Towards a General Theory of Environmental Inequality: Social Characteristics of Townships and the Distribution of Pollution in China’s Jiangsu Province

Ethan D. Schoolmana* and Chunbo Mab

aDepartment of Sociology, University of Michigan, 500 S. State St., Rm. 3001, Ann Arbor, MI 48109, USA. Email: [email protected] bSchool of Agricultural and Resource Economics, Centre for Environmental Economics and Policy, The University of Western Australia, Crawley, WA 6009, Australia. Email: [email protected]

*Corresponding author: Tel 001-518-257-0471; Fax 001-734-763-6887

07 November 2011 Working Paper 1123 School of Agricultural and Resource Economics http://www.are.uwa.edu.au

Citation: Schoolman, Ethan D. and Chunbo Ma. (2011) Towards a General Theory of Environmental Inequality: Social Characteristics of Townships and the Distribution of Pollution in China’s Jiangsu Province, Working Paper 1123, School of Agricultural and Resource Economics, University of Western Australia, Crawley, Australia.

© Copyright remains with the authors of this document. Towards a General Theory of Environmental Inequality: Social Characteristics of Townships and the Distribution of Pollution in China’s Jiangsu Province

Key Words: Environmental Inequality, Hukou System, Pollution, China

JEL: D63, J15, J61, Q53, R12, R23

1. Introduction

Social scientists have explored the distribution of pollution and pollution-based health risk in the U.S. for nearly three decades. Consensus on whether environmental inequality exists in the U.S., and if so why, has evaded researchers for much of this time, and the scientific merit of early work in particular has been called into question (Bowen 2002). In recent years, however, data and methods of analysis have diversified and improved, in ways that we review below, and the findings of researchers have begun to converge. Since the early 2000s, every major study has found that African-American communities and those of other racial minorities are exposed to more pollution than those of whites, controlling for income, land use, and other potential confounding factors (Mohai and Saha 2007; Morello-Frosch and Jesdale 2006; Morello-Frosch, Pastor, and Sadd 2001; Pastor, Morello-Frosch, and Sadd 2005). There is still disagreement over whether income is also a valid predictor of pollution exposure. But knowledge on race-based environmental inequality in the U.S., in particular, has made important strides over the course of thirty years.

In this paper we ask whether the theory and methods developed to test for, analyze and make sense of environmental inequality in the U.S. can be extended to a major developing country like China. The environmental implications of China’s economic expansion over the past two decades have been staggering. China is now the world’s second largest economy, and its overwhelming reliance on coal has made it the world’s largest emitter of greenhouse gasses (Zeng et al. 2008). Urban air quality is so poor that Chinese authorities shuttered factories and power plants throughout the Beijing area in the weeks leading up to the 2008 Olympics, in order to ensure clean air for athletic events. Yet compared to the U.S., research on environmental inequality in China is underdeveloped. Journalism and qualitative research have helped to communicate the depth of the environmental challenges facing all Chinese, from farmers to coal miners to the burgeoning urban middle class (Knup 1997; Palmer 2007; Pan 2001). Yet systematic, quantitative research on who is most affected by pollution in China is in its infancy (Jian 2005; but see Ma and Schoolman 2010 and 2009, which we review in greater detail below). Lack of data and the slow diffusion of sociological theory and methods across the Pacific are major reasons why quantitative explorations of the environmental challenges to equality posed by China’s fearsome economic transformation have yet to take form.

This paper represents, to the best of our knowledge, the most thorough systematic study to date of environmental inequality in China. We are interested in two main questions. First: does environmental inequality exist in China, and if so, does it break down along the same lines as in the U.S.—i.e. race, and to a lesser extent class—or along different lines? Second: if environmental inequality does exist in China, then why does it exist? Might similar historical processes be held to account for its existence in both China and the U.S., or do the

2 two countries need different explanations for a shared sociological phenomenon? In order to answer these questions, we employ methods developed for the U.S. context to analyze an original dataset on the location and emissions of pollution-producing facilities in one of China’s fastest growing provinces. We find that migrants from China’s poor countryside are exposed to a disproportionate amount of pollution, even after controlling for other factors, such as the presence of “dirty and hard” industries in which rural migrants are most likely to find work. Moreover, despite the many stark differences between Chinese and U.S. society, we theorize that environmental inequality exists in both countries for the same underlying reason. Specifically, we theorize that the benefits of industrialization accrue in ways both unequal and predictable. Those who occupy the low rungs of the social ladder, whether racial minorities in the U.S. or rural migrants in China, are those who bear the brunt of the massive environmental consequences of a changing economy.

In the next section we abstract from the history of industrialization and urban expansion in the U.S. a set of general processes which we argue can be deployed to predict and explain environmental inequality in China. We then introduce our data and methods of analysis, which draw heavily on recent innovations in the use of GIS to analyze the spatial distribution of pollution and pollution-producing facilities. Finally, we present our analyses, discuss our findings, and suggest future directions for the study of environmental inequality in China and other developing countries.

2. Towards a Theory of Environmental Inequality

Two main findings have emerged from three decades of research into environmental inequality in the U.S. First, as we have already noted, every major study since the early 2000s has found that racial minorities are disproportionately exposed to environmental pollution and hazards, and that this effect persists across income levels. The consensus that appears to have emerged is especially noteworthy, given that the first twenty years of research saw significant disagreement. Many early studies did indeed find hazardous waste facilities, superfund sites, and other pollution-producing facilities to be located overwhelmingly in minority, and to a lesser extent low-income, communities (CRJ 1987; Daniels and Friedman 1999; GAO 1983; Goldman and Fitton 1994; Hamilton 1995; Hird 1993; Hird and Reese 1998; Lester, Allen, and Hill 2001; Perlin et al. 1995; Ringquist 1997; Zimmerman 1993). But many others found just the opposite: that neither race nor income were significantly associated with proximity to environmental hazards (Anderton et al. 1994; Anderton, Oakes, and Egan 1997; Been and Gupta 1997; Davidson and Anderton 2000; Greenberg 1993; Oakes, Anderton, and Anderson 1996). The inconsistency of these early studies has been attributed to validity problems inherent in their measures of pollution exposure (Mohai and Saha 2006; Downey 2006), which we discuss in greater detail when we introduce our own data and methods below. More recent research builds measures of exposure by using either geographic information systems (GIS) or sophisticated environmental models (Mohai and Saha 2007; Morello-Frosch and Jesdale 2006; Morello-Frosch, Pastor and Sadd 2001; Pastor, Morello- Frosch, and Sadd 2005). The consensus achieved by these newer studies suggests that race- based environmental inequality has been and continues to be a significant problem in the U.S.

The second main finding to have emerged, based on historical and qualitative research in particular, is that race-based environmental inequality in the U.S. has multiple causes, including politically-enabled residential sorting and siting decisions that are both unintentionally and intentionally prejudicial (Brown 1995; Bryant and Mohai 1992; Szasz and Meuser 1997). The complex interplay of urban expansion, industrialization, and racial

3 discrimination in the U.S. is critical for the purposes of this paper, because, in our view, the underlying mechanisms are applicable to the case of China, as well. In the U.S., prior to widespread automobile ownership, industry maximized access to employees and transportation networks by locating facilities in densely populated urban areas. At a time when increases in urban population far outpaced increases in land area, exposure to environmental pollution from industrialization was unequal, but widespread (Fogelson 2001). Once the automobile and post-war policies like the G.I. Bill made mass suburbanization possible, however, whites fled dirty, crowded urban cores. Ensconced in gleaming new communities, whites worked to keep suburbs closed to minorities through community pressure, discriminatory lending, minimum lot sizes, and opposition to affordable housing (Fishman 1987; Fogelson 2005; Jackson 1985). Minorities and the poor were left to deal with the environmental legacy of urban industrialization: brownfields, power plants, highways, and little money for remediation (Bullard 2000; Checker 2005; Hurley 1995; Lerner 2005). Political marginality and environmental degradation reinforced one another. Polluted areas lost the very middle-class residents whose influence might have kept dirty industrial and waste-management facilities from continuing to locate near minority communities. At the same time, low rents on land in polluted areas represented a perverse draw to those who could not afford to live in cleaner communities, adding another layer to the cycle of inequality (Been and Gupta 1997).

