INTRODUCTION The economic successes of the

The economic successes of the United States and the other advanced industrial

nations were largely predicated upon their relatively early entry into manufacturing and

their concomitant rise into the position of manufacturing powers in the 19th and 20th

centuries (Peskin 2003). Jobs in manufacturing have been long viewed as relatively good

sources of employment, providing high wages and benefits compared to other sectors of

the economy (Bivens, Scott and Weller 2003; Bluestone and Harrison 1982). Also,

communities, regions, or nations with higher proportions of workers in manufacturing

have generally experienced lower levels of inequality (Flynn 2003; Harrison and

Bluestone 1988; Kalleberg, Reskin and Hudson 2000).

Manufacturing in the United States has generally been in a period of decline over

most of the past fifty years, as the share of workers in manufacturing peaked at

twentyeight percent in 1965 (Rowthorn and Ramaswamy 1997), dropping to just over

eleven percent in 2004 (Bureau of Labor Statistics 2005). These declines in

manufacturing employment are commonly referred to as deindustrialization, and are

usually characterized as highly problematic. Richard Florida (2005: 96) points out that

today policy makers are concerned with outsourcing of jobs to other countries, but this

represents a far smaller threat to U.S. workers than deindustrialization, which "destroyed

millions of jobs in the industrial Midwest during the 1980s." The overall downward trend

in manufacturing employment has not been uniform across different regions of the

nation, first beginning in the Northeast “Rust Belt” region in the 1960s and 1970s.

However, during this same time period, manufacturing jobs were growing in the South,

particularly in the Sunbelt states (Jackson, Masnick, Bolton, Bartlett and Pitkin 1981;

Kasarda 1995). The de facto industrial policy termed the “selling of the South” by Cobb

(1993) did not lead to long-term manufacturing success for the region, as the share of

manufacturing jobs in the South declined in the 1980s and 1990s (see also Falk and

Lyson 1988).

Deindustrialization has forced many low-skill workers into lower paying service

jobs or into unemployment, and has also served as a catalyst for the increasing income

inequality observed over the past thirty years (Alderson 1999; Bivens, Scott and Weller

2003; Bluestone and Harrison 1982; Flynn 2003; Nielsen and Alderson 1997; Rowthorn

and Ramaswamy 1997). Even given these declines in manufacturing and long-term

indicators of the rise of the service sector (Rowthorn and Ramaswamy 1997), countless

local officials and development experts still today expend both energy and tax dollars in

an attempt to lure manufacturing industries into their communities. This effort is often

referred to as “smokestack chasing” since the very nature of the manufacturing process

involves the creation of waste by-products, often released or expelled from the facility

into the surrounding air, land or water.

Simultaneously over the past several decades, pro-environmental values have

been increasingly expressed by American citizens (Dunlap 1992, 2000), as well as

citizens of other advanced industrial nations (Inglehart 1977, 1995, 1997). In Silent

Spring, a book viewed by many as critical to the development of the modern

environmental movement, Rachel Carson (1962: 16) noted that in the 20th century,

humanity, largely through industrial development, had “acquired significant power to

alter the nature of (the) world.” The Club of Rome report entitled The Limits to Growth

(Meadows, Meadows, Randers and Behren 1972) was one of the first to argue that

economic growth could not continue indefinitely without ultimately disrupting global

ecological equilibrium, leading to what William Catton (1980) would subsequently refer

to as ecological overshoot and collapse. Industrialism, predicated on endless economic

growth and found in capitalist and communist societies alike, was thus believed to be the

key source of ecological problems in modern society (Porritt 1984).

Such concerns over the effects to nature and humanity of both the increasing

volume and intensity of pollution produced by modern industry was shared by many

scientists, policy makers, and the general public, reaching a critical mass by the first

Earth Day celebration in 1970 (Dunlap 1992). Declines in overall concern regarding the

environment were modest in the 1970s, as levels of support for environmental protection

remained high (Dunlap 1992). Support for the environment amongst the public began to

climb again in the 1980s and into the 1990s, as it was perceived that the federal

government was not paying enough attention to the environment or was acting in a

manner that was potentially harmful towards it (Dunlap 1992). Concerns particularly

were expressed that problems in the environment were unequally distributed amongst the

population, disproportionately affecting the poor and racial minorities (Bullard 1983,

1992; Bryant and Mohai 1992; United Church of Christ 1987). However, support for

environmental protection is broad-based and is still relatively high today, reflecting the

entrenchment of it as a general concern amongst the American public (Dunlap 2000).

This broad-based, long-term support for environmental protection is no doubt

linked to the steady stream of problems and issues people hear about and experience in

their day to day lives. From global problems such as climate change, acid rain, and ozone

depletion to more localized problems including habitat loss, bioinvasion, overfishing, or

water or air pollution, plus a nearly infinite additional number of problems and potential

issues, the state of the environment is frequently reflected upon by individuals,

organizations, and societies. Further, the issue of how to actually address many of these

problems without irrevocably doing harm to the modern way of life is a vexing one for

politicians, business leaders, social activists, and ordinary citizens alike, primarily

because at its core, the industrial activity that undergirds much of our modern existence

appears responsible for many of these problems.

Private industry and government have each been frequently scrutinized by

environmentalists and the public at large for not paying enough attention to concerns

about environmental protection or to problems apparent in the natural environment. This

criticism has often been misplaced or overdrawn, however, since there has been a

longstanding recognition of both the high level of concern for environmental protection

as well as the damage that has been and is continuing to be done to the environment.

Even in the very pro-business climate of the Sunbelt states in the 1980s, protecting the

relatively unspoiled natural environment was seen as a central problem for developers

wary of public scrutiny (Cobb 1993). This reflected recognition by industry and

government of the fears of environmentalists and environmentally-oriented policy

makers over the impact of economic growth upon the natural environment, rooted in the

mounting empirical evidence of environmental problems. Grievances were voiced over

the effect of land-use patterns on general environmental quality (Carson 1962; McHarg

1969; Norton 2005), as well as about the overall impact of pollution generated by

manufacturing facilities (Carson 1962; Meadows et al. 1972). Thus manufacturing has

been particularly criticized by environmentalists, since the very nature of the

manufacturing process leads to the production of external wastes, usually in the form of

air, water or ground pollution, which are generally unaccounted for by the manufacturing

facility and must thus be dealt with or faced by the public.

Governmental efforts to clean up polluted areas or to limit future pollution have

also been often viewed as limited or inefficient, particularly those undertaken at the

federal level. One notable example of these failed efforts involves the Superfund

program, instituted by the Environmental Protection Agency (EPA) during the Carter

administration, and designed to clean up the most heavily polluted industrial sites in the

country. In the first decade following implementation of the Superfund program, over

34,000 potential sites were identified by the EPA, with 1,236 being placed on the

National Priorities List (indicating that they were the most urgently in need of

remediation). Of those sites, only 33 had been certified as “clean” after a decade of the

Superfund program (Barnett 1994). The overall failures of the Superfund program to

clean up or remediate these sites reflects the failure of governmental regulation to

successfully balance “between an industrial drive for profit and public demands for

protection from production-related harms to health, safety, and the environment” (Barnett

1994: 31).

This tension between economic interests and environmental protection was

utilized by those in certain branches of the federal government to challenge the premise

of environmental regulation. In 1971, the Commerce Secretary warned that new

environmental regulations being considered by Congress at the time would “throw

thousands of productive people out of work,” leading to such severe economic

disruptions that major new relief programs, similar in scale to those created during the

Great Depression, would be necessary to counter the widespread unemployment and

social problems that would ensue (Kazis and Grossman 1983: 17).

Communities that sought remediation of environmental problems caused by

industry often faced what Kazis and Grossman (1983) referred to as “job blackmail”: the

threat of job loss or plant closure. This threat was particularly problematic for poorer

communities or those comprised primarily of racial-ethnic minorities (Bullard 1992). As

a result of this inherent tension between economic interests and environmental protection,

particularly exacerbated by overall declines in manufacturing employment, an intractable

dividing line has developed between those who focus on manufacturing as a primary

source of local economic growth and employment opportunities, and those who

emphasize environmental quality and protection (Bullard 1994; Green 1998; Kazis and

Grossman 1983).

This dividing line has played out in the form of a societal debate, viewed as

pitting “jobs versus the environment,” which has been a central frame utilized by

academics and policymakers on all sides of the issue. This “jobs versus the environment”

divide has been a dominant one in the research by environmental sociologists and

environmental economists, as well as for those researchers who examine economic and

community development efforts. Competing (and in some cases complementary)

theoretical approaches have been developed within sociology and economics which are

centrally focused upon this specific issue; these include the Human Ecology, Treadmill of

Production, neo-Marxist, Ecological Modernization and Environmental Justice

approaches within environmental sociology, and the Environmental Kuznets Curve and

Power-Inequality approaches developed within the field of economics. Fleshing out of

the commonalities and differences between and among these approaches will take place

below and in Chapter Two. Also, from the standpoint of economic and community

development, resolving this tension between economic growth and environmental

sustainability has important implications for future policies and development endeavors.

