial fractionalization index:
The variable race i is the fraction of county population that self-identifies as the ith group: white, black, Asian and Pacific Islander, American Indian, and Other. Higher values represent greater fragmentation.
Education represents an educational attainment index comprised of the percentage of college graduates in a given county, minus percentage of high school dropouts (multiplied by 100). This index accounts for both high and low attainment levels, which jointly contribute to the aggregate education level of a county population. It is expected that more educated populations are inversely related to pollution because of increased access to information and understanding of the implications of higher pollution levels (Gawande et al. 2000; Rupasingha et al. 2004).
Three measures of county growth are included in the empirical model to account for the context in which the cross-sectional snapshot of the income–pollution relationship is taken. These include PopChange, percentage change in population from 1990 to 2000 (multiplied by 100); IncChange, percentage change in per capita income from 1990 to 2000 (multiplied by 100); and UemployChange, percentage change in the annual average unemployment rate from 1990 to 2000 (multiplied by 100). 4 Neoclassical economic theory suggests that changes in population reflect the growth or loss of job opportunities. From this perspective, one could argue that population growth is indicative of productivity increases and, thus, pollution increases. On the other hand, Deller and Lledo (2007) show that areas with high values of environmental amenities (i.e., low pollution levels) tend to attract new populations seeking enhanced quality of life. By this line of reasoning, an increase in population would reflect low pollution levels.
With regard to income changes, the expected sign of this variable seems to be dependent on the location of the bulk of US counties along the EKC. If most counties are located on the left-hand side of the EKC, growth in income over the preceding decade could be associated with higher pollution levels. Alternatively, if most counties are located on the right-hand side of the EKC, preceding growth could be associated with lower pollution levels.
Changes in unemployment rates are logically associated with changes in production; however, there is also an argument that they factor into local levels of social capital. When people lose their jobs, they tend to pull back from voluntary activities; activities that reflect civic engagement and community participation. The expected effect of unemployment depends on which mechanism of correlation is stronger. Unemployment could reflect a loss in production activities, which would lead to lower pollution levels. Alternatively, unemployment could be linked to lower levels of social capital, which are associated with higher levels of pollution.
Many EKC studies include variables reflecting industry composition to account for variation in major pollution-generating processes (Suri and Chapman 1998; Kaufmann et al. 1998; Rupasingha et al. 2004). The relative size of a county’s manufacturing sector is likely highly correlated with air pollution concentrations, while the farming sector is likely closely tied to higher use of diesel-powered equipment. A ratio of industry sector employment to total employment is used as a proxy for sector size. 5 Farming employment includes proprietors and “wage and salary” workers.
Finally, state-level dummy variables are included to capture qualitative regulations, policies, and incentive structures implemented by state governments that have a significant effect on industry composition and relative pollution levels within each state. Counties with a shared state, regardless of their relative location with the state, are subject to similar impacts from state legislatures and regulatory bodies. The signs of these variables are not of particular concern to this analysis (and will not be reported); however, they are included to mitigate bias among the variables of interest. Basic descriptive analysis of the data is provided in Table 1.
Measures of Social Capital
The most difficult metric to develop and implement is social capital. In essence, social capital is easy to describe but difficult to empirically operationalize. Where human capital describes individual qualities such as education and skill levels, social capital describes the quality of connections between individuals and how well they are leveraged to achieve success within the region—creating a whole that is greater than the sum of the parts. In his 1995 article Blowing Alone: America’s Declining Social Capital, which helped popularize the term social capital, Putnam (1995, 67) defines social capital as “features of social organization such as networks, norms, and social trust that facilitate coordination and cooperation for mutual benefit.”
Staatz (1998, 2) builds on the idea that social capital facilitates social cohesion, claiming that it acts “to increase the ‘liquidity’ of social interaction…much like an expansion of the money supply.” Schmid and Robinson (1995, 66) view social capital as a productive asset, “like money in the bank, [social capital] makes assets more productive and saves costs—besides being valuable in itself,” as does Woolcock (2001, 67) who suggest that “the basic idea of ‘social capital’ is that one’s family, friends and associates constitute an important asset, one that can be called upon in a crisis, enjoyed for its own sake and/or leveraged for material gain.” Woolcock (2001, 69) further clarifies by arguing that “human capital resides in individuals, social capital resides in relationships.”