The precise race-based mechanisms driving environmental inequality in the U.S. cannot be transposed directly to a country like China, where ethnic Han make up 92 percent of the population and race is relatively unimportant (Quan 2002). But underlying the specific history of environmental inequality in the U.S. are three more general processes, which may, in theory, be applicable outside the U.S. context. First, individuals of high social status use their economic, political, and social capital to distance themselves from the environmental externalities of industrialization. Second, poverty, political weakness, and social exclusion present significant obstacles to individuals of low social status seeking to avoid the worst pollution that urban areas have to offer. Third, the relationship between social status and pollution exposure is mediated by infrastructure: improvements in transportation allow those with sufficient resources, both economic and social, to live and work progressively further from undesirable areas, while adding to the downward socioeconomic and environmental spiral of low-status communities. These three general processes can be extended to a country like China, if we accept that the role that race has played in the U.S. is a role that any invidious distinction can accomplish which successfully, either through cultural norms or the law, assigns special privileges to one group over another.

In the U.S., it is racial minorities whose social position has led, through the mechanisms outlined above, to a disproportionate environmental burden. In Chinese history, an arguably analogous low-status group has been people born in rural areas. The superior social status of urban residents over those from rural areas is codified and expressed through the official household registry system known as hukou. Though a crucial tool of the modern Chinese state, the roots of hukou go back over two thousand years, when emperors mobilized a vast bureaucracy to monitor and extract revenue from the population (Wang 2005). The contemporary hukou system was established in the 1950s by the post-war communist government under the tutelage of the Soviet Union (Chan and Zhang 1999). Each individual was, and still is, required to register in either the city or the village or commune of a parent (typically the mother). From the beginning, privileging urban residents over villagers was a prime consequence of the registry, with often tragic results. During the Great Leap Forward (1958-61), grain allotments skewed toward the cities resulted in the death of over twenty

4 million people, almost all from rural areas (Wang 2005). By the mid-1960s, the hukou registry had facilitated the forced movement of nearly forty million urban residents from their homes in cities to their former villages or the villages of their parents. The remaining urban population, the elite of the nation, “became a truly privileged minority in China for the next three decades and is still heavily subsidized and favoured by the state today” (Wang 2005:47).

The socioeconomic consequences of the hukou system have taken on new meaning in recent decades, as tens of millions of Chinese have streamed into urban areas in search of jobs in the export economy. Urban workers with a rural hukou constitute an enormous “floating population” whose size, measured at approximately 79 million in 2000 (Liang and Ma 2004), exceeded 120 million in 2007—over one-third the size of the current U.S. population (Zhu 2007). Rural migrants in China’s teeming, modern cities are systematically disadvantaged by the hukou system—though the outcomes are perhaps not surprising, if one takes the effect of race in the U.S as a model. Urban workers with official rural residency have lower wages and returns to education than those with urban residency (Fan 2001; Knight and Song 1999; Du, Gregory, and Meng 2006). Children of such workers, whose official status is rural even if they have been born and raised in the city, go to inferior schools, score lower on competitive exams, and are much less likely to attain a university education (Wu and Treiman 2004). Rural designation exacerbates the negative effects of gender on income (Huang 2001). A rural hukou harms one’s chances of becoming a member of the communist party—an indispensable gateway to the middle class (Wu and Treiman 2004). Moreover, without legal right to work and live in the city, and with limited access to services, members of the floating population, like undocumented workers in the U.S., are at the mercy of employers and the state. On the other hand, individuals with an official urban residence are guaranteed access to the “subsidized education system, subsidized housing, welfare programs, and community cultural activities,” while migrants “have no such entitlements” (Liu 2005:136). In sum, the hukou system “was and remains the institutional guardian of the deep urban-rural divide that has characterized China since the mid-1950s” (Cheng and Selden 1994:667).

Like the colour of one’s skin in the U.S., one’s hukou, or place of residence, is a fundamental and only marginally more malleable aspect of one’s identity in China, with many of the same consequences for people on the wrong side of the status boundary. Indeed, the migration of villagers to cities in China bears a striking resemblance to the American Great Migration of the early 1900s, when over a million poor African-Americans from rural areas left the South for work in heavy industry in Northern cities. Our main argument in this paper is that hukou in China, like race in the U.S., will also be found to drive environmental inequality through the processes described above. As members of China’s new middle class trade bicycles for automobiles, they are already colonizing inexpensive land on urban outskirts and establishing new suburban developments that emphasize cleanliness, family life, and escape from urban ills (Campanella 2008). Rural migrants are being isolated still further, left to live in close proximity to factories, waste-treatment facilities, and other environmental hazards. Our principle hypothesis in this paper, which we test in the analyses below, is thus that urban areas with large numbers of rural migrants are exposed to more pollution than those without, controlling for other factors.

As we note above, there has been little systematic inquiry into environmental inequality in the developing world, let alone in China, using methods developed for the U.S. case. The sole exception, to the best of our knowledge, is an examination of power plant and factory locations in China’s Henan province (Ma and Schoolman 2010, 2009). The present

5 study improves upon this earlier work in several crucial ways. First, using previously unavailable data on the actual emissions, as well as locations, of pollution-producing facilities, we test our models against a wide variety of weighted and unweighted dependent variables; this ability to check results across related but analytically distinct outcomes leads, ultimately, to more robust conclusions. Second, using new data on the geographical distribution of different industries, we address the question of whether a relationship between rural migrants and pollution may simply be a function of the tendency of migrants to work in certain industries, like mining and heavy manufacturing. Third, we offer in this study a more fully- realized description of the parallels between the social history of African-Americans in the U.S. and rural migrants in China—parallels whose possible environmental implications are the subject of the following analysis.

3. Data and Methods

3.1. General Method There are three main kinds of environmental inequality research, each corresponding to a different way of measuring the geographic distribution of pollution. Unit-based studies measure the presence of environmental hazards, like toxic waste disposal sites and incinerators, within spatial units like zip codes or census tracts. Until the early 2000s, most research was of this type (CRJ 1987; Anderton et al. 1994; Anderton, Oakes, and Egan 1997; Been and Gupta 1997; Daniels and Friedman 1999; Davidson and Anderton 2000; GAO 1983; Greenberg 1993; Hamilton 1995; Hird 1993; Hird and Reese 1998; Lester, Allen, and Hill 2001; Oakes, Anderton, and Anderson 1996; Perlin et al. 1995; Ringquist 1997; Zimmerman 1993). Unit-based studies exhibit two main problems. First, spatial units which are very near to one or more environmental hazards, but which do not actually contain any hazards within their boundaries, may be just as or even more affected by these hazards as the host units themselves. Second, the populations of large host units may be located farther from resident environmental hazards than those of small host units—but the unit-based method treats the impact of environmental hazards on the populations of all host units, no matter what their size, as the same.

Newer distance-based studies address this problem by using GIS to address the fact that pollution does not stop at administrative boundaries and that host unit size matters when calculating the impact of nearby pollution sources. Downey (2006) outlines a five-step process in his study of pollution in metropolitan Detroit: 1) break each census tract into a grid of identical cells; 2) use GIS to draw containment areas of a one-quarter kilometer radius around the center of each cell; 3) use the U.S. Toxics Release Inventory (TRI) to find the amount of pollution released within the containment area of each cell; 4) sum the pollution load of all cells in each census tract to find the total pollution load of each tract; 5) determine whether particular tract-level socioeconomic characteristics are associated with increased levels of pollution. Mohai and Saha (2006, 2007) follow a different approach, first using GIS to draw circular containment areas around hazardous waste treatment, storage, and disposal facilities (TSDFs) nationwide, and then analyzing the socioeconomic characteristics of census tracts located, in whole or in part, within these containment areas. Each study analyzes data with both distance-based and older unit-bases techniques, in order to compare the two approaches side by side, and finds that the choice of methodology yields substantively different results. For Mohai and Saha’s (2006, 2007) studies of TSDFs, distance-based analysis reveals race-based environmental inequalities that unit-based analysis fails to detect. Distance-based studies have their shortcomings: for instance, they typically do not take into

6 account mobile sources of pollution. But they represent a significant improvement over unit- based research.

Finally, exposure/risk-based studies use sophisticated atmospheric models to predict the dispersion of pollution from point and mobile sources over enormous geographical areas, and their constituent administrative units (Morello-Frosch and Jesdale 2006; Pastor, Morello- Frosch, and Sadd 2005). Because these kinds of studies rely on data that will likely not be available for China or other developing countries for the foreseeable future, we do not consider this particular methodology in detail in this paper.