However, the question itself has broader appeal in the social sciences, as it is

more generally about the competing interests and values of central social actors in the age

of late modernity; thus theorists of society and social change broadly understood should

be interested in this debate as well. Therefore in this dissertation an examination is

undertaken of the relationship between patterns of growth and decline in the

manufacturing sector of the economy and the state of the natural environment. In the next

section a statement of the problem being addressed here is provided, followed by a

discussion of the importance of this study, focusing primary attention on what it

contributes to the existing literature surrounding the relationship between economic

growth and environmental conditions.

Statement of the Problem

This study examines the utility of competing theoretical approaches regarding a

central question within environmental sociology and environmental economics: what is

the empirical relationship between economic growth and environmental quality? As

mentioned in the prior section and elaborated upon in the next chapter, this issue is one

which has been pivotal in the development of environmental sociology and

environmental economics, but which to date has not been satisfactorily resolved. Since

the question has been of such importance in defining these areas of inquiry, the failure of

past work to effectively or thoroughly address the issue is problematic and in need of

rectification. This dissertation is an important step towards a better understanding of this

key question.

Careful review of the literature suggests a pronounced lack of agreement

regarding the specification of the direction of this relationship. Many researchers argue

that higher levels of manufacturing activity (generally utilized as a proxy for economic

growth) lead to higher levels of pollution (Cable and Cable 1995; Catton 1980; Foster

1999; O‟Connor 1988; Schnaiberg 1980; York, Dietz and Rosa 2003). This is the basic

position of most who call themselves environmental sociologists, particularly within the

United States. Supporters of this view can be broadly characterized as operating within

an Environmental Conflict Theory (ECT) perspective, which includes most human

ecologists, as well as adherents of the Treadmill of Production and neo-Marxist

approaches.

Another camp, composed of some environmental sociologists, but primarily made

up of environmental economists, argues for a curvi-linear relationship between economic

growth and environmental quality (Grossman and Krueger 1991; Kuznets 1955; Mol and

Sonnenfeld 2000; Mol and Spaargaren 1993; Stern 2004, 2005). Drawing from earlier

modernization theories, the logic of this approach begins with the premise that in early

stages of “development” broadly construed, pollution is expected to increase. However,

at a certain point of development, it is expected that environmental quality will begin to

improve. This is the basic position of those who utilize the Ecological Modernization and

Environmental Kuznets‟ Curve (EM/EKC) perspectives.

Finally, a third group has emerged within sociology and economics, focusing not

just on absolute levels of pollution, but rather on the distribution of the burdens of

pollution between social groups. Here the emphasis is placed upon the inequality of

power, broadly conceived, and its effect upon levels of pollution, particularly as it

impacts traditionally less-advantaged social groups (racial-ethnic minorities and the poor)

(Boyce 1994; Bullard 1983, 1994, 2005; Torras 2005; Torras and Boyce 1998). It is

argued that as the inequality of power increases, two results will typically be observed.

First, absolute levels of pollution will tend to be higher, and second and most importantly,

the relative distribution of pollution will be more unequal, disproportionately impacting

disadvantaged social groups. This is the general position of the Power-

Inequality and Environmental Justice (P-I/EJ) perspectives.

In this study these competing approaches are empirically tested within common

empirical models were possible, in an attempt to better understand the relationship

between economic activity and environmental quality. Data from a variety of sources are

analyzed with conventional descriptive and inferential statistical techniques, as well as

recently developed and emergent spatial statistical techniques. In the next section, the

importance of this dissertation and its specific contributions to the existing literature is

spelled out in greater detail.

Importance of the Study

Numerous researchers have attempted to examine the empirical relationship

between economic change and environmental quality, particularly within the

Environmental Kuznets Curve approach; typically environmental quality has been

operationalized using some measure of the level of pollution in a monolithic geographic

space (Coursey and Hartwell 2000; Grossman and Krueger 1995; Raymond 2004; Seldon

and Song 1994). Few studies have assessed this relationship over time or particularly

across space, or have attempted to explore issues of time and space simultaneously. Thus

acquiring a more complex and nuanced understanding of the relationship between

economic change and environmental quality is the primary contribution of this

dissertation.

Second, the links between what are macro-structural theories of environmental

change and typically community-level processes of environmental inequity will be

fleshed out and comparatively tested. Results derived from this study will inform and add

to the evolving debate between the competing theoretical camps, which is another

substantial contribution of this dissertation.

A third contribution of this dissertation is the refinement and localization of

research in this area of study. Past research on the relationship between economic trends

and environmental conditions has relied upon national or state-level analyses, whereas

this study focuses on locale/community-level data (counties). This is advantageous

because this is where both manufacturing facilities are found and the majority of any

pollution emanating from the facility remains (a key contention of Environmental Justice

researchers is that pollution tends to be localized and clustered, often in areas with high

concentrations of racial-ethnic minorities or those in poverty).

Fourth, this dissertation will employ recently developed spatial statistics, allowing

for more refined and appropriate statistical analyses than can be performed through the

use of a-spatial regression techniques or other similar statistical approaches.

These are not the only contributions that this dissertation will make to the existing

literature; however, these are certainly among the most substantial to note. In the next

chapter, a review of the relevant literature will be undertaken. This chapter will introduce

the reader to the relevant issues and parse out the critical elements for empirical analysis

designed to explore this relationship between economic change and environmental

conditions. This literature review accomplishes three central tasks: 1) it provides

additional context about the overall relationship between economic change and

environmental conditions; 2) there is a delineation of the different theoretical

perspectives and their key statements involving the link between economy and

environmental conditions; and 3) hypotheses from these perspectives are developed,

which can be tested using the available data. In the concluding section of Chapter Two,

these hypotheses are proposed and discussed. The data utilized to test these hypotheses

and the methods of analysis to be used are described in extensive detail in Chapter Three.

Chapters Four through Six report the results of the data analyses. Chapter Four

presents a descriptive portrait of the spatial character and temporal patterns in the

manufacturing and service sectors of the U.S. economy between 1988 and 1997, and of

toxic pollution output from manufacturing during that same period, emphasizing the

intersection of the changing character of the economy across local space and the

differential effects on environmental quality in those areas. The analyses in Chapter Five

address the central issue of whether or not pollution is the “cost of doing business” for

U.S. communities; in other words, is pollution a necessary trade-off for jobs or higher

wages in the local area? In Chapter Six, the integrative Domestic Pollution Havens

(DPH) concept is developed and tested in empirical analyses as an effort to move beyond

the rhetorical „environment or economy‟ tradeoff, by focusing on those localities in

which both economic conditions are relatively poor and environmental quality is highly

degraded. Finally, the concluding Chapter Seven includes a summary of the research

findings from the previous three chapters, linking these findings to the existing body of

theory on environment and society, and providing suggestions for the direction of future

research.

CHAPTER II

REVIEW OF THE LITERATURE

Introduction: The Industrial Production Process and Environmental Quality

As discussed in Chapter One, there exists a long-standing academic and public

concern over the relationship between growth, particularly economic growth, and

environmental quality. The basic argument goes that there is inherent tension between the

two, such that social groups must forego one to achieve the other. On one side,

environmentalists emphasize sustainable practices, while most in the business community

and many politicians characterize side effects from economic growth as “the cost of

doing business.”

This divide in thinking is not a recent development, as it extends back to the

earliest writings of environmentalists. The tenor of the debate intensified in the 1960s and

1970s, after works such as The Closing Circle (Commoner 1971), The Limits to Growth

(Meadows et al. 1972), and Overshoot (Catton 1980) all warned of potential (and even

impending) ecological catastrophes rooted largely in human production practices. Within

the sub-discipline of environmental sociology and within the environmental social

sciences more generally, this “enduring conflict” has been a central and recurring theme

(see Humphrey, Lewis and Buttel 2002; Schnaiberg and Gould 1994).

In his recent work Collapse (2005), Jared Diamond nicely captures the character

of this conflict, which can be briefly restated as such: All modern societies depend on

extracting natural resources, both non-renewable and renewable ones. The only questions

involve where, in what amounts, and by what means we choose to do so. Today, due to

economic costs involved with such extraction, big businesses do most of this work.

Environmentalists tend to view big businesses as an enemy, and vice versa. Diamond

(2005:441-42) goes on to state that:

“(e)nvironmentalists blame businesses for harming people by damaging

the environment, and routinely putting the business‟s financial interests

above the public good. Those accusations are often true. Conversely,

businesses blame environmentalists for routinely being ignorant of and

uninterested in business realities, ignoring the desires of local people and

host governments for jobs and development, placing the welfare of birds

above that of people, and failing to praise businesses when they do

practice good environmental policies. Yes, those accusations too are often

true.”

At the same time that this debate over the effects of economic growth upon

environmental quality has intensified, manufacturing in the United States has generally

been in a period of decline (Rowthorn and Ramaswamy 1997; Bureau of Labor Statistics

2005). The deindustrialization of America is well underway, and is viewed by many

policymakers as a source of great concern. Some blame environmentalists or

environmental policy and regulations for the decline in manufacturing. Within this

framework, it is contended that an overemphasis on environmental quality has come at

the cost of economic growth and progress, as manufacturing industries and jobs have

generally been considered higher “quality” than most of the service jobs that replace

manufacturing.