Fukuyama (2002, 7) argues that “social capital is an instantiated informal norm that promotes cooperation between two or more individuals [and gives way to] trust, networks, civil society and the like.” Pretty and Ward (2001, 210) write that social capital captures “the idea that social bonds and social norms are an important part of the basis for sustainable livelihoods.” Adler and Kwon (2002, 17) define it as “[t]he goodwill that is engendered by the fabric of social relations and that can be mobilized to facilitate action.” Similarly, Rupasingha, Goetz, and Freshwater (2006, 84) write about social capital as an intermediate “good” that contributes to more efficient means of production “by investing in relationships that reduce transactions costs, we can reduce the friction in productive activities.”
Shaffer, Deller, and Marcouiller (2004, 203–204) suggest that
[S]ocial capital refers to features of social organization such as networks, norms, and social trust that facilitate coordination and cooperation for mutual benefit. Networks of civic engagement foster norms of general reciprocity and encourage the emergence of social trust. Social capital consists of the social networks in a community, the level of trust between community members, and local norms. These networks, norms and trusts help local people work together for their mutual benefit.
Flora and Gillespie (2009, 117) write that “[s]ocial capital is made up of the interactions among groups and individuals for mutual support.” They distinguish between bonding social capital, which builds stronger ties among similar people and groups, and bridging capital, which builds ties across differences of place, class, gender, ethnicity, and religion. Paldam (2000) suggests that there are three “families” of social capital concepts that hinge on (1) trust, (2) ease of cooperation, and (3) networks. Higher levels of trust allow for ease of volunteer cooperation and expanded networks create a larger pool of individuals with which to cooperate. Social capital is the “glue” that ties the three “families” together.
Clearly, there is consensus about the general notion of social capital, but its definition is nebulous. This diverse understanding of social capital has prompted criticism that “social capital [has] become all things to all people, and hence nothing to anyone” (Woolcock 2001, 69). Further, Durlauf (2002, 459) claims “the concept itself [is] too vague to permit analysis whose clarity and precision matches the standards of the [economics] field.” Furthermore, among a number of definitions tangle the “causes and effects of social capital as a collective trait…giving rise to much circular reasoning” (Portes 2000, 4). Despite these limitations of the definition of social capital, the concept is applied to a variety of social and economic phenomenon.
There is a long literature on social capital and environmental quality (Pennington and Rydin 2000; Rydin and Pennington 2000; Pretty and Ward 2001; Grafton and Knowles 2004); however, relatively little analysis exists on the role of social capital within the specific framework of the EKC (Paudel and Schafer 2009; Paudel et al. 2011). In their work on nongovernmental groups and resource management, Pretty and Ward (2001) identify a maturation process through which group formation transforms into successful group management. Group formation generally occurs in reaction to an outside threat or crisis and as group abilities and confidence grows, accumulating interpersonal connections and trust, it starts to become more inward looking and gain momentum independent from the initial threat(s). In essence, the group is building the fundamental foundations of social capital. As groups continue to mature, they tend to have strong ties to external agencies and are proactive, rather than reactive, in their agendas. The mature group demonstrates high levels of social capital: strong interpersonal connections, ties to external entities and, most importantly, proactivity.
Grafton and Knowles (2004) come close to an analysis of social capital within the context of the EKC with their study of the linkages between social capital and “national environmental performance.” The concept of social capital is deconstructed into four subconcepts, the components of which are regressed on six different measures of environmental performance, including “positive” measures like an environmental sustainability index and “negative” measures like urban concentration of sulfur. The four subgroups of social capital are: civic social capital (community engagement, social associations, and trust); public social capital (institutional quality and accountability); social divergence (barriers to communication between individuals and groups, such as ethnic and religious fragmentation); and social capacity (aggregate individual qualities that reflect ability to achieve human potential, such as education and health). Grafton and Knowles (2004) control for a linear income effect as opposed to a nonlinear or quadratic specification which categorically differentiates it from the EKC studies.
Given the lack of consensus on the definition of social capital, it is not surprising that there is also a lack of consensus with regard to quantitative measurement. Historically, social capital has been crudely proxied by crime rates or poverty rates, but our thinking of how to measure social capital has moved beyond these simplistic measures along two lines: stated versus revealed preferences. Surveys capture stated preferences and are thus limited by the subjective nature of the responses. Per capita counts of membership organizations, and other institutions associated with social capital, are more objective in the sense that they are observed; however, significant assumptions must be made about the likelihood that a given type of organization (church, Elks club, etc.) actually reflects the civic engagement and networking we are trying to measure.