In our study, like Downey (2006), we work with actual emissions data. Thus, we adopt a version of Downey’s approach. We use GIS software to give each spatial unit—in our case, the townships of Jiangsu province, as we explain below—a buffer of five miles. The result is that each township is associated with a containment area that is mathematically similar to its actual borders. Relative to the size of the average urban township (22 square miles), a five-mile buffer is in line with the methods of previous studies: Downey (2006) uses circular containment areas with a 250-meter radius, centered on 25 x 25 meter cells, while Mohai and Saha (2007) build their areas using radii of 1, 2, and 3 miles from pollution sources. In order to address the possibility that our containment areas may be capturing sources of pollution that are too far away to be having a significant impact on township populations, we also use GIS to weight the emissions of pollution sources by their distance from township centers.

After constructing the containment areas, we test the ability of primarily township- level socioeconomic characteristics to predict three different dependent variables: 1) the number of pollution-producing facilities located within each containment area; 2) the per capita amount of major pollutants emitted within each containment area; 3) the per capita amount of these same pollutants, with the emissions of each contributing facility weighted by the facility’s distance from the geocenter of the township in question. The use of GIS thus allows us to address the two major shortcomings of unit-based research. First, by creating buffered containment areas around our spatial units, we do not treat administrative entities which are close to, but which do not actually contain, environmental hazards, as ipso facto different from true host units. Second, by incorporating distance-based weights into one of our three dependent variables, we compensate for the fact that pollution sources located within the containment areas of small host units are likely, due to increased proximity, to impact the populations of these units more severely than those located within the containment areas of large host units.

3.2. Sources of Pollution

Two kinds of data are indispensable for environmental inequality research: data on sources and emissions of pollution, and data on the socioeconomic characteristics of spatial units. Historically, reliable data on pollution in China have been non-existent or not publicly available. Our pollution data come from a 2005 dataset on manufacturing and waste- treatment facilities that has recently been made available by the Environmental Protection Bureau (EPB) of Jiangsu province (the province is the largest administrative unit in Chinese government, after the nation itself). Jiangsu is a microcosm of contemporary China: it is the third-wealthiest province, but there exists a stark divide between its prosperous south, centered on the former national capital of Nanjing, and its poor, rural north. In the absence of

7 nationwide data, Jiangsu’s economic dynamism and lingering inequalities make it an ideal venue for a study of how pollution distribution in China may be shaped by social forces.

The Jiangsu EPB dataset contains two kinds of data. First, it identifies, based on a nationwide database from the China State Environmental Protection Agency (SEPA) that ranks over eighty thousand facilities by their emissions of sulfur dioxide (SO2), smog, 1 ammonia nitrogen (NH3-N), and waterborn organic compounds (COD ), the names and host administrative units of facilities whose summed emissions constitute 85 percent of the total amount of each pollutant emitted in Jiangsu. Second, it gives the actual emissions of the four pollutants for each facility on the list. The resulting dataset describes the emissions of the manufacturing facilities, power plants, and waste treatment facilities which together represent the 647 most significant sources of pollution in Jiangsu.

Precise geographical coordinates of sources were not reported in the published dataset, and we used several digital map databases in order to determine the coordinates for each facility. In this way we were able to determine precise coordinates for 541 facilities, or 83.6 percent of all pollution sources on the list. For 85 sources, we generated approximate coordinates by using either the facility’s street address, when it could be found online, or the geocenter of the facility’s host village, a sub-unit of the township, when it was given in the data. These imputed coordinates fully satisfy the research need: street address gives a very precise location; and the average village unit in China is 17.5 times smaller than the average township. For 15 pollution sources for which neither street address nor village was available, we used the geocenter of the host township, an alternative which we believed was preferable to dropping the source altogether. We dropped 6 sources for lack of location information. Our final dataset thus contains data for 641 major sources of pollution in Jiangsu province.

3.3. Spatial Units of Analysis

China has five levels of administrative division: 1) province level (34); 2) prefecture level (333); 3) county level (2,862); 4) township level (41,363); 5) village level (703,786).2 We used the township as our spatial unit of analysis, because it is the smallest administrative unit for which detailed socioeconomic data are widely available. Data for Jiangsu townships were collected from the China Census 2000 (CC 2000), the Statistical Yearbook of Jiangsu 2000 (SYJ 2000), and annual reports of local governments. For Jiangsu province, data were available for 1802 township units, including urban (1446), rural (272) and virtual units (84)3. We use only urban townships as units of analysis, because, as can be seen in Figure 1, which depicts township boundaries and pollution sources in Jiangsu, there is little doubt that rural areas are exposed to much less industrial pollution than urban areas. The more interesting question, and the one which motivates our study, is whether all urban areas are equally likely to contain, or to be near to, major sources of pollution—and, by extension, whether all individuals in urban areas are equally likely to suffer from pollution’s ill effects.

1 COD, or “chemical oxygen demand,” is a widely-used measure of water pollution by organic compunds from wastewater treatment and manufacturing.

2 The numbers in parentheses mark the total units nationally at each level as of the end of 2009.

3 Urban townships refer to “Zhen” and “Jie Dao”. Rural townships refer to “Xiang”. Some areas (and associated population) are not classified in the formal Nation – Province – Prefecture – County – Township – Village administrative division. For statistical purpose, they are named as “Xu Ni” townships (i.e. virtual units).

8 [Figure 1 about here]

3.4. Dependent Variables4

As explained above, we use three different kinds of dependent variable in order to capture three different ways of measuring pollution at the township level. First, we use the number of pollution sources located within the containment area of each township. Second, we use the per capita amount, in kilograms per resident, of each of four major pollutants (SO2, smog, NH3-N, and COD) that is emitted within each containment area. Third, we again use the per capita amount of each pollutant emitted, but first weight the contribution of each facility to the township total by its distance to the geocenter of the township in question. Thus, we use nine dependent variables overall. All pollution data come from the 2005 Jiangsu EPB dataset.

3.5. Independent Variables

3.5.1. Independent Variable of Main Interest

Percentage of Rural Migrants in Township Population (MIGRANTS). China Census 2000 reports the total migrant population of townships but does not differentiate between migrants from rural areas and the much small number of migrants from other urban areas, who may be changing cities due to job relocation. In order to obtain a more accurate estimate of the number of rural migrants, we subtract the number of migrants in each township with an education level of college or above from the total migrant population.

3.5.2. Control Variables

Percentage of employment in “dirty and hard” industries (INDUSTRY). This variable captures the impact of industrial structure on township pollution by describing the percentage of township residents employed in mining, manufacturing, and electricity generation. By controlling for employment in highly polluting industries, we are able to test whether, if rural migrants are indeed associated with increased pollution, the effect is simply due to the fact that migrants who work in pollution-producing facilities are also likely to live near them.

Percentage of Ethnic Minorities in Township Population (MINORITY). The Chinese government recognizes 55 official ethnic minorities, but together they make up only about 8 percent of the population. More importantly, racial tension is far less of an issue in China than in the U.S. (Quan 2002). Nevertheless, we include the percentage of racial minorities in each township as a control in order to compare its effect with the theoretically much more important effect of race on environmental inequality in the U.S.

Log of County-Level Per Capita Disposable Income (INCOME). Data on income at the township level was not available. Instead, we use the log of urban per capita disposable income at the county level as an indicator of income for all urban townships within each county. A recent study suggests that the majority of income inequality at the provincial level is primarily due to inequality between counties rather than within counties (Gustafsson and Li 2002). Thus, while not ideal, we believe that using county-level per capita income as an indicator of township-level income should not pose a serious problem to the analysis.

4 Sources and compressed definitions of all variables are summarized in the Data Appendix.

9

There also exists the potential for an endogeneity issue between the income control variable and the dependent variables (the pollution indicators). A substantial portion of pollution derives from industrial production, which also contributes to local income. Thus, it may also be that income is higher because of the presence of more industrial activity (and therefore of more pollution sources). In order to address the endogeneity issue, the current income is instrumented with the one-year- and two-year-lagged income. Although current income might be endogenous to the current level of pollution, it is less likely that past income is subject to the same problem. In all regressions we have performed, the weak instrument hypothesis is clearly rejected.

Log of Distance to the Nearest River (RIVER). Township proximity to a main river may be an important factor in determining facility location, because many industries require access to freshwater for cooling purposes and waste disposal. The log of the distance between the township’s center and the nearest river is intended to control for this effect.

Other Township-Level Controls. We also control for the population density of each township (DENSITY), measured in thousands of persons per square kilometer, the percentage of residents with a high-school education (EDUCATION), and the percentage of working-age (15-64) persons (LABOR) in the population.