Thus the debate is mutually reinforcing and infused with rhetoric and evidence on

all sides. However, the economic growth versus environmental quality frame has to date

largely been utilized as a polemic or a rhetorical characterization, with few attempting (or

having the data available) to analyze the interrelationship of both issues simultaneously,

particularly at a more localized level of analysis such as the county-level, where both

economic growth (or contraction) and environmental quality issues is most likely to be

directly felt (studies examining economic development issues at the county-level include

Partridge and Rickman 2005 and Mencken and Tolbert 2005; studies examining

environmental quality at the county-level include Hunter and Sutton 2004 and Bowen,

Salling, Haynes and Cyran 1995; for the combination of the two at the county-level, see

Pagoulatos, Goetz, Debertin and Johannson 2004).

The economic growth versus environmental quality frame also serves as a key

element in most of the leading theoretical perspectives within the environmental social

sciences. There are several leading approaches within environmental sociology and the

environmental social sciences, which can be cast under four general umbrellas:

1) Environmental Conflict Theory

2) Environmental Kuznets Curve and Ecological Modernization

3) Power/Inequality and Environmental Justice

4) Social Constructionist

While important and worthy of consideration, the Social Constructionist

perspective does not directly address the economic growth versus environmental quality

frame at an empirical level. Rather, the Social Constructionist approach focuses attention

on the way in which social issues are framed or constructed discursively. Thus this

perspective is not included in subsequent discussion, though it certainly informs the

manner in which the oppositional character of the environment-economy duality is

understood. In the next sections of this chapter, each of the remaining three leading

perspectives in environmental sociology and the environmental social sciences is

described, noting here the central manner in which the approach addresses the tension

between economic growth and environmental quality.

Environmental Conflict Theory Perspectives

The first approach includes several distinctive perspectives which have been

linked together in prior theoretical and empirical studies (Fisher and Freudenburg 2001;

York, Dietz and Rosa 2003; Mol and Spaargaren 2005). These approaches are generally

characterized as political-economy, neo-Marxist, or treadmill of production perspectives,

but in this dissertation they will be referred to as “Environmental Conflict Theory”

perspectives given their emphasis on conflictual social relations as the driving force

behind the antagonistic environment-economy duality. The starting point underlying each

of these perspectives is that the environment is being exploited as a result of the nature of

the logic of capitalist accumulation (Cable and Cable 1995; O‟Connor 1988; Schnaiberg

1980). Furthermore, there is an unstoppable cultural exhortation which impacts workers

and owners alike, emphasizing continuous, uninterrupted economic growth as the

primary objective for all (Schnaiberg and Gould 1994). Thus any efforts to redress this

exploitation must be initiated by key social-institutional actors, either through a

revolutionary social movement or particularly through the State (Cable and Cable 1995;

O‟Connor 1988; Schnaiberg 1980; York 2003).

It would be safe to argue that most environmental sociologists subscribe to some

variant of this thinking, though the extent to which the focus of blame is placed upon

capitalism or other social institutions differs sufficiently so that combining each

perspective into one of the existing rubrics (either political economy, neo-Marxist, or

treadmill of production) seems unfair and overly dismissive of the differences that exist.

It should be noted that as regards the relationship of economic conditions and

environmental quality, adherents of the Environmental Conflict perspectives generally

see this relationship in a similar way. As economic growth occurs, environmental

conditions should worsen. This may be due to the nature of capitalist accumulation, the

character of productive processes, and/or a lack of oversight by the State or other social

actors, but each comes to the relatively same conclusion: stopping the inexorable pursuit

of economic growth at all costs is essential to avoid ecological overshoot and collapse,

and to ensure planetary survival (Cable and Cable 1995; Catton 1980; Diamond 2005;

Foster 1999; O‟Connor 1988; Schnaiberg 1980; York, Dietz and Rosa 2003).

These approaches tend to share a common focus on at least five key issues, which

leads them to differ significantly from the other perspectives, most notably the Ecological

Modernization and Environmental Kuznets Curve perspectives that we will discuss next.

The key elements which the Environmental Conflict theory perspectives share include: 1)

there is a focus on changing the capitalist system, which is argued will resolve

environmental problems almost as a byproduct outcome; 2) these approaches argue for

“absolute” sustainability and ecological survival; 3) there is a shared sense that the

structural and institutional changes which have occurred since the late 1980s regarding

the environment are largely “window dressing”; 4) there is a lack of practical suggestions

for concrete trajectories towards social change and ecological sustainability; and 5) the

environmental problems focused on tend to be more “apocalyptic,” such as climate

change and biodiversity (Mol and Spaargaren 2005: 92-93). The Ecological

Modernization approach is largely dismissed by adherents of the Environmental Conflict

Theory perspectives due to its unrealistic optimism and somewhat overinflated hyping of

the “successes” of ecological modernization (Schnaiberg, Pellow and Weinberg 2002).

So in the final analysis for each of the perspectives contained with this

Environmental Conflict Theory approach, the only possible resolution to the intractable

tension between economy and environment would involve either one of the following, or

some combination of these scenarios: large-scale social reform, a successful

revolutionary social movement, or ecological collapse of some sort which forces societal

restructuring for the remaining population. In the next section, a description will be

provided of the two perspectives which are generally seen as oppositional to the

Environmental Conflict Theory.

Ecological Modernization and Environmental Kuznets Curve Perspectives

The second approach discussed in this dissertation includes the Ecological

Modernization (EM) and Environmental Kuznets Curve (EKC) perspectives (see

especially Mol and Sonnenfeld 2000 and Mol and Spaargaren 1993 for Ecological

Modernization; Grossman and Krueger 1991, Kuznets 1955, Stern 2004, 2005 for

Environmental Kuznets Curve). These perspectives are often presented as separate and

distinctive, given that they come from different academic disciplines (EKC from

economics and EM from sociology). However, adherents of each perspective see the

relationship of economic conditions and environmental quality as having a curvi-linear

relationship: as economic growth occurs, environmental conditions should worsen at

first, until sufficient growth has occurred to make environmental quality concerns no

longer a luxury. This will lead to either a shift in demand from the public towards more

environmentally-friendly production processes, or an increased environmental activism

on the part of the State, or both in tandem. The result will be improved environmental

conditions.

The EKC perspective follows the more general work by Nobel Prize award

winning economist Simon Kuznets (1955), in which an inverted or U-shaped relationship

is expressed between level of economic development and income inequality. Essentially,

it is argued that income inequality will increase in the early stages of economic

development, yet decrease as development reaches a certain tipping point. Similarly, the

central premise underlying the EKC perspective is that the environment is a luxury good

demanded only by richer individuals or countries when other more pressing needs (in

effect a sufficiently high economic standard of living) have already been met (Torras

2005).

The core assumption (which is shared by the EM theorists) derives in form and

spirit, if not in content, from the modernization approach popular within economics and

the sociology of development in the 1950s and 1960s. In early stages of “development”

broadly construed, pollution is expected to increase. However, at a certain point of

development, it is expected that environmental quality will begin to improve. It is here

that the EM and EKC approaches part company somewhat, though not due to substantive

differences as much as alternate foci deriving from core disciplinary concerns. Briefly,

for the EM proponents, progress in environmental quality results primarily from

institutional and organizational improvements reflective of the development of an

ecological rationality separate and undetermined by economic rationality, while in the

EKC approach the operating assumption is that higher per capita incomes allow for

nations (or other entities) to focus greater attention and resources on improving

environmental quality.

The EM approach did not develop in an intellectual vacuum, as it synthesized

primarily in response to the confluence of several social-structural, institutional, and

ecological factors (Hajer 1995:94-95): 1) the decline of radical environmentalism in the

late 1970s, largely in response to the economic recession which gripped most of the

Western world; 2) the professionalization of the environmental movement; 3) the rise of

issues such as acid rain and ozone depletion which begged for pragmatic solutions and

not societal overhaul; and 4) a nascent set of concepts and language developed in the

academic and policy arenas which were geared towards solving environmental problems

within the general structural-institutional framework of society (italics added).

Essentially, from the beginning the EM approach involved a focus on “(existing or

programmed) environmental reforms in social practices, institutional designs and societal

and policy discourses to safeguard societies‟ sustenance bases” (Mol and Sonnenfeld

2000).

The central argument of EM is that it represents a shift away from the 1970s

“ecoalarmism” and programs of (unfulfilled) radical social transformation, towards

ecological modernization and institutional reform, reflecting the changing values and

environmental awareness of the late 1980s and early 1990s (Mol and Spaargaren 1993).