Surveys such as the national Opinion Research Council’s General Social Survey and University of Michigan’s World Values Survey provide primary, individual-level data relating to community engagement and social norms. The benefit of a survey is that questions can target the specific topic of interest. How involved are you in the community planning process? How well do you know your neighbors? The limitation, however, is that respondents’ opinions about their behavior may differ from actual behavior. Furthermore, survey responses measure the connectedness of the individual, not the quality or effectiveness of the connections between individuals, which define social capital. Nevertheless, survey responses are commonly used to measure social capital (Grafton and Knowles 2004; Paudel et al. 2011), particularly in country-level analyses where other country-to-country comparable data are limited.
The social capital index created by Rupasingha, Goetz, and Freshwater (2006) uses secondary data from the US Census’ County Business Patterns (CBP) and the National Center for Charitable Statistics to construct their associational density measure using the density of local groups per capita of the following entities: (1) civic organizations, (2) bowling centers, (3) golf clubs, (4) fitness centers, (5) sports complexes, (6) religious entities, (7) political organizations, (8) labor organizations, (9) business associations, and (10) professional organizations. To effectively weight each individual measure in a combined index, Rupasingha, Goetz, and Freshwater (2006) employ principal component analysis to determine the subgroup of variables that explain the bulk of the variation among all variables. 6
Paudel and Schafer (2009) adopt a similar approach, using a single index comprised of various counts of civic associations as a proxy for social capital within Louisiana parishes. The organizational categories employed by Paudel and Schafer (2009) vary slightly from those used by Rupasingha, Goetz, and Freshwater (2000, 2006) by including dance studios and music organizations, but remove voter turnout, census response rate, and nonprofit organizations. They use the following components, measured as entities per 10,000 persons: (1) dance studios, schools, and halls, (2) bowling centers, (3) music, amusement, and recreational services, (4) public golf courses, (5) membership sports and recreation clubs, (6) business associations, (7) professional, (8) labor, (9) civic and social, (10) political, and (11) religious organizations.
The social capital measure used in this analysis combines the work of Putnam (1995), Alesina and La Ferrara (2000), Rupasingha, Goetz, and Freshwater (2000, 2006), Rupasingha et al. (2004), Rupasingha and Goetz (2008), and Deller and Deller (2010). We use the Census Bureau’s CBP which provides aggregate organizational data, by economic sector, for each county in the United States. As an indicator of local engagement, this analysis incorporates the total number of establishments per county population 1,000 for a range of sectors, as defined by the North American Industry Classification System, into the social capital index:
Environment and wildlife organizations;
Grant making and giving services;
Museums, historical sites, and like institutions;
Civic and social organizations;
Business associations;
Professional organizations;
Labor unions and similar labor organizations;
Political organizations.
Civic and social organizations, as well as business, professional, labor, and political groups, are considered the most direct measure of social capital (Deller and Deller 2010). Establishments in these sectors often reflect membership bases and networks that cross socioeconomic or demographic lines. Environmental organizations, grant-making institutions, and museum and historical sites are often reliant on charitable contributions and local volunteers, both of which reflect community or regional engagement from individuals and businesses. Environment and wildlife organizations are a direct indicator of social capital within an environmental context and help tailor the social capital measure to an EKC application.
The density of religious organizations is also incorporated into the social capital index used in this analysis. Churches, synagogues, and mosques serve as meeting and networking places, and often facilitate volunteer-based programs aimed at filling local community needs (Deller and Deller 2010). This is particularly true in rural communities. Furthermore, Putnam (1995, 68) states that “[r]eligious affiliation is by far the most common associational membership among Americans…. For example, the United States has more houses of worship per capita than any other nation on earth.” Hence, the concentration of congregations per population 1,000 across different types of religions is used as a measure of social capital:
Evangelical congregations;
Catholic congregations;
Jewish congregations;
Muslim congregations;
Other congregations.
Local densities of organizations that require financial support and volunteers, membership-based associations, and religious organizations, measured on a per population 1,000 basis, are simply summed together to create an index of social capital for US counties. The index ranges from 0.38 to 11.07 and averages 2.58. The distribution of the index is somewhat right skewed, suggesting there are a few counties with relatively high levels of social capital. It is hypothesized that higher levels of social capital are associated with lower levels of social capital and would decrease the income turning point of the EKC.