Regional-Level Variation (PREFECTURE). Many studies of environmental inequality point to the need to account for regional differences and suggest the introduction of a control for such factors (Bowen et al. 1995; Yandle and Burton 1996; Bowen 2001; Pastor, Sadd, and Morello-Frosch, 2004). The thirteen prefectures of Jiangsu province are certainly ripe for such variation. Province-level planners may implement different industrial location policies in different prefectures. Officials may be more aggressive in promoting environmental protection in some prefectures than in others. Differences may also exist in resource availability (water, energy, or labor) and production across prefectures. We generate a dummy variable for each of Jiangsu’s thirteen prefectures5 in order to account for regional characteristics which are not captured by other independent variables. Standard procedure with dummy variables is to omit one region as a reference category; however, this practice is less informative when little knowledge exists of the selected reference (Haisken-DeNew and Schmidt 1997; Kennedy 1986; Suits 1984). The approach taken here follows Kennedy (1986) and uses the provincial average as the reference category. The coefficient of a regional dummy thus has a straightforward interpretation: the extent to which each region is more or less pollution intensive than the provincial average, ceteris paribus.

Table 1 reports summary statistics for all variables. The number of pollution sources per township includes all sources of pollution that are located within each township’s containment area. Per capita emissions of each pollutant are calculated for the same containment areas.

[Table 1 about here]

3.6. Model and Hypothesis

5 Nanjing (Capital of Jiangsu Province), Wuxi, Xuzhou, Changzhou, Suzhou, Nantong, Liangyungang, Huaiyin, Yancheng, Yangzhou, Zhenjiang, Taizhou, and Suqian.

10

The basic model can be specified as follows:

POLLUTIONi = 0 + 1 MIGRANTSi + 2 INDUSTRYi + 3 MINORITYi + 4 INCOMEi + 5 DENSITYi + 6 EDUCATIONi + 7 LABORi + 8 RIVERi +  j PREFECTUREij + i

For each dependent variable we experiment with three different models, all of which are variations on the basic model specified above. In each case, Model 1 and Model 2 are both two-stage least squares regression (2SLS)—that is, they include the one-year and two-year lagged county income as instrumental variables—but only Model 2 includes the INDUSTRY control variable. Model 3 is an ordinary least squares regression (OLS) that, like Model 2, includes all controls.

Our primary hypothesis is that pollution in urban townships is positively associated with the percentage of township residents who are rural migrants—that is, with the percentage of residents who, despite working in an urban area, have their official residence, or hukou, in a rural area. We expect that, due to the social mechanisms outlined above, the effect of rural migrants on township pollution will remain, even after controlling for confounding factors like the role of heavy industry in a township’s economic base.

4. Results

Table 2 gives the results of regressions of the number of pollution sources. Table 3 and Table 4 provide the results of regressions of actual emissions. Results of regressions of weighted emissions are shown in Table 5 and Table 6. For each dependent variable, there is little difference between the results of Models 2 and 3, suggesting that the possibility of an endogenous relationship between the dependent variable (essentially, industrial pollution) and a key independent variable (income) does not pose a serious problem. Moreover, there are no major differences between the unweighted and weighted results for any given pollutant. All significance tests are two-tailed.

Overall, the analyses provide support for our primary hypothesis that rural migrants in Jiangsu province are disproportionately exposed to industrial pollution in urban areas. The effect of the percentage of rural migrants in the population—those who have their hukou in rural areas but who have migrated to the city in search of work—on township pollution levels is positive and highly significant across all dependent variables, with the exception of SO2. Townships with a higher percentage of rural migrants are close to more major sources of pollution, and are exposed to higher levels of smog, organic pollutants (as measured by COD), and NH3-N, even when other factors are taken into account. In particular, the effect of the percentage of rural migrants does not lose significance—though it tends to decrease in magnitude—when a key confounding variable, the percent of township residents employed in dirty, hard industries, is added to the regression model.

In the full model (used for Models 2 and 3, the former being a two-stage least squares regression and the latter an ordinary least squares regression) the magnitude of the effect of rural migrants on township pollution is comparable to that of reliance on dirty, hard industries. For instance, each additional percentage point of rural migrants in the population is associated with .119 more major sources of pollution (p < .01, Table 1) and 2.284 more kilograms of smog per capita (p < .01, Table 5). Similarly, each additional percentage point of township residents employed in mining, manufacturing, and electricity generation is

11 associated with .106 more sources of pollution (p < .01, Table 1) and 3.137 more kilograms of smog per capita (p < .01, Table 5). As we discuss in greater detail below, the fact that the effect of rural migrants on township pollution remains significant and substantial, even when employment in dirty industry is included in the regression models, suggests that migrants are not disproportionately exposed to pollution simply because they tend to work in the dirtiest industries.

The effects of control variables differ across the four pollutants but are largely consistent across the three models for each pollutant. The effect of population density, though generally significant, is very small. For the actual and weighted emissions results, the effect of income is positive and very large for SO2 and smog, but—once all variables are included—the effect of income is not significant for COD or for the weighted NH3-N, and it is significant but of modest size for the unweighted NH3-N. The uneven impact of income could be a residue of the fact that precise income statistics for townships were not available, and so we used county income in its place. The percentage of highly educated residents is significant only for COD, where its effect is approximately equal to that of migrants.

[Table 2 about here] [Table 3 about here] [Table 4 about here] [Table 5 about here] [Table 6 about here]

5. Discussion and Conclusion

We have argued that environmental inequality is not a contingent effect of industrialization, but rather a predictable consequence of the way that pre-industrial social class structures, like Weber’s cultural switchman, channel the environmental consequences of development toward the most vulnerable populations. We have also suggested that social forces leading to race-based environmental inequality in the U.S. provide a template for thinking about how status distinctions embedded in other cultural frameworks may operate to perpetuate existing social hierarchies into the environmental realm. The results of our analyses, by illuminating the web of social factors affecting the distribution of pollution in one of China’s most dynamic provinces, provide support for the idea that, with respect to pollution exposure, the state-sponsored hukou system is playing the role that racial classification has historically played in the American context.

The culture, institutions, and political traditions of China could hardly be more different from those of the U.S. Yet given the results of our analyses, it is not difficult to see how people with an official “rural” designation could, over time, suffer an environmental fate analogous to that of African-Americans and other racial minorities. The most intriguing finding of this study is that a relationship between pollution and rural migrant populations in Jiangsu exists even when employment in heavy, dirty industry is taken into account. The persistence of this relationship indicates that rural migrants and high levels of pollution do not hang together simply because of where the former are likely to work. Rather, we would argue that factors are in play which a general theory of environmental inequality, centered on social mechanisms implicated in American race-based environmental inequality, readily suggests. Most members of the “floating population” likely can only afford to live in the most polluted areas of modern Chinese cities, even when they do not actually work in the industries causing the pollution. As happened with African-Americans, social prejudice may

12 be keeping cleaner, healthier, more desirable Chinese communities closed to people bearing the stigma of a rural hukou. The construction of new factories, power plants, and waste treatment facilities may be being taking place largely in areas occupied by low-status social groups partly because the land is less expensive, but also because the political opposition is less viable. Any of these factors, or all of them, could be in play, as well as social processes unique to the Chinese context. As members of China’s burgeoning middle class use their newfound wealth, and newly bought cars, to move away from pollution, members of the floating population are likely have little choice but to live with it and even move to it—when pollution does not come to them first. Indeed, the results of our study suggest that this dynamic has already taken root in Jiangsu.

The first reports on environmental inequality in the U.S. were published over thirty years ago. But research into environmental inequality in China, let alone in other parts of the developing world, is in its early stages. To the best of our knowledge, our study provides the first systematic evidence that Chinese communities with large populations of rural migrants are disproportionately exposed to industrial pollution, even after taking into account possible confounding factors. But future investigators will find many ways to extend and improve upon our methods, and to subject both our ideas and our results to rigorous test. New data, such as China’s in-process National Inventory of Pollution Sources, will allow investigators to look for nationwide variation in how exposure to pollution is distributed across social groups, and to probe whether socioeconomic characteristics of cities and provinces influence the extent to which environmental inequality arises. Sophisticated models of how pollution disperses into the atmosphere and waterways may open the door to “exposure/risk-based” studies of how pollutants from non-point sources are distributed across China. Future census data may contain finer measurements of key socioeconomic statistics: in particular, data on income at the township level would enable researchers to more precisely assess the effect of a key explanatory variable. Finally, and perhaps most importantly, we hope that penetrating case studies of specific Chinese communities will begin to elaborate in greater detail the confluence of social mechanisms driving the inequalities suggested by our study.