Besides changing ecological conditions and concerns, this shift was significantly

influenced by the widespread adoption of the concept of “sustainable development” as a

worthwhile and necessary social and political program, particularly following the

Brundtland report of 1987 (Mol and Spaargaren 1993; see also Stokke 1991). However,

what was most critical for Mol and Spaargaren (1993) were substantive institutional

reforms, particularly within the economy, the result of which was the emergence of an

autonomous ecological sphere within social life. The (relative) disembedding of the

ecological sphere from the economic sphere (where ecological concerns were largely

subordinated under processes of production and consumption) is highly significant, as it

involves the institutionalizing of ecological interests and “the transformation of central

institutions of modern society within the boundaries of modernity, in order to solve the

ecological crisis” (Mol and

Spaargaren 1993: 437). This involves for Mol (1996: 307) the “emergence of an

ecological rationality parallel with an economic rationality,” which, like past social

struggles to impose limits on economic rationality (such as the labor movement), is the

result of social debate and contestation, in this case the debate and struggle which took

place in the 1970s and 1980s involving ecological concerns.

As opposed to the Environmental Conflict Theory approaches described above,

Mol (2005:92-93) contends that five characteristics of the EM approach (which are

generally shared by EKC adherents) include: 1) work within the EM tradition is centered

around environmental reforms first and foremost; 2) EM focuses on “relative” (typically

quantifiable) environmental improvements; 3) EM contends that real structural and

institutional changes have occurred since the late 1980s regarding the environment; 4)

EM links theory with practice more effectively; and 5) the environmental problems EM

adherents address are more “conventional” problems such as surface water pollution, or

solid waste disposal. These characteristics also describe the EKC approach, and as a

result, these two approaches are treated as one in this dissertation. In the next section, two

other approaches which are frequently characterized as distinctive are described as one

approach: the Power/Inequality and Environmental Justice perspectives.

Power-Inequality and Environmental Justice Perspectives

The first two perspectives described engage each other rather directly; the third

perspective addresses the relationship between economic conditions and environmental

quality in a distinctive way. The Power/Inequality and Environmental Justice

perspectives shift the focus towards vulnerable populations, though in many ways each

parallels the Environmental Conflict approach, with its emphasis on deteriorating

environmental conditions following economic growth. Drawing this central contrast,

Torras (2005: 1354) argues that the other approaches ignore the fact that

“environmentally degrading economic activities generate winners as well as losers.” In

other words, not all social actors experience the same environmental conditions that

result from economic growth.

This has recently been extended by Freudenberg (2005, 2006), who has

introduced the disproportionality hypothesis. This approach focuses on the inequality in

the production of harm, NOT the inequality of the experiencing of harmful outcomes

focused on the Power/Inequality and Environmental Justice camps. Freudenberg

contends that most environmental destruction is created by a privileged few industries or

facilities, yet the costs are passed down to the entire society, thus adding a separate layer

to the inequalities which emerge in the production process. However in the empirical

portion of this study, focus will be on the more traditional distribution of harm and risk

which emerges from the production process, but Freudenberg‟s work needs to be

considered as extending this relatively underdeveloped theoretical perspective.

Those within this tradition, which has emerged as a strand of environmental

economics, argue that social decisions governing environmental issues almost always

favor more powerful interests and social actors, leading to disproportionate

environmental degradation where wider inequalities of social power exist (Boyce 1994;

Torras 2005; Torras and Boyce 1998). Early formulations focused greater attention on

income inequality (Boyce 1994; Torras and Boyce 1998), while more recent work has

shifted to a more general view of inequality as rooted in social power inequities (Boyce

1999; Torras 2005). Most of the empirical work done within the Power/Inequality

approach has focused on levels of power across nations or states, and the differences in

environmental quality that are theorized to exist.

The Power/Inequality approach parallels the theoretical and practical orientation

of the Environmental Justice perspective, largely developed within environmental

sociology. Within this approach, the focus was placed on disproportionate environmental

impacts faced by African-Americans (Bullard 1983, 1994), and later by the poor and

other socially disadvantaged groups (Bullard 2005). Early empirical work done within

the Environmental Justice approach involved the construction of community-level case

studies regarding the historical trajectory of environmental degradation faced by socially

disadvantaged groups (Bryant and Mohai 1992; Bullard 1983, 1990; United Church of

Christ 1987). However, over the past decade a strand of statistical research focused on

environmental equity at the local level has emerged (Bowen et al. 1995; Downey 1998,

2005, 2006; Sadd et al. 1999). While the findings have been mixed (Downey 2005),

generally the results have pointed to some form of existing environmental inequality,

with African Americans and the poor being more likely to face poor environmental

conditions.

Linking these two approaches is logical, given that the Power/Inequality

perspective offers a more explicit theory for understanding how and why disadvantaged

groups face poor environmental conditions, an empirical reality observed within

countless Environmental Justice studies. What the EJ perspective brings to bear is the

more localized nature of these problems, something that to date the Power/Inequality

perspective has not addressed with its national or state-level studies. Thus for both

perspectives, the shared key problematic here is not the amount of pollution or

environmental problems per se, but it is the distribution of environmental conditions

within and across space. In the next section, the key hypotheses to be tested in this

dissertation will be drawn out for each approach.

Hypotheses

In this section, the first characterization of the hypotheses to be tested in the

analysis chapters will be introduced. The key commonality that each of these

perspectives shares is that they do clearly articulate some sort of directional relationship

regarding the link between economic change and environmental quality. As regards the

Environmental Conflict Theory and Ecological Modernization/Environmental Kuznets

Curve perspectives, there is a relatively straight forward relationship that allows

relatively direct comparisons between the two perspectives. Essentially, the

Environmental Conflict Theory adherents contend that economic growth is related to

worsened environmental conditions. This is contrasted by the Ecological

Modernization/Environmental Kuznets Curve perspectives which see the relationship as

curvi-linear: at first, the ECT perspective is accurate, but this is up to a certain point, at

which time economic growth continues yet environmental quality actually improves.

The Power/Inequality and Environmental Justice perspectives offer a slightly less

easy to compare relationship regarding economic change and environmental quality. For

adherents of these perspectives, what is most important to understand is not the

relationship between these two, but the level of inequality that exists within a social

space. In other words, regardless of level of economic growth or decline, if there is

considerable social, political or economic inequality across groups, environmental

conditions will be worse. This critical distinction can be empirically modeled within a

common framework alongside the other two perspectives, allowing for a comparative

assessment of each of these three approaches. This represents a significant opportunity

which previous studies in the environmental social sciences have not been able to

accomplish. However, it is critical to appropriately set up the data and the analytical

models for such an endeavor. In the next chapter, the data and methods utilized in this

study are described in considerable detail.

CHAPTER III

METHODS

The primary objective of this study is to examine the character and content of the

empirical relationship between economic change (growth or decline) and environmental

quality in the United States. To accomplish this goal, data sources are utilized which were

gathered at multiple points in time, at the county-level for the contiguous forty-eight

states in the United States. The data are drawn from multiple sources, which are

described in further detail in the next section. These sources include: 1) Environmental

Protection Agency (EPA) Toxic Release Inventory (TRI) and Risk Screening

Environmental Indicators (RSEI); 2) EPA DOCKET data; 3) Census of Population and

Housing; 4) Census Bureau County Business Patterns establishment data on

manufacturing; 5) the Hall and Kerr Green Policies Index; 6) the League of Conservation

Voters (LCV) score; 7) Dave Leip‟s Atlas of U.S. Elections; and 8) National Center for

Charitable Statistics (NCCS) Non-Profit Organizations data.

In this chapter, the data sources and methodological procedures will be presented

and discussed. First, the data sources used to construct the final dataset will be described.

Second, a description of the unit of analysis to be used throughout will be given. In the

next three sections, the dependent variables, independent variables, and control variables

(respectively) to be used in the statistical analyses will be described. Finally, the

analytical strategy will be presented.

Data

EPA TRI and RSEI Data

The EPA Toxic Release Inventory includes data on releases and transfers of toxic

chemicals from large facilities (U.S. Environmental Protection Agency 2006a). The TRI

data used here is drawn from individual year datasets for the years 1988 through 1997. It

is taken from the forty-eight contiguous states and examined at the county-level, either

for individual years, in multiple year blocks, or across the entire 1988-1997 time period.

TRI data has been criticized by some researchers (Downey 1998, 2005, 2006; Sadd et al.

1999; Szasz and Meuser 1997) on a variety of grounds.

A broad critique of TRI data includes the following points: 1) it is based on

selfreporting by manufacturing facilities of ten or more employees, thus it excludes

smaller facilities and does exhibit a sectoral bias; 2) it is not a measurement of actual

exposures but of crude potential risk, as one cannot assess the direction chemicals move

once released and who, if anyone, is exposed to them; 3) TRI data is only inclusive of the

U.S., a concern for those who seek cross-national studies; and 4) the data is only

available going back approximately twenty years (the first TRI data was released in

1989), thus longer-term historical processes cannot be examined. However, Downey

(2006: 27) notes that “(d)espite these limitations, TRI-based studies make up a significant

proportion of the environmental inequality literature,” due principally to the fact that TRI

data is the best available national-level data source for researchers to utilize. Thus the

TRI data should be viewed carefully as a crude measure of local toxicity, though it is

more than acceptable for that purpose, and furthermore for county-level analysis of the

entire United States, it is all that we have.

The Risk Screening Environmental Indicators is an enhanced and expanded data

source based upon the TRI data, which the EPA touts as a “state of the art” screening tool

that compares toxic chemicals released to the environment from industrial sources (U.S.