Finally, we include a handful of variables that are intuitively connected to social capital as exploratory variables. ResidentialStability is the percentage of county residents over five years old who lived in the same county in 1995 and 2000 (multiplied by 100). It is hypothesized to have an inverse correlation with pollution because long-term residents are likely more invested and connected to their regions and thus more concerned with local pollution levels. Similarly, HomeOwnership, the percentage of owner-occupied homes, is thought to be negatively correlated with pollution. Homeowners are more vested with the community and sensitive to activities that might distract from the value of their houses. In essence, the likelihood of the NIMBY syndrome is more likely to be present in communities with higher concentrations of homeownership. Commute reflects the average commute time, in minutes, that county residents working outside the home spend in transport between their homes and their place of employment. Longer commute times are thought to reflect increased segmentation of residents’ lives: live in one area, work in another. Such a division is likely to detract from awareness of local issues and the time commitment of long commutes competes with time availability for activities that foster civic connectedness in the home community. Furthermore, there is likely to be some direct association between commute time and air pollution due to the contributing factor of vehicle exhaust, however, mobile sources such as vehicles account for only one-tenth of PM2.5 emissions, on average.
Spatial Econometrics
Four spatial specifications of the reduced-form EKC model, along with a basic ordinary least squares (OLS) regression are employed to explore the sensitivity of results across estimation methods. If the overall empirical results and policy insights are consistent across the different estimators, then a certain level of confidence is given to our inferences. The general spatial autocorrelation (SAC) model is expressed by:
Estimation of the SAC results in parameter values
The sign of
The other variable of interest, social capital, has a marginal impact represented by:
The mixed spatial autoregressive (SAR) model, also called the spatial lag model, and the spatial error model (SEM) are specific instances of the general model. In the case, where λ = 0, the general model reduces to the SAR:
Thus, the estimate of
The spatial Durbin model expands on the spatial lag model by adding a term of lagged explanatory variables. In reference to the EKC, the spatial Durbin reflects not only the effect of neighboring county pollution levels on County i’s pollution level but also the effect of explanatory variables in neighboring counties on pollution in County i. LeSage and Pace (2009) argue that the spatial Durbin model is perhaps the most general specification of the spatial model and the more traditional spatial lag, error and general models are special cases. Indeed, Elhorst (2010) very effectively argues that the movement to the spatial Durbin family of estimators is “raising the bar” in applied spatial econometrics.
The Durbin specification is expresses by:
where, ρ is the parameter on the lagged dependent term and βL is the vector of parameters on the lagged explanatory variables (W 2 X; Anselin 1999; LeSage and Pace 2009; Elhorst 2010). As with the general spatial model, W 1 and W 2 do not have to be constructed using the same strategy, but they can be. Each explanatory variable enters the equation twice, as a direct modifier of the county of interest and as an indirect modifier of neighboring counties. The total effect of a unit change in a given variable is the sum of that variable’s direct and indirect effects (LeSage and Pace 2009). OLS estimates of the original linear model, along with maximum likelihood estimates for the four spatial models discussed in this section are compared in the following section.
Empirical Results
There are five sets of results to review: the regression results from OLS and the first three spatial models using concentrations (Table 2), the spatial Durbin results (Table 3). Given the volume of individual results, a detailed discussion of each variable in each model is impractical and as such we will limit ourselves to general patterns in the results with a focus on income, social capital, and ruralness. Based on the various permutations of the R 2, the model explains about three-quarters of the variation in the concentration data and given the cross-sectional nature of the data this relatively strong.
Regression Results for PM2.5 Concentrations.
Note: OLS = ordinary least squares; PM2.5 = fine particulate matter.
p Values in parentheses: *p < .05. **p < .01. ***p < .001.
Durbin Model Results for PM2.5 Concentrations.
Note: PM2.5 = fine particulate matter.
p Values in parentheses: *p < .05.
In terms of the spatial dependency, the spatial lagged parameter ρ as well as the SEM parameter λ are statistically significant, again suggesting that the aspatial least squares may provide inaccurate insights. From the spatial Durbin, there is also evidence of spatial spillover effects (indirect) for a handful of variables. A casual comparison of the estimated coefficients across the four models provided in Table 2 suggests stability in the results. As argued above, this stability across estimators lends a level of confidence to our results.