Peter Hessler, a former Peace Corps volunteer and trenchant observer of modern China, recently wrote that “to drive across China was to find yourself in the middle of the largest migration in human history—nearly one-tenth of the population was on the road, finding new lives away from home” (2010:33). As China continues to grow and change, the need will only increase for research into how the most vulnerable members of Chinese society are coping with the epochal transformation of their nation. Researchers have a role to play, not only in chronicling the present, but also, by exposing inequality and injustice to light, in helping to change the future for the better.

References

Anderton, Douglas L., Andy B. Anderson, John Michael Oakes, and Michael R. Fraser. 1994. “Environmental Equity: The Demographics of Dumping.” Demography 31:229-48. Anderton, Douglas L., John Michael Oakes, and Karla L. Egan. 1997. “Environmental Equity in Superfund: Demographics of the Discovery and Prioritization of Abandoned Toxic Sites.” Evaluation Review 21(1):2-26. Been, Vicki and Francis Gupta. 1997. “Coming to the Nuisance or Going to the Barrios? A Longitudinal Analysis of Environmental Justice Claims.” Ecology Law Quarterly 24:1-56.

13 Bowen, William M. 2001. Environmental Justice through Research-Based Decision-Making. New York: Garland Publishing ———. 2002. “An Analytical Review of Environmental Justice Research: What Do We Really Know?” Environmental Management 29(1):3-15. Bowen, William M., Mark J. Salling, Kingsley E. Haynes, and Ellen J. Cyran. 1995. “Toward Environmental Justice: Spatial Equity in Ohio and Cleveland.” Annals of the Association of American Geographers 85(4):641-63 Brown, Phil. 1995. “Race, Class, and Environmental Health: A Review and Systemization of the Literature.” Environmental Research 69:15-30 Bryant, Bunyan and Paul Mohai, eds. 1992. Race and the Incidence of Environmental Hazards: A Time for Discourse. Boulder, CO: Westview Press. Bullard, Robert D. 2000. Dumping in Dixie: Race, Class, and Environmental Quality. Third Edition. Boulder, CO: Westview Press. Campanella, Thomas J. 2008. The Concrete Dragon: China’s Urban Revolution and What It Means for the World. Princeton, NJ: Princeton Architectural Press. Chan, Kam Wing and Li Zhang. 1999. “The Hukou System and Rural-Urban Migration in China: Processes and Changes.” The China Quarterly 160:818-855. Checker, Melissa. 2005. Polluted Promises: Environmental Racism and the Search for Justice in a Southern Town. New York: NYU Press. Cheng, Tiejun and Mark Selden. 1994. “The Origins and Social Consequences of China’s Hukou System.” The China Quarterly 139:644-668. China Census 2000. China Data Center. University of Michigan. CRJ/UCC (Commission for Racial Justice, United Church of Christ). 1987. Toxic Wastes and Race in the United States: A National Report on the Racial and Socioeconomic Characteristics of Communities with Hazardous Wastes Sites. New York: Public Data Access. Daniels, Glynis and Samantha Friedman. 1999. “Spatial Inequality and the Distribution of Industrial Toxic Releases: Evidence from the 1990 TRI.” Social Science Quarterly 80: 244-62. Davidson, Pamela and Douglas L. Anderton. 2000. “Demographics of Dumping II: A National Environmental Equity Survey and the Distribution of Hazardous Materials Handlers.” Demography 37(4):461-466. Downey, Liam. 2006. “Using Geographic Information Systems to Reconceptualize Spatial Relationships and Ecological Context.” American Journal of Sociology 112: 567-612. Du, Yang, Robert Gregory, and Xin Meng. 2006. “The Impact of the Guest-Worker System on Poverty and the Well-Being of Migrant Workers in Urban China,” in Ross Garnaut and Gigang Song (eds.), The Turning Point in China’s Economic Development. Canberra: Asia Pacific Press at the Australian National University. Fan, C. Cindy. 2001. “Migration and Labor-Market Returns in Urban China: Results from a Recent Survey in Guangzhou.” Environment and Planning A 33(3):479-508. Fishman, Robert. 1987. Bourgeois Utopias: The Rise and Fall of Suburbia. New York: Basic Books. Fogelson, Robert M. 2001. Downtown: Its Rise and Fall, 1880-1950. New Haven, CT: Yale University Press. ———. 2005. Bourgeois Nightmares: Suburbia, 1870-1930. New Haven, CT: Yale University Press. General Accounting Office (U.S.). 1983. “Siting of Hazardous Waste Landfills and Their Correlation with Racial & Economic Status of Surrounding Communities.” Gaithersburg, MD: U.S. General Accounting Office.

14 Goldman, Benjamin A. and Laura Fitton. 1994. Toxic Wastes and Class Revisited: An Update of the 1987 Report on the Racial and Socioeconomic Characteristics of Communities with Hazardous Waste Sites. Washington, D.C.: Center for Policy Alternatives. Greenberg, Michael. 1993. “Proving Environmental Inequity in Siting Locally Unwanted Land Uses.” Risk: Issues in Health and Safety 4:235-245. Guftasson, Björn and Li Shi. 2002. “Income Inequality Within and Across Counties in Rural China in 1988 and 1995.” Journal of Development Economics 69(1):179-204 Haisken-DeNew, John P. and Christoph M. Schmidt. 1997. “Inter-Industry and Inter-Region Differentials: Mechanics and Interpretation.” Review of Economics and Statistics 79:516-521. Hamilton, James T. 1995. “Testing for Environmental Racism: Prejudice, Profits, Political Power?” Journal of Policy Analysis and Management 14(1):107-132. Hessler, Peter. 2010. Country Driving: A Journey Through China from Farm to Factory. New York: Harper. Hird, John A. 1993. “Environmental Policy and Equity: The Case of Superfund.” Journal of Policy Analysis and Management 12(2):323-43. Hird, John A. and Michael Reese. 1998. “The Distribution of Environmental Quality: An Empirical Analysis.” Social Science Quarterly 79(4):693-716. Huang, Youquin. 2001. “Gender, Hukou, and the Occupational Attainment of Female Migrants in China (1985-1990).” Environment and Planning A 33(2):257-279. Hurley, Andrew. 1995. Environmental Inequalities: Race, Class, and Industrial Pollution in Gary, Indiana, 1945-1980. Chapel Hill: University of North Carolina Press. Jackson, Kenneth T. 1985. Crabgrass Frontier: The Suburbanization of America. New York: Oxford University Press. Jian, Ke. 2005. “Environmental Justice: Can an American Discourse Make Sense in Chinese Environmental Law?” Temple Journal of Science, Technology and Environmental Law 24:253-285. Kennnedy, Peter. 1986. “Interpreting Dummy Variables.” Review of Economics and Statistics 68:174-175. Knight, John and Lina Song. 1999. The Rural-Urban Divide: Economic Disparities and Interactions in China. Oxford: Oxford University Press. Knup, Elizabeth. 1997. “Environmental NGOs in China: an Overview.” China Environment Series 1:9-15 Lerner, Steve. 2005. Diamond: A Struggle for Environmental Justice in Louisiana’s Chemical Corridor. Cambridge, MA: MIT Press. Lester, James P., David W. Allen, and Kelly M. Hill. 2001. Environmental Injustice in the United States: Myths and Realities. Boulder, CO: Westview Press. Liang, Zai and Zhondong Ma. 2004. “China’s Floating Population: New Evidence from the 2000 Census.” Population and Development Review 30(3):467-488. Liu, Zhiqiang. 2005. “Institution and Inequality: The Hukou System in China.” Journal of Comparative Economics 33(1):133-157. Ma, Chunbo, and Ethan D. Schoolman. 2009. “Understanding Environmental Inequality in China: Is New Theory Needed?” Presented at the Annual Meeting of the Society for the Study of Social Problems, August 7, San Francisco, CA. Ma, Chunbo, and Ethan D. Schoolman. 2010. “Who Bears the Environmental Burden in China? An Analysis of the Distribution of Industrial Pollution Sources.” Ecological Economics 69:1869-1876; non-substantive corrigendum forthcoming. Mohai, Paul and Robin Saha. 2006. “Reassessing Racial and Socioeconomic Disparities in Environmental Justice Research.” Demography 43(2):383-399