Environmental Protection Agency 2006b). This allows interested parties to assess the

relative hazard and risk of chemicals, facilities, regions, industries, or many other factors

(U.S. Environmental Protection Agency 2006b). As with the TRI data, the RSEI data

used here is drawn from individual year datasets for the years 1988 through 1997. It is

taken from the forty-eight contiguous states and examined at the county-level, either for

individual years, in multiple year blocks, or across the entire 1988-1997 time period.

Since the RSEI data is designed explicitly to provide some assessment of risk, it can be

argued that it represents an improvement over the “raw” TRI data, though measures from

both sources will be used throughout this study, as each source proves more useful to

address the different questions and hypotheses examined here.

EPA DOCKET Data

The EPA DOCKET dataset includes information on compliance enforcement of

major environmental laws through administrative actions and civil court cases, for the

years 1971 through 2000 (Right to Know Network 2006). The DOCKET data used here

is for the forty-eight contiguous states and examined at the county-level, and examined

across the entire 1971-2000 time period.

Census Demographic Data

The Census of Population and Housing dataset includes information taken from

both the 1990 and 2000 Censuses on population size and characteristics, such as race,

income, age, Census region, metropolitan status, et cetera (U.S. Census Bureau 2006).

These data are aggregated to the county-level for the forty-eight contiguous states, and

are drawn either from the 1990 or 2000 Censuses, as is relevant for the specific analysis.

County Business Patterns Data

The Census Bureau County Business Patterns dataset includes annual

establishment data for the manufacturing and service sectors, as well as for overall

industry (Bureau of Labor Statistics 2006). Information is utilized for the number of

establishments, number of employees, and average wages for all industries, all

manufacturing activity, and all service activity (Bureau of Labor Statistics 2006). This

data is aggregated to the county-level for the contiguous forty-eight states, and are

utilized either for individual years, in multiple year blocks, or across the entire 1988-

1997 time period.

Environmental Policies Data

The Hall and Kerr Green Policies Index, which is a state-level composite measure

of the relative “toughness” of environmental regulations and policies based on 77 specific

indicators, was developed in the years 1991 and 1992 (Hall and Kerr 1991). This

measure, drawn from the broader two-hundred fifty-six indicator Green Index, is still

being widely used to compare U.S. states on their environmental politics (see Allen 2001;

Andrews and Edwards 2004; Grant and Jones 2003; Grant, Bergesen and Jones 2002).

Legislator Rankings Data

The League of Conservation Voters (LCV) Score is a state-level ranking of each

congressional delegation and their environmental voting record (League of Conservation

Voters 2006). The construction of this measure by the LCV involves the scoring of each

legislator as either pro- or anti-environment on all environmentally-oriented legislation

brought before her/his house of Congress. The next step involves the calculation of a

statewide average of U.S. Senators and Representatives for each state. The premise

underlying the aggregation of all the legislators is that if a state elects pro-environmental

legislators to serve in Congress, it's more likely to have pro-environmental policies at the

state level also (League of Conservation Voters 2006). The measure used here is an

aggregation of scores for the years 1977 through 1994 (see Levinson 1999).

Voting Data

Dave Leip‟s Atlas of U.S. Elections provides state and county-level data on

national Presidential election results (Leip 2006). In this study, voter-turnout results from

the 1988 and 1992 Presidential elections are examined (Leip 2006). County-level data for

the contiguous forty-eight states is utilized from the average voter-turnout for the 1988

and 1992 Presidential elections, to produce a circa-1990 characterization of the strength

of local political culture.

Environmental Non-Profit Organizations Data

The National Center for Charitable Statistics (NCCS) data on non-profit

organizations (the National Taxonomy of Exempt Entities) is taken from IRS records of

tax-exempt organizations across the U.S., including those characterized as working on

issues related to the “environment and animals” (National Center for Charitable Statistics

2007). In this study, county-level data for the contiguous forty-eight states on the

presence of environmentally-focused organizations between 1989 and 2000 is used.

Unit of Analysis

This study examines the relationship between economic change and

environmental quality in U.S. counties. The unit of analysis for this study will be the

counties and county-equivalent areas of the 48 contiguous states of the U.S. While a

debate exists about the appropriate unit of analysis in environmental sociology (Allen

2001; Downey 1998), counties are chosen as the unit of analysis because of the

availability of data at this level, and also due to the widespread use of the county as a unit

of analysis in macro-level research (Allen 2001; Blanchard 2007; Blanchard and

Matthews 2006).

In the following section, the variables utilized in the analyses will be described,

including the dependent variables, independent variables, and control variables.

Measurement of Variables

Dependent Variables

The dependent variables examined in the statistical analyses include: Number of

TRI Facilities; Log Number of Pounds of TRI releases; Log Number of Pounds of TRI

releases per Facility; Log Median Risk Score; Dirtiness Index; Change in Manufacturing

Establishments 1988-1997; Change in Manufacturing Employees 1988-1997; Change in

Manufacturing Wages 1988-1997; Domestic Pollution Haven categorization; Domestic

Pollution Haven level characterization; Log Number of Pounds of TRI Releases per

Manufacturing Job, 1988-1997; Log Number of Pounds of TRI Releases per Average

Manufacturing Wage, 1988-1997; and Log Median Risk Score per Average

Manufacturing Wage, 1988-1997. Each of these dependent variables is described below.

Number of TRI Facilities

This most basic environmental risk measure involves a count of TRI facilities in a

given year or years (1988, 1997, and change between these years) in a given area (in this

study, counties are the geographic area of interest). The underlying assumption is that, all

things being equal, those counties with more TRI facilities face an increased risk of

negative outcomes from toxic releases and pollution. This measure is also used in certain

analyses as an independent variable.

Log Number of Pounds of TRI releases

This commonly used measure of pollution output in a geographical space refers to

the sum total of TRI reported releases in a given year or years (1988, 1997, the average

of all years 1988-1997, and change between these years) across U.S. counties. A natural

log is performed to account for the non-normal distribution of the variable. This measure

is also used in certain analyses as an independent variable.

Log Number of Pounds of TRI releases per Facility

This measure refers to the sum total of county-level TRI reported releases in a

given year or years (1988, 1997, the average of all years 1988-1997, and change between

these years), divided by the number of TRI facilities. To address non-normality, this

measure is logged as well. This measure is also used in certain analyses as an

independent variable.

Log Median Risk Score

A criticism of the use of the simple pounds of TRI releases measure is that it

includes some releases which may not pose a clear risk to human health; thus researchers

often utilize a measure which takes into account only those pounds of releases which are

characterized as toxic. Thus this measure refers to the sum total of county-level TRI

reported toxic releases in a given year or years (1988, 1997, the average of all years

1988-1997, and change between these years). Again, to address non-normality, this

measure is logged. This measure is also used in certain analyses as an independent

variable.

Dirtiness Index

This is an index measure of the proportion of the total number of manufacturing

industries in a county which report to the TRI. Since the TRI only requires certain

industries (which are generally the most pollution-intensive industries) to report

information on releases and transfers, counties which have a higher proportion of

facilities which report to the TRI could be characterized as housing more “dirty”

industries proportionally. This measure is taken for the years 1988 and 1997, as well as

the average of all years 1988-1997, and change between 1988 and 1997. A natural log is

performed to account for the non-normal distribution of the variable. This measure is also

used in certain analyses as an independent variable.

Change in Manufacturing Establishments 1988-1997 This is a

measure of the percentage change in number of manufacturing establishments in a county

between the years 1988 and 1997. A positive percentage indicates an increase in the

number of manufacturing establishments in 1997 versus 1988, while a negative

percentage indicates a decline.

Change in Manufacturing Employment 1988-1997 This is a

measure of the percentage change in number of manufacturing employees in a county

between the years 1988 and 1997. A positive percentage indicates an increase in the

number of manufacturing employees in 1997 versus 1988, while a negative percentage

indicates a decline.

Change in Manufacturing Wages 1988-1997

This is a measure of the percentage change in manufacturing wages in a county

between the years 1988 and 1997. Actually, the comparison is between 1988

inflationadjusted wages and 1997 wages. A positive percentage indicates an increase in

the average manufacturing wage in 1997 versus the inflation-adjusted average

manufacturing wage in 1988, while a negative percentage indicates a decline.

Domestic Pollution Haven categorization

Domestic Pollution Havens (DPHs) are groups of localities with relatively high

manufacturing pollution per economic reward to the community and with low

government “controls” of these polluters via enforcement actions. Domestic Pollution

Havens are operationalized using a classification of pollution-per-economic reward

(these rewards include jobs or average wage) in manufacturing by the level of social

controls (including EPA enforcement actions and State environmental regulations). See

Chapter 6 for a more detailed discussion of the character of DPHs.

The DPH categorization variable was created using the DPH count variable

described immediately below. In sum, the indicators used to construct the count variable

include three pollution-per-reward indicators (Log Number of Pounds of TRI Releases

per

Manufacturing Job, 1988-1997; Log Number of Pounds of TRI Releases per Average

Manufacturing Wage, 1988-1997; and Log Median Risk Score per Average

Manufacturing Wage, 1988-1997) and three government controls indicators (DOCKET

number of cases; DOCKET percentage of cases which involved court actions; and the

Green Policies Environmental Policies score by state) described below. The DPH count

variable was then analyzed using the Local Indicator of Spatial Autocorrelation (LISA)

statistic to create a dummy classification of DPH=1 or else=0, based on the LISA results.