Higher median age, education levels as well as change in population, and income tends to put downward pressure on concentrations. Somewhat unexpected, higher poverty rates also place downward pressure on concentration rates. Increased levels of racial heterogeneity and income inequality are associated with higher PM2.5 concentrations. Higher commute times along with greater dependency on farming and manufacturing for economic activity are also associated with higher concentration levels. Higher levels of residential stability, a very simple measure of social capital, are associated with higher concentrations along with higher levels of homeownership. The latter result is weaker from a statistical significance across the different estimation methods.
For the spatial Durbin, the total income effect is not statistically significant, but the coefficients are of the correct signs with only the direct effect on income squared being statistically significant. Despite this last result, there is largely consistent evidence of the presence of the EKC with the PM2.5 concentration model. The income tipping point is estimated to range from a low estimate of US$24,000 for the spatial lag model to a high of US$25,500 for the SEM model (OLS estimate US$24,600 and general estimate US$24,800). This tipping point is comparable to the work of List and Gallet (1999) who estimate the tipping point to be US$22,675 using US state-level emissions data from 1929 to 1994. Perhaps most comparable to our work, Rupasingha and his colleagues’ (2004) analysis of toxic releases to air, water, and land using US county data for the turning point for the air pollution spatial model is US$22,127—while the turning point for the air pollution nonspatial model is US$21,459.
The results for the two other variables of particular interest, measures of social capital and ruralness, are encouraging and consistent with our prior expectations: higher levels of both social capital and ruralness are associated with lower levels of PM2.5 concentrations. As outlined in detail above, communities with higher levels of social capital, broadly defined, are less likely to tolerate higher levels of pollution and they are in a better position to organize the community to act. It also makes sense that more rural US counties are less likely to have high levels of PM2.5 concentrations when compared to more urban settings. The levels of emissions, and as a result potential concentration levels, are likely to be higher in urban areas. Also, rural residents are less likely to accept higher levels of PM2.5 concentrations as preferences for cleaner air appear to be higher in more rural areas. The interplay between social capital and ruralness suggests that the impact of social capital is weaker in more rural areas. Again, this makes sense; rural areas are less likely to have high concentrations of the types of institutions that define our social capital measure.
Consistency of results across the various estimators lends confidence to our overall results. First, the data support the presence of the EKC with a tipping point that is within a reasonable range of the income data. Second, as expected, social capital, as captured by our index, is associated with lower levels of PM2.5 concentration. Third, all else held constant, PM2.5 concentrations are lower in more rural areas. Fourth, the dampening effect that social capital places on PM2.5 concentrations is weaker in more rural areas. Fifth, most of the control variables, exempt for residential stability, homeownership and poverty, behave as expected.
Conclusions
Using the framework of the EKC, we test the hypothesis that communities with higher levels of social capital, all else held constant, will tend to have lower levels of pollution. The rationale is that communities with higher levels of social capital are in a better position to place downward pressure on pollution. Using PM2.5 emissions and concentration data for US counties, we find consistent evidence of the EKC across a range of spatial estimators. We find the tipping point, or peak of the EKC, between US$27,100 and US$28,200 (per capita income) for PM2.5 emissions and between US$24,000 and US$25,500 for PM2.5 concentration.
We also find that higher levels of social capital, measured through concentrations of business, social, professional and advocacy organizations, along with religious institutions, are consistently associated with lower levels of both PM2.5 emissions and concentrations. This result supports the central hypothesis that communities with a stronger sense of civic solidarity, ability to network, and work together in collective action are better position to promote environmental quality by minimizing pollution. We also find that the ability to act on this social capital is weaker in more rural counties. This result is not unexpected because the sense of individualism tends to be stronger in rural areas and collective action tends to be less prevalent.
The results here are important for several reasons. First, the presence of the EKC suggests that as income grow, pressures to reduce pollution, specifically PM2.5 emissions and concentrations, will grow. The implication is that one mechanism to promote a cleaner environment is to promote income growth. Second, the promotion of social capital has multiple implications one of which is the ability to facilitate lower levels of pollution at the local level. Third, how this relationship plays out will vary over the urban–rural spectrum. Fourth, this work adds to a small but growing economics literature that explores ways in which to quantify metrics of social capital using widely available secondary data sources.
Footnotes
Acknowledgment
This work has benefited from the support of Paul Mitchell and the helpful comments of Jennifer Alix-Garcia and David Marcouiller
Authors’ Note
All errors are the responsibility of the authors.
Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: Partial support for this work was provided by the Wisconsin Agricultural Experiment Station, University of Wisconsin–Madison.