15 ———. 2007. “Racial Inequity in the Distribution of Hazardous Waste: A National-level Reassessment.” Social Problems 54(3):343-370 Morello-Frosch, Rachel and Bill Jesdale. 2006. “Separate and Unequal: Residential Segregation and Estimated Cancer Risk Associated with Ambient Air Toxics in U.S. Metropolitan Areas.” Environmental Health Perspectives 114(3):386-393. Morello-Frosch, Rachel, Manuel Pastor, Jr., and James Sadd. 2001. “Environmental Justice and Southern California’s ‘Riskscape’: The Distribution of Air Toxics Exposures and Health Risks Among Diverse Communities.” Urban Affair Review 36(4):551-578 Oakes, John Michael, Douglas L. Anderton, and Andy B. Anderson. 1996. “A Longitudinal Analysis of Environmental Equity in Communities with Hazardous Waste Facilities.” Social Science Research 23:125-148. Palmer, Michael. 2007. “Towards a Greener China? Accessing Environmental Justice in the People’s Republic of China,” in Andrew Harding (ed.), Access to Environmental Justice: a Comparative Study, pp. 205-235. Leiden: Martinus Nijhoff. Pan, Philip P. 2001. “Cancer-stricken Chinese Village Tries to Pierce a Wall of Silence.” Washington Post, Nov. 5, A19. Pastor, Jr., Manual, James L. Sadd, and Rachel Morello-Frosch. 2004. “Waiting to Inhale: The Demographics of Toxic Air Release in 21st Century California.” Social Science Quarterly 85(2):420-440 Pastor, Jr., Manual, Rachel Morello-Frosch, and James L. Sadd. 2005. “The Air is Always Cleaner on the Other Side: Race, Space, and Ambient Air Toxics Exposures in California.” Journal of Urban Affairs 27(2):127-148. Perlin, Susan A., R. Woodrow Setzer, John Creason, and Ken Sexton. 1995. “Distribution of Industrial Air Emissions By Income and Race in the United States: An Approach Using the Toxic Release Inventory.” Environmental Science & Technology 29:69–80 Quan, Ruixue. 2002. “Establishing China’s Environmental Justice Study Models.” Georgetown International Environmental Law Review 14(3):461-487 Ringquist, Evan A. 1997. “Equity and the Distribution of Environmental Risk: The Case of TRI Facilities.” Social Science Quarterly 78:811-829. Suits, Daniel B. 1984. “Dummy Variables, Mechanics v. Interpretation.” Review of Economics and Statistics 66:177-180. Szasz, Andrew and Michael Meuser. 1997. “Environmental Inequalities: Literature Review and Proposals for New Directions in Research and Theory.” Current Sociology 45(3):99-120. ———. 2000. “Unintended, Inexorable: The Production of Environmental Inequalities in Santa Clara County, California.” American Behavioral Scientist 43:602-632 Wang, Fei-Ling. 2005. Organizing Through Division and Exclusion: China’s Hukou System. Stanford, California: Stanford University Press. Wu, Xiaogang and Donald J. Treiman. 2004. “The Household Registration System and Social Stratification in China: 1955-1996.” Demography 41(2):363-384. Yandle, Tracy and Dudley Burton. 1996. “Reexamining Environmental Justice: A Statistical Analysis of Historical Hazardous Waste Landfill Siting Patterns in Metropolitan Texas.” Social Science Quarterly 77:477-492 Zeng, Nina, Yihui Ding, Jiahua Pan, Huijun Wang, and Jay Gregg. 2008. “Sustainable Development: Climate Change—The Chinese Challenge.” Science 319:730-731. Zhu, Yu. 2007. “China’s Floating Population and Their Settlement Intention in the Cities: Beyond the Hukou Reform.” Habitat International 31(1):65-76. Zimmerman, Rae. 1993. “Social Equity and Environmental Risk.” Risk Analysis 13:649-666.

16

Figure 1. Pollution Sources in Jiangsu Province

Data sources: CC 2000; authors’ collection.

17 Table 1. Summary Statistics

Mean Std. Dev. Min Max

Dependent Variables* Number of pollution sources 5.827 6.886 0.0 36.0

** Per capita emission of SO2 305.5 1097.7 0.0 26575.7 Per capita emission of SMOG 77.0 241.5 0.0 5110.4 Per capita emission of COD 40.7 90.7 0.0 1329.9

Per capita emission of NH3-N 4.0 9.0 0.0 120.2 Independent Variables Percentage of rural migrants 13.3 12.1 0.4 68.5 Percentage of employment in “dirty & hard” industry*** 25.3 19.1 0.1 90.6 Percentage of minorities 0.4 0.8 0.0 21.5 Per capita disposable income (Yuan) 9.3 0.3 8.6 9.8 Population density (1,000 person/km2) 4390.6 10845.2 34.6 84167.3 Percentage of senior high-school educated 17.2 13.8 3.9 71.9 Percentage of population of age 15 - 64 71.7 4.5 59.1 87.5 Distance to nearest river 8.0 1.3 0.1 10.4

*For containment areas with a buffer width of 5 miles around the borders of each township. **All per capita emissions statistics are in kilograms per resident. ***Mining, Manufacturing and Power Production. Data Sources: CC 2000; Annual reports of local governments, 2004-2005.

18 Table 2. Regression Results for Number of Pollution Sources

Dependent: Number of Pollution Sources Model 1: 2SLS Model 2: 2SLS Model 3: OLS Independent: Coef. Z Coef. Z Coef. Z MIGRANTS 0.166** 9.02 0.119** 6.42 0.119** 6.38 INDUSTRY 0.109** 9.59 0.106** 9.36 MINORITY -0.137 -0.84 -0.182 -1.15 -0.176 -1.11 INCOME 1.309 1.14 -2.344* -2.00 -1.489 -1.36 DENSITY -0.00001 -0.70 -2.17e-06 -0.15 -2.96e-06 -0.20 EDUCATION 0.067** 4.03 0.086** 5.29 0.085** 5.19 LABOR 0.119* 1.99 -0.015 -0.25 -0.022 -0.36 RIVER -0.281** -2.89 -0.214* -2.26 -0.220* -2.31 Dummies NANJING 0.5 1.07 2.0** 3.95 1.8** 3.59 WUXI 7.3** 12.77 7.0** 12.60 6.7** 12.40 XUZHOU -1.2** -2.61 -2.2** -4.60 -1.9** -4.22 CHANGZHOU 4.5** 7.65 5.0** 8.64 4.7** 8.35 SUZHOU 1.0 1.61 1.2 1.90 0.8 1.39 NANTONG -0.7 -1.82 -0.3 -0.72 -0.4 -0.99 LIANYUNGANG -2.1** -3.18 -3.0** -4.55 -2.7** -4.21 HUAIYIN -0.8 -1.39 -0.6 -1.17 -0.4 -0.82 YANCHENG 2.9** -6.49 -2.4** -5.39 -2.2** -5.09 YANGZHOU -1.0* -2.20 -1.5** -3.23 -1.4** -3.09 ZHENJIANG -1.0 -1.94 -0.8 -1.71 -1.0* -1.96 TAIZHOU -1.2** -2.71 -1.0* -2.31 -1.1* -2.36 SUQIAN -0.6 -0.76 -1.9* -2.54 -1.5* -2.06 Constant -16.1 -1.57 24.39* 2.27 17.1 1.68 N of Obs. 1446 1446 1446 R-Squared 0.5797 0.6039 0.6040 Wald chi2 1993.56** 2206.95** F-Statistic 108.69** Durbin chi2 0.3321 3.867* Wu-Hausman 0.3273 3.818 *significant at p < 0.05; **significant at p < 0.01; all tests are two-tailed