Further details on the LISA statistic can be found in the analytical strategy section of this

chapter.

Domestic Pollution Haven level characterization

The DPH level characterization variable was created using the six potential

indicators of DPHs noted immediately above (this number was reduced following a

principal components analysis). Each of the three pollution-per-reward and three

governmental controls indicators were either coded as 1 if “high” (i.e. meeting the

criteria for DPH classification) or 0 if “medium” or “low.” The three pollution-per-

reward indicators were then compared with each of the three governmental controls

indicators to determine if counties could be characterized as DPHs, for each of the nine

possible crosstabulation comparisons. The cells which corresponded to high pollution-

per-economic reward AND low environmental controls were coded as an indicator of a

Domestic Pollution Haven. The results of these nine comparisons are summed to obtain

the number of DPH-conditions met in each county. The sum ranges from 0-9.

Again, the indicators used to construct the DPH count variable include the three

pollution-per-reward variables described immediately below (Log Number of Pounds of

TRI Releases per Manufacturing Job, 1988-1997; Log Number of Pounds of TRI

Releases per Average Manufacturing Wage, 1988-1997; and Log Median Risk Score per

Average Manufacturing Wage, 1988-1997) and three government controls indicators

(DOCKET number of cases; DOCKET percentage of cases which involved court actions;

and the Green Policies Environmental Policies score by state). These three government

controls indicators are described now, followed by the three pollution-per-reward variables.

EPA DOCKET Number of cases

This variable is a count of the total number of DOCKET actions by the EPA in a

county between the years 1971 and 2000.

EPA DOCKET Percentage of Cases which Involved Court Actions

This variable is measured as the percentage of all DOCKET enforcement actions

resulting in court litigation, which represents the most serious enforcement action

possibly undertaken by the EPA.

Toughness of State Environmental Regulations

The toughness of state environmental regulations variable is taken from the Green

Index Environmental Policies score for each state, which measures the relative

“toughness” of environmental regulations and policies based on 77 specific indicators.

Log Number of Pounds of TRI Releases per Manufacturing Job, 1988-1997

Several variables have been created for this study which can be characterized as

pollution-per-reward, or what economists refer to as “Pollution Output Intensity” (POI).

The POI concept was developed to capture the relative risk-reward tension between

economic activity and environmental quality (Hettige, Martin, Singh and Wheeler 1995).

Three possible aspects of POI are considered here, including the Log Number of Pounds

of TRI Releases per Manufacturing Job, 1988-1997. This variable assesses the

countylevel inter-annual mean of the ratio of logged number of pounds of TRI releases

per manufacturing job, for the years 1988 to 1997.

Log Number of Pounds of TRI Releases per Average Manufacturing Wage, 1988-1997

Another POI indicator is Log Number of Pounds of TRI Releases per Average

Manufacturing Wage, 1988-1997. It is calculated as the county-level inter-annual mean

of the ratio of log TRI pounds per average manufacturing wage for the years 1988 to

1997.

Log Median Risk Score per Average Manufacturing Wage, 1988-1997

The third and final POI indicator is Log Median Risk Score per Average

Manufacturing Wage, 1988-1997. This measure moves beyond potential exposure

modeling and examines actual risk by utilizing RSEI data (see above discussion). This

indicator is calculated as the county-level inter-annual mean of the ratio of the log

median risk score per average manufacturing wage for the years 1988 to 1997.

Independent Variables

The independent variables utilized in the statistical analyses can be combined into

the following categories: Economic Change; Economic Standing; Environmental Justice;

Local Political Culture; Local Social Organization; and Other Factors of Interest. In the

analytical strategy section below, the location of these variables in analyses is laid out.

The following section will describe the independent variables utilized within each of

these broad categories.

Economic Change

Nine variables are used to measure economic change, which it is argued is a

critical component in changing environmental quality across counties. These variables

include: Change in Manufacturing Establishments 1988-1997; Change in Manufacturing

Employees 1988-1997; Change in Manufacturing Wages 1988-1997; Change in Service

Establishments 1988-1997; Change in Service Employees 1988-1997; Change in Service

Wages 1988-1997; Manufacturing/Service Sector Change in Establishments 1988-1997;

Manufacturing/Service Sector Change in Employees 1988-1997; and Manufacturing/

Service Sector Change in Wages 1988-1997. Each of these variables is derived from their

respective 1988 and 1997 characteristic measures, which are described and characterized

in the next chapter. The three variables documenting change in the manufacturing sector

are also described in the section above, as they are also used in certain analyses as

dependent variables.

The first service sector variable is the Change in Service Establishments

19881997. This is a measure of the percentage change in service establishments in a

county from the year 1988 to 1997. A positive percentage indicates an increase in the

number of service establishments in 1997 versus 1988, while a negative percentage

indicates a decline.

The next service sector variable is the Change in Service Employees 1988-1997.

This is a measure of the percentage change in service employees in a county from the

year 1988 to 1997. A positive percentage indicates an increase in the number of service

employees in 1997 versus 1988, while a negative percentage indicates a decline.

The final service sector variable is the Change in Service Wages 1988-1997. This

is a measure of the percentage change in service establishments in a county from the year

1988 to 1997. As with manufacturing wages, the comparison is between 1988

inflationadjusted wages and 1997 wages in the service sector. A positive percentage

indicates an increase in the average service wage in 1997 versus the inflation-adjusted

average service wage in 1988, while a negative percentage indicates a decline.

The remaining three variables in this category are: Manufacturing/Service Sector

Change in Establishments 1988-1997; Manufacturing/Service Sector Change in

Employees 1988-1997; and Manufacturing/Service Sector Change in Wages 1988-1997.

The construction of these variables is described in detail in Chapter 4; however, it should

be noted now that each of these measures characterizes the shift between 1988 and 1997

in both the manufacturing and the service sectors for each county in the contiguous U.S.

Thus the possible categories include: growth or decline across both sectors; growth in

one and decline in the other (and vice versa); or no change in both. The thresholds for

assessing change in each of the three areas (establishments, employees, and wages) were

determined by sensitivity analyses designed to determine the optimal percentage of

growth or decline needed to capture variation across space. Again, more detail on these

measures can be found in Chapter Four.

Economic Standing

Two variables are used to examine the general level of economic standing in the

county, which is a key component in both the Environmental Conflict Theory and

Environmental Kuznets Curve/Ecological Modernization approaches. These two

approaches have different views regarding the relationship of this variable to

environmental quality: the Environmental Conflict Theory approach argues that as

economic development increases, environmental quality should worsen. The

Environmental Kuznets Curve/Ecological Modernization approach agrees with this

assessment, up to a point, arguing that once economic growth reaches a certain tipping

point, environmental quality should begin to improve as social actors seek out

environmental improvements. Thus the empirical relationship for this approach is not

linear, but is instead curvi-linear (see Chapter 2 for further elaboration of these

approaches).

As a result, two variables are used here to measure economic standing: per capita

income and per capita income squared. Per-capita income is simply measured as the

average household income per county in 1990. Per-capita income squared is calculated

by squaring per-capita income. This is done to assess the curvi-linear character of the

theorized relationship (i.e. if per-capita income is positively related with negative

environmental conditions, yet per-capita income squared is negatively related, this

supports the Environmental Kuznets Curve/Ecological Modernization argument; if it is

not, and each are positively related with negative environmental conditions, this would

support the Environmental Conflict Theory approach).

Environmental Justice

Five variables are used to examine aspects of the composition of the population

that represent socially disadvantaged groups, which are likely to face disproportionate

environmental risk, according to the Power-Inequality/Environmental Justice

perspectives (see Chapter 2 for further discussion of these approaches). It should be noted

that several of the following variables were logged because of the non-normality of their

distribution. The first variable is the log of percent black in the county in 1990. The

second variable is the log of percent in poverty in the county in 1990. Both of these are

standard variables frequently used in studies of environmental justice, as

AfricanAmericans and the poor are two particularly disadvantaged groups in the U.S.

The third variable is the log of percent elderly in the county in 1990. This is less

frequently used as an indicator of environmental equity, but it is also a measure of a

potentially vulnerable population that could experience environmental inequities.

The fourth variable used to address environmental inequities is the log of the

Black-White Median Income Ratio in 1990. This is a measure which expresses the ratio

of county-level income of blacks to that of whites. As the ratio moves closer to 1, there is

greater median income equity, and vice versa. The environmental equity argument is that

in areas where there is relative economic inequality, higher levels of environmental

problems should follow, as the Black (or other group) population is less able to

“purchase” environmental equity.

The final variable used to address environmental inequities is the log of the 1990

Black-White Dissimilarity Index. This is a well known segregation measure which

reflects the level of racial residential segregation in a county (Blalock 1967; Massey and

Denton 1993). It is commonly described as the percentage of Blacks that would have to

move in order for them to be evenly distributed across a given county; from an

environmental justice standpoint, if this measure is relatively low, then that means that

the population in that area is relatively evenly dispersed and thus environmental problems

would relatively equally affect different groups.

Local Political Culture

Two variables are used to examine the character of the local political culture,

which is argued to positively affect environmental quality (i.e. stronger local political

culture should be associated with lower levels of toxic releases or negative environmental

conditions). The first indicator of local political culture is the League of Conservation

Voters state delegation score. This is a measure of the environmental voting record of the

Congressional delegation in a state (including U.S. Senators and Representatives). A

higher score indicates a more pro-environment voting record of the delegation. The

second indicator is the average percentage voter turnout in 1988 and 1992 Presidential

Elections. This is a commonly used measure of citizen involvement and political efficacy,

indicating here the degree of strength of the local citizenry to affect environmental social

controls, with a higher turnout representing a stronger or more efficacious citizenry.

Local Social Organization

One variable is used here to examine the character of the local social organization,

which is argued to positively affect environmental quality (i.e. stronger local social

organization should be associated with lower levels of toxic releases or negative

environmental conditions). The variable is the presence of any Environmental 501c3

Organizations by county, 1989-2000. This captures local environmental activism which is

separate from the political process (i.e. from the regulatory control mechanisms), but

which does involve the capacity for working on the improvement of environmental

conditions in an area.

Other Factors of Interest

The other variables of interest include: the log of percent county in-migration

between 1985 and 1990; log of county population density 1990; and metro/adjacency to

metro area in 1990.

The first variable in this category is the log of percent county in-migration

between 1985 and 1990. This is a measure which captures the percentage of migrants

coming into a county, with the argument being that areas which are relatively polluted or

toxic, or which lack sufficient economic rewards per unit of pollution will be less

attractive to potential in-migrants.

The next variable is the log of county population density 1990. This captures the

number of people per square mile in a county, logged due to the non-normality of the

distribution. The typical argument is that areas with a higher population density will

experience higher levels of pollution.

The final variable in this category is the metropolitan status for the county as of

1990. The possible categories for this indicator include: location in a metropolitan area;

adjacency to a metropolitan area; or being located in a non-metropolitan and non-

adjacent county. This is a measure which captures the degree of urbanization in a county,

with the argument being that counties which are more urbanized are more likely to have

manufacturing facilities and are thus more likely to be polluted, all else being equal.

Analytical Strategy

The analysis which follows includes several distinct components and is carried

out over three chapters. These chapters, respectively, describe trends in pollution and

industrial change; examine the relationship between manufacturing pollution and

economic standing in the manufacturing and service sectors; and the spatial

correspondence of high manufacturing emissions and low governmental enforcement

actions, resulting in the characterization of these sub-regions as Domestic Pollution

Havens. The analyses will address the specific hypotheses outlined in Chapter Two in the

following manner.

Chapter Four

In Chapter Four, the focus is on identifying and describing the spatial character

and temporal patterns of growth and decline within two key industrial sectors of the U.S.

economy, manufacturing and service, and toxic pollution from manufacturing, as

measured in the TRI and RSEI. Particular attention is paid to differences along the lines

of Census regions or metropolitan area status. Another central issue involves the

identification of significant spatial concentrations in the manufacturing and service

sectors, and significant spatial concentrations of TRI facilities, releases, and “dirty”

facilities which overlap with significant clusters of manufacturing and service sector

changes. This is accomplished through the use of descriptive and exploratory data

techniques, including the reporting of measures of central tendency and dispersion,

percentage and proportion change, cross-tabulation, and univariate and bivariate Local

Indicators of Spatial Autocorrelation (LISA). Given the relatively unknown character of

the LISA statistic, a brief overview of what it is and what it does is in order.

Univariate measures of spatial concentration (either Moran‟s I or LISA) relate the

value of a variable at a location to that of the same variable at neighboring locations. For

example, the “high-high and low-low locations suggest clustering of similar values,

whereas the high-low and low-high locations indicate spatial outliers” (Anselin 2003:

100).

Bivariate measures of spatial autocorrelation (again either Moran‟s I or LISA)

provide evidence of the link between the value of a variable at one location compared to

that of a different variable at neighboring locations (Anselin 2003). In other words, is

there local spatial dependence in a variable or between two variables, or does space not

matter? The LISA statistic is calculated by comparing counties with their neighbors. The

powerful aspect of this analysis is that the results provide visual evidence of both the

presence and type of local spatial autocorrelation for either univariate measures or in

bivariate relationships, conveniently displayed in a map to show where this localized

spatial autocorrelation exists (Anselin 2003).

In this study, comparisons are made using a Queens Weight Matrix, though others

were utilized for comparison with no significant differences found. To understand the

Queens Weight Matrix, think of a chessboard, and the manner in which different pieces

move. Queens have the most latitude in terms of the moves that they can make, in that

they can move in any direction (horizontally, vertically, and diagonally) on the board.

This type of matrix allows one to examine a fuller range of surrounding neighbors

(including those which share boundaries or vertices) to determine if there are similarities

across space (see Anselin 2002).

If there is statistically significant spatial autocorrelation, it can be characterized as

High/High (i.e. you and your neighbor(s) have a High score on the variable of interest, or

in a bivariate LISA, you have a High score on the X variable of interest and your

neighbor(s) have a High score on the Y variable of interest); Low/Low (you and your

neighbor(s) have a Low score on the variable of interest, or in a bivariate LISA, you have

a Low score on the X variable of interest and your neighbor(s) have a Low score on the Y

variable of interest), Low/High (you have a Low score, while your neighbor has a High

one, or in a bivariate LISA, you have a Low score on the X variable of interest and your

neighbor(s) have a High score on the Y variable of interest), or High/Low (you have a

High score, while your neighbor has a Low one, or in a bivariate LISA, you have a High

score on the X variable of interest and your neighbor(s) have a Low score on the Y

variable of interest).

Chapter Five

In Chapter Five, the primary objective is to address the central issue of whether or

not pollution is the “cost of doing business” for U.S. communities; i.e. is pollution a

necessary trade-off for jobs or higher wages in the local area? This debate has been

largely a rhetorical one; thus a central goal of this dissertation is to offer some empirical

direction for those interested in this important divide. To accomplish this goal, there are

two sub-questions which are addressed here through the use of cross-sectional

comparisons and regression-based analyses: 1) is pollution part of the cost of doing

business associated with job or wage growth in U.S. communities; and 2) has the “great

slide” or decline of manufacturing affected local pollution levels? However, the questions

are not addressed a-theoretically, as several hypotheses proposed in Chapter Two are

comparatively tested in this chapter.

To accomplish the goals of Chapter Five, regression modeling is utilized,

including zero-inflated negative binomial regression and ordinary least squares

regression. Zero-inflated negative binomial regression is utilized because the number of

TRI facilities is an over-dispersed count variable, characterized by excess zeroes. Poisson

regression, which is typically utilized for count variables, is inappropriate here due to the

distribution of the dependent variable (Long 1997). Zero-inflated negative binomial

regression utilizes two equations: one estimates the factors that produce zero scores while

the other equation models the factors that produce integer counts above one (Jacobs and

Carmichael 2004). Thus the model takes the form:

Pr (yi=0│xi) = ψi + (1 – ψi) exp (-µi)

Pr (yi│xi) = (1 - ψi) exp(-µi) µiyi /yi! for yi > 0 (Long 1997).

The other models in Chapter Five utilize Ordinary Least Squares (OLS)

regression. In OLS regression, the model takes the general form:

Yj = B0 + B1X1j + B2X2j+…+BkXkj + Ej where: “a dependent variable, Y, is

assumed to be a function of a set of k independent variables (or regressors), X1, X2, …

Xk, in some population” (Berry 1993: 3; italics in original). Additionally, an error term

(Ej) exists to account for “the deviation between the observed value of Y for observation

j, Yj, and the mean value of the Y distribution for cases in the population having the X

values of observation” (Berry 1993: 6; italics in original).

Chapter Six

In Chapter Six, the main goal is to address the spatial character and location of

domestic pollution havens in the United States. Domestic pollution havens are

conceptualized as areas which have higher levels of environmental pollution per unit of

economic reward and also lower levels of social controls, specifically environmental

regulations and enforcement actions. Several hypotheses are proposed to address the

factors which may impact the existence of a Domestic Pollution Haven. These

hypotheses were described in Chapter Two, and are comparatively examined in Chapter

Six.

To accomplish the goals of Chapter Six, regression modeling is utilized, including

logistic regression, negative binomial regression, and spatial maximum likelihood

regression. Logistic regression is utilized due to the dichotomous nature of the dependent

variable, with the logistic regression equation taking the form:

Ө = e (b0 + b1*x2+ b2*x2 …+ bk*xkj) ÷ 1 + e (b0 + b1*x2+ b2*x2 …+ bk*xkj) (see Menard 2002).

Negative binomial regression is utilized because the dependent variable is a count

variable. Poisson regression, which is typically utilized for count variables, is

inappropriate here due to the distribution of the dependent variable. The negative

binomial regression equation takes the form:

log( ) = b0 + b1*x1 + b2*x2…+ bk*xkj + е (see Long 1997).