19 Table 3. Regression Results for SO2 and SMOG Emissions Dependent: SO2 Emission SMOG Emission Model 1: 2SLS Model 2: 2SLS Model 3: OLS Model 1: 2SLS Model 2: 2SLS Model 3: OLS Independent: Coef. Z Coef. Z Coef. t Coef. Z Coef. Z Coef. t MIGRANTS 10.934** 4.17 2.929 1.12 2.926 1.11 3.188** 4.85 1.708* 2.56 1.708* 2.54 INDUSTRY 18.127** 11.28 18.185** 11.30 3.353** 8.14 3.340** 8.11 MINORITY 12.629 0.28 6.403 0.15 6.281 0.15 -2.694 -0.24 -3.845 -0.35 -3.819 -0.35 INCOME 1278.1** 7.84 687.0** 4.18 668.1** 4.35 320.0** 7.82 210.6** 5.01 214.7** 5.45 DENSITY -0.010** -4.77 -0.009** -4.33 -0.009** -4.29 -0.003** -4.63 -0.002** -4.26 -0.002** -4.24 EDUCATION 0.927 0.39 4.007 1.74 4.033 1.74 0.243 0.40 0.813 1.38 0.807 1.36 LABOR 7.589 0.89 -13.992 -1.66 -13.852 -1.64 0.420 0.20 -3.572 -1.66 -3.602 -1.66 RIVER -21.622 -1.56 -10.587 -0.80 -10.450 -0.78 -10.058** -2.90 -8.017* -2.36 -8.047* -2.35 Dummies NANJING -262.4** -3.68 -28.6 -0.38 -24.0 -0.32 -57.3** -3.04 -14.7 -0.80 -13.9 -0.73 WUXI -623.8** -7.68 -680.1** -8.72 -674.3** -8.86 -148.9** -7.31 -159.4** -7.98 -160.6** -8.23 XUZHOU 715.8** 10.48 575.1** 8.65 570.2** 8.77 155.8** 9.09 129.8** 7.62 130.9** 7.86 CHANGZHOU -413.6** -4.96 -345.7** -4.32 -340.3** -4.33 -114.3** -5.47 -101.7** -4.97 -102.9** -5.11 SUZHOU -732.5** -7.96 -718.1** -8.15 -710.4** -8.34 -191.7** -8.31 -189.1** -8.38 -190.7** -8.74 NANTONG -79.7 -1.36 -4.8 -0.09 -2.4 -0.04 -25.7 -1.75 -11.9 -0.82 -12.4 -0.86 LIANYUNGANG 466.2** 4.92 332.4** 3.64 326.6** 3.62 125.9** 5.30 101.2** 4.32 102.5** 4.44 HUAIYIN 362.3** 4.64 390.1** 5.20 386.0** 5.19 149.5** 7.61 154.5** 8.04 155.4** 8.16 YANCHENG 80.1 1.25 170.9** 2.77 167.6** 2.74 20.6 1.29 37.4* 2.37 38.1* 2.43 YANGZHOU 66.1 0.97 -11.5 -0.18 -12.7 -0.19 -9.2 -0.54 -23.5 -1.41 -23.2 -1.38 ZHENJIANG -174.9* -2.41 -156.7 -2.25 -154.0* -2.21 -43.4* -2.38 -40.0* -2.24 -40.6* -2.28 TAIZHOU -135.7* -2.09 -101.3 -1.63 -100.7 -1.61 -42.4** -2.60 -36.1* -2.26 -36.2* -2.25 SUQIAN 602.2** 5.67 -396.6** 3.85 387.9** 3.88 144.9** 5.45 106.9** 4.05 108.8** 4.25 Constant -12030.5** -8.23 -5501.2** -3.64 -5340.3** -3.74 -2868.4** -7.82 -1660.8** -4.29 -1695.8** -4.63 N of Obs. 1440 1440 1440 1440 1440 1440 R-Squared 0.1652 0.2344 0.2344 0.1711 0.2078 0.2078 Wald chi2 287.55** 439.20** 293.74** 372.67** F-Statistic 21.72** 18.61** Durbin chi2 3.224 0.096 0.715 0.069 Wu-Hausman 3.184 0.094 0.705 0.068 *significant at p < 0.05; **significant at p < 0.01; all tests are two-tailed

20 Table 4. Regression Results for COD and NH3-N Emissions Dependent: COD Emission NH3-N Emission Model 1: 2SLS Model 2: 2SLS Model 3: OLS Model 1: 2SLS Model 2: 2SLS Model 3: OLS Independent: Coef. Z Coef. Z Coef. t Coef. Z Coef. Z Coef. t MIGRANTS 1.814** 5.42 1.303** 3.82 1.263** 3.66 0.147** 4.27 0.117** 3.29 0.117** 3.26 INDUSTRY 1.275** 6.11 1.304** 6.22 0.069** 3.17 0.072** 3.29 MINORITY -8.251** -2.79 -28.96** -5.17 -8.856** -3.01 -1.554** -2.64 -1.575** -2.69 -1.581** -2.68 INCOME 70.620** 3.40 25.515 1.19 17.747 0.88 7.861** 3.67 5.539** 2.46 4.655* 2.21 DENSITY -0.001** -4.24 -0.001** -3.51 -0.001** -3.86 -0.0001** -2.91 -0.0001** -2.73 -0.0001** -2.68 EDUCATION 1.117** 3.70 1.544** 5.10 1.353** 4.48 0.049 1.54 0.061 1.92 0.062 1.94 LABOR -1.313 -1.20 -2.895** -2.63 -2.806* -2.52 -0.160 -1.43 -0.246* -2.13 -0.239* -2.06 RIVER -1.588 -0.90 -1.007 -0.58 -0.728 -0.41 -0.319* -1.75 -0.277 -1.52 -0.270 -1.47 Dummies NANJING -1.4 -0.15 15.9 1.61 17.8 1.83 -1.2 -1.17 -0.2 -0.21 -0.004 -0.00 WUXI 5.5 0.53 2.1 0.20 4.5 0.45 -2.7* -2.49 -2.8** -2.67 -2.6* -2.46 XUZHOU 12.7 1.48 1.9 0.21 -0.2 -0.02 -0.4 -0.48 -1.0 -1.12 -1.2 -1.41 CHANGZHOU 51.0** 4.81 56.2** 5.36 58.4** 5.67 4.2** 3.84 4.5** 4.10 4.7** 4.41 SUZHOU -47.5** -4.07 -45.7** -3.97 -42.5** -3.82 -2.9* -2.39 -2.8* -2.31 -2.4* -2.09 NANTONG -8.7 -1.16 -3.3 -0.44 -2.3 -0.31 -1.4 -1.86 -1.1 -1.47 -1.0 -1.33 LIANYUNGANG -9.2 -0.76 -19.1 -1.60 -21.5 -1.82 -0.4 -0.30 -0.9 -0.73 -1.2 -0.96 HUAIYIN 10.4 1.04 12.0 1.22 10.3 1.06 1.5 1.44 1.6 1.53 1.4 1.35 YANCHENG -7.9 -0.97 -1.5 -0.19 -2.9 -0.36 -0.1 -0.09 0.3 0.31 0.1 0.13 YANGZHOU 4.8 0.55 -0.6 -0.07 -1.1 -0.13 6.3** 7.01 6.0** 6.68 5.9** 6.58 ZHENJIANG 5.0 0.54 6.6 0.73 7.7 0.85 -1.9* -2.04 -1.9 -1.95 -1.7 -1.82 TAIZHOU -11.2 -1.35 -8.7 -1.06 -8.4 -1.03 -0.9 -1.03 -0.7 -0.87 -0.7 -0.84 SUQIAN 21.2 1.58 6.0 0.44 2.4 0.18 2.8* 2.04 2.0 1.43 1.6 1.17 Constant -515.8** -2.79 -42.1 -0.21 24.4 0.13 -56.6** -2.96 -30.9 -1.50 -23.4 -1.19 N of Obs. 1442 1442 1442 1442 1442 1442 R-Squared 0.2105 0.2316 0.2209 0.1422 0.1488 0.1489 Wald chi2 387.84** 435.38** 241.72** 253.42** F-Statistic 21.42** 12.43** Durbin chi2 3.409 0.953 2.324 1.128 Wu-Hausman 3.368 0.939 2.294 1.111 *significant at p < 0.05; **significant at p < 0.01; all tests are two-tailed

21 Table 5. Regression Results for Weighted SO2 and SMOG Emissions Dependent: Weighted SO2 Emission Weighted SMOG Emission Model 1: 2SLS Model 2: 2SLS Model 3: OLS Model 1: 2SLS Model 2: 2SLS Model 3: OLS Independent: Coef. Z Coef. Z Coef. t Coef. Z Coef. Z Coef. t MIGRANTS 11.518** 3.34 3.622 1.03 3.633 1.03 3.686** 4.53 2.282** 2.73 2.284** 2.71 INDUSTRY 17.855** 8.25 17.636** 8.15 3.178** 6.15 3.137** 6.07 MINORITY -12.102 -0.21 -18.780 -0.33 -18.301 -0.32 2.216 0.31 3.364 0.25 3.453 0.25 INCOME 1269.2** 5.94 684.020** 3.11 754.9** 3.66 296.0** 5.86 191.7** 3.64 204.9** 4.16 DENSITY -0.014** -5.00 -0.013** -4.64 -0.013** -4.63 -0.003** -5.05 -0.003** -4.72 -0.003** -4.71 EDUCATION -3.059 -0.97 0.001 0.00 -0.095 -0.03 -0.535 -0.73 -0.012 -0.02 -0.030 -0.04 LABOR 28.338* 2.52 7.258 0.64 6.723 0.59 4.000 1.50 0.229 0.08 0.129 0.05 RIVER -28.473 -1.57 -17.507 -0.98 -18.026 -1.00 -13.396** -3.12 -11.451** -2.69 -11.548** -2.69 Dummies NANJING -163.4 -1.65 73.7 0.73 56.3 0.57 -36.3 -1.55 5.9 0.24 2.6 0.11 WUXI -591.7** -5.55 -646.8** -6.19 -668.6** -6.54 -143.5** -5.69 -153.3** -6.14 -157.4** -6.44 XUZHOU 769.9** 8.63 629.8** 7.09 648.4** 7.47 156.2** 7.40 131.3** 6.19 134.8** 6.49 CHANGZHOU -489.6** -4.48 -422.2** -3.94 -442.5** -4.20 -125.1 -4.83 -113.1** -4.41 -116.9** -4.64 SUZHOU -796.1** -6.60 -781.3** -6.62 -810.1** -7.10 -189.5** -6.63 -186.8** -6.63 -192.2** -7.05 NANTONG -208.5** -2.70 -130.5 -1.72 -139.7 -1.84 -52.1** -2.85 -38.2* -2.10 -39.9* -2.20 LIANYUNGANG 630.7** 5.07 498.2** 4.06 520.0** 4.30 157.3** 5.34 133.7** 4.56 137.8** 4.77 HUAIYIN 377.6** 3.67 403.0** 4.01 418.7** 4.20 155.9** 6.40 160.4** 6.67 163.3** 6.85 YANCHENG 20.3 0.24 108.5 1.31 120.9 1.47 7.4 0.37 23.1 1.17 25.4 1.29 YANGZHOU 37.0 0.42 -40.1 -0.46 -35.3 -0.40 -22.4 -1.06 -36.1 -1.73 -35.2 -1.67 ZHENJIANG -231.3* -2.42 -213.5* -2.29 -223.6* -2.40 -47.0* -2.08 -43.8* -1.96 -45.7* -2.05 TAIZHOU -89.7 -1.05 -56.2 -0.67 -58.5 -0.70 -30.2 -1.49 -24.2 -1.21 -24.7 -1.23 SUQIAN 651.2** 4.68 447.6** 3.24 480.4** 3.59 145.1** 4.41 108.9** 3.30 115.0 3.60 Constant -13258** -6.92 -6813.7** -3.37 -7419.7** -3.88 -2858.5** -6.31 -1711.6** -3.54 -1824.3** -3.99 N of Obs. 1440 1440 1440 1440 1440 1440 R-Squared 0.1326 0.1713 0.1714 0.1468 0.1683 0.1684 Wald chi2 216.41** 293.88** 243.69 287.21** F-Statistic 14.68** 14.36** Durbin chi2 0.073 0.757 0.030 0.459 Wu-Hausman 0.072 0.746 0.030 0.452 *significant at p < 0.05; **significant at p < 0.01; all tests are two-tailed

22 Table 6. Regression Results for Weighted COD and NH3-N Emissions Dependent: Weighted COD Emission Weighted NH3-N Emission Model 1: 2SLS Model 2: 2SLS Model 3: OLS Model 1: 2SLS Model 2: 2SLS Model 3: OLS Independent: Coef. Z Coef. Z Coef. t Coef. Z Coef. Z Coef. t MIGRANTS 2.358** 6.66 1.716** 4.75 1.717** 4.71 0.174** 4.02 0.128** 2.86 0.128** 2.84 INDUSTRY 1.483** 6.70 1.473** 6.65 0.107** 3.92 0.109** 4.00 MINORITY -28.033** -4.65 -28.475** -4.80 -28.452** -4.76 -1.324 -1.80 -1.356 -1.85 -1.360 -1.84 INCOME 7.589 0.35 -42.179 -1.85 -38.654 -1.81 5.713* 2.13 2.119 0.75 1.405 0.53 DENSITY -0.002** -5.14 -0.001** -4.81 -0.001** -4.79 -0.0001** -3.12 -0.0001** -2.90 -0.0001** -2.86 EDUCATION 0.784* 2.42 1.042** 3.25 1.037** 3.21 0.007 0.17 0.025 0.64 0.026 0.66 LABOR 1.627 1.41 -0.204 -0.17 -0.230 -0.20 0.097 0.68 -0.036 -0.25 -0.030 -0.21 RIVER -1.900 -1.02 -0.985 -0.53 -1.011 -0.54 -0.709** -3.10 -0.643** -2.82 -0.638** -2.78 Dummies NANJING 24.1* 2.37 44.2** 4.24 43.3** 4.21 -0.1 -0.11 1.3 1.03 1.5 1.18 WUXI 10.3 0.95 6.4 0.59 5.3 0.50 -3.8** -2.84 -4.1** -3.07 -3.9** -2.97 XUZHOU -4.3 -0.47 -16.8 -1.84 -15.9 -1.78 -0.6 -0.50 -1.5 -1.30 -1.7 -1.50 CHANGZHOU 47.8** 4.25 53.9** 4.85 52.9** 4.84 2.9* 2.11 3.3* 2.44 3.5** 2.63 SUZHOU -29.8* -2.41 -27.7* -2.27 -29.2* -2.47 -2.3 -1.54 -2.2 -1.45 -1.9 -1.30 NANTONG 6.6 0.84 12.9 1.65 12.5 1.60 -0.4 -0.45 0.01 0.02 0.1 0.11 LIANYUNGANG -18.0 -1.41 -29.6* -2.33 -28.5* -2.28 -0.5 -0.33 -1.4 -0.87 -1.6 -1.02 HUAIYIN -6.1 -0.57 -4.2 -0.40 -3.4 -0.33 1.4 1.10 1.6 1.21 1.4 1.10 YANCHENG -20.0 -2.32 -12.7 -1.48 -12.1 -1.42 -0.9 -0.81 -0.3 -0.31 -0.5 -0.43 YANGZHOU -5.1 -0.56 -11.3 -1.25 -11.1 -1.22 6.4** 5.70 6.0** 5.30 5.9** 5.23 ZHENJIANG 14.9 1.52 16.8 1.74 16.3 1.69 -1.8 -1.49 -1.7 -1.39 -1.6 -1.30 TAIZHOU 4.2 0.48 7.1 0.82 7.0 0.80 -0.3 -0.25 -0.1 -0.06 -0.04 -0.03 SUQIAN 1.4 0.10 -16.3 -1.14 -14.7 -1.06 2.7 1.54 1.4 0.79 1.1 0.63 Constant -159.4 -0.82 391.9 1.87 361.7 1.83 -51.3* -2.14 -11.5 -0.44 -5.4 -0.22 N of Obs. 1442 1442 1442 1442 1442 1442 R-Squared 0.1955 0.2198 0.2088 0.1102 0.1202 0.1079 Wald chi2 350.47** 406.28** 180.12** 197.37** F-Statistic 20.01** 9.71** Durbin (score) 0.299 0.174 1.570 0.469 Wuchi2(1)-Hausman 0.294 0.172 1.548 0.462 *significant at p < 0.05; **significant at p < 0.01; all tests are two-tailed

23 Table 7: Data Appendix: Variable Definitions and Sources

Variable Definition Source

Number of pollution sources / Per capita POLLUTION Jiangsu EPA, 2005 emission of SO2, SMOG, COD, NH3-N (kg)

China Census 2000 with MIGRANTS Percentage of rural migrants (%) Maps, China Data Center

Percentage of employment in “dirty & hard” China Census 2000 with INDUSTRY industry (%) Maps, China Data Center

China Census 2000 with MINORITY Percentage of minorities (%) Maps, China Data Center

Annual County Reports, INCOME Log of per capita disposable income (Yuan) 2004 and 2005

China Census 2000 with DENSITY Population density (1,000 person/km2) Maps, China Data Center

China Census 2000 with EDUCATION Percentage of senior high-school educated (%) Maps, China Data Center

China Census 2000 with LABOR Percentage population of age 15 – 64 (%) Maps, China Data Center

China Census 2000 with RIVER Log of distance to nearest river (meter) Maps, China Data Center

China Census 2000 with PREFECTURE Dummies for 13 prefectures (1/0) Maps, China Data Center

24