The remaining regression analyses utilize Spatial Maximum Likelihood (ML)

regression models. Spatial regression is utilized to account for possible spatial

dependence which can lead to the violation of classical regression assumptions (Anselin

2005; Anselin and Bera 1998; Brasier 2002). The decision to use OLS estimators, or to

utilize Spatial Lag or Spatial Error models is based on an interpretation of the LaGrange

multiplier error and lag probability tests (for further explication, see Brasier 2002). Both

Spatial Lag and Spatial Error models are used to account for spatial dependence in the

model.

Spatial Lag equations take the form:

y = ρWy + βX + е where: “y is a vector of observations on the dependent

variable, Wy is a spatially lagged dependent variable for weights matrix W, X is a

matrix of observations on the explanatory variables, е is a vector of i.i.d. error terms,

and ρ and β are parameters” (Anselin 2005: 201). Spatial Lag models include a lag

coefficient, Rho, which accounts for “spatial processes in the dependent variable which

are not fully captured by the regressors in the model” (Land and Deane 1992: 223). In

other words, an unmeasured event or process is occurring which has an effect across

spatial units (Brasier 2002).

Spatial Error equations take the form:

y = Xβ + е, with e = λWе + μ

where: “y is a vector of observations on the dependent variable, W is the spatial weights

matrix, X is a matrix of observations on the explanatory variables, e is a vector of

spatially autocorrelated error terms, μ a vector of i.i.d. errors, and λ and β are parameters”

(Anselin 2005: 213). Spatial Error models are appropriate when the spatial dependence is

likely the result of measurement error within the model (Brasier 2002).

Conclusion

This multi-step analytical strategy will address the substantive and theoretical

issues discussed above by: 1) allowing for a more complex and spatially-explicit

examination of the economy-environment relationship over time and especially across

space, particularly through the use of spatial visualization, descriptive and inferential

statistics; 2) bringing together in common analyses indicators which capture the leading

theories addressing the economy-environment relationship, allowing for comparison and

critique of each approach; and 3) providing for the refinement and localization of

findings at the county-level, as opposed to prior studies which focused on the national or

state-levels. While useful in their own right, national or state-level analyses may cover up

patterns of spatial clustering or dispersion which occur at a more regional or local level

of social life. In the next chapter, descriptive analyses are performed as a first step

towards accomplishing these goals.

CHAPTER IV

SPATIAL AND TEMPORAL PATTERNS OF INDUSTRIAL GROWTH AND

DECLINE AND POLLUTION IN THE UNITED STATES, 1988-1997

This chapter presents a descriptive portrait of the spatial character and temporal

patterns of growth and decline within two key industrial sectors of the U.S. economy,

manufacturing and service, and toxic pollution from manufacturing, as measured in the

Toxic Release Inventory (TRI). As described in Chapters One and Two, the underlying

tension driving the analyses in this study involves the scale and scope of manufacturing

sector activity and the level of toxic pollution present in localities, a relationship which is

theorized to be positive in character (i.e. increased levels of manufacturing activity are

generally expected to be linked to increased levels of toxic pollution, and vice versa).

This tension is particularly problematic since, as was shown above, both economic

development and environmental protection are highly valued by most Americans.

As described in Chapter Three, the following three chapters offer analyses which

seek to describe, explore and model this underlying relationship in a variety of ways. In

this chapter, the initial focus is on characterizing the manufacturing and service sectors,

and toxic releases by manufacturing industries between 1988 and 1997, as well as

changes in these variables during this time period. This chapter also provides the first

attempt to empirically assess the bivariate relationship between changes across the

manufacturing and service sectors and the toxic release levels within localities, as well as

examining this relationship while controlling for census region or metropolitan status.

Finally, through the use of descriptive spatial statistical techniques, attention is paid to

the examination of the spatial concentration of manufacturing and service sector

composition, toxic release levels, and the changes in each over time. In sum, there are

four primary issues addressed in this description:

1. What is the composition and character of the manufacturing and service sectors

and of manufacturing‟s toxic releases in 1988 and 1997 across U.S. counties?

2. Did changes across manufacturing and/or service sectors affect levels of toxic

releases across U.S. counties between 1988 and 1997?

3. Was there a spatial pattern involved in changes which took place for industry

and/or toxic releases between 1988 and 1997? Specifically, are there regional or

metropolitan status differences in industry and/or toxic release changes between

1988 and 1997?

4. Are there significant spatial concentrations in the composition of manufacturing

and service sectors, or of changes across these sectors between 1988 and 1997?

Also, are there significant spatial concentrations of TRI facilities, releases, and

“dirty” facilities which overlap with significant clusters of manufacturing or

service change?

Addressing these questions through descriptive and exploratory approaches will facilitate

the use of more advanced statistical techniques in the subsequent analysis chapters. The

techniques utilized in this chapter were described in detail in Chapter Three.

In the next section, descriptive economic and toxic characteristics are presented.

Subsequent sections describe the changes over time in these characteristics, incorporate

bivariate analyses involving the economic and toxic characteristics, and bivariate

analyses assessing the relationship between spatial variables (Census region and

metropolitan status) and either the economic or toxic measures.

Descriptive Characteristics of the Manufacturing and Service Sectors, and of Toxic

Releases, 1988 and 1997

In Table 4.1, county-level means and standard deviations of economic

characteristics for 1988 and 1997 are reported in the manufacturing and service sectors,

as well as the overall economic conditions for the county. These economic characteristics

include number of employees, number of establishments, and average wage. Imputation

techniques were utilized where possible for manufacturing and service sector data,

though this involved only a handful of cases.

With respect to the average number of employees, there was an average growth of

5,594 employees per county between 1988 and 1997, from 27,934 to 33,528. Most of that

growth appears to have occurred in the service sector, where the average number of

employees increased by approximately 3,928 per county, from 7,971 to 11,869. In the

same time period, employment in the manufacturing declined slightly, from 6,248 to

6,079 employees per county, reflecting the continuation during the early and mid 1990s

of the general pattern of deindustrialization observed by Bluestone and Harrison (1982)

and others.

Table 4.1

Total Economic, Manufacturing and Service Characteristics, 1988 and 1997

1988

1997

S.D.

All employees

27934.10

115183.51

3107

33528.41

125454.92

3107

All establishments

1916.13

6774.29

3107

2196.94

7322.93

3107

Average wage

21289.60

5148.06

3102

21223.52

4918.84

3092

Manufacturing

employees

6247.90

24797.26

3067

6079.08

20572.09

3074

Manufacturing

establishments

117.75

511.32

3066

127.24

488.37

3074

Manufacturing

average wage

26858.73

8308.45

2909

27411.33

8911.93

2932

Service employees

7971.47

36527.25

3100

11869.29

49219.61

3106

Service

establishments

617.35

2372.47

3100

809.56

2917.71

3106

Service average

wage

16722.86

8562.77

3068

18312.51

4724.24

3052

Percent or

ratio

S.D.

Percent or

ratio

S.D.

Manufacturing

employees

26.39

18.07

3063

23.82

16.55

3063

Manufacturing

establishments

6.31

3.49

3066

6.42

3.37

3074

Manufacturing

wage Vs. total wage

1.25

0.25

2905

1.28

0.27

2921

Service employees

22.44

9.09

3095

29.04

10.88

3092

Service

establishments

27.57

4.70

3100

31.88

5.07

3106

Service wage Vs.

total wage

0.79

0.29

3063

0.87

0.16

3038

In terms of establishments, the pattern is similar, though growth occurred overall,

as well as across the service and manufacturing sectors. The average number of

establishments per county grew by 281, from 1,916 to 2,197. Most of the growth (193, or

69%) in total number of establishments came in the service sector, where the number of

establishments grew from 617 to 810. However, the number of manufacturing

establishments also grew slightly, from 118 to 127, an increase in facilities of 7.6%.

Regarding wages, the adjusted average wage per employee is reported for 1988

and the average wage per employee is reported for 1997 (1988 wages = $1.36 to 1997

wages = $1) (BLS 2006). Wages grew between 1988 and 1997 across sectors and overall;

though inflation-adjusted overall wages actually declined slightly, while average wages

grew in both the manufacturing and service sectors. The overall average wage between

1988 and 1997 declined by $66 (-0.31%) when inflation is factored in, going from

$21,290 to $21,224. Wages in the service sector grew from $16,723 to $18,313, an

increase of $1,590 or 9.51% when inflation is considered. Wages in the manufacturing

sector grew by $552 (2.06%) when adjusting for inflation, growing from $26,859 in 1988

to $27,411 in 1997.

As relates to the “de-industrialization of America” literature, it would appear that

there is something to it across counties during the time period of interest (1988-1997), as

the average number of employees in manufacturing per county declined while the

average number of establishments increased slightly (though at a much smaller clip than

for the service sector or overall). Also of interest is that this is a period generally

characterized as moving from relatively stagnant economic conditions to one in which

significant growth occurred in the American economy, yet when controlling for inflation

overall wages actually declined. However, inflation-adjusted wages grew slightly for

manufacturing sector employees and significantly for service sector employees during

this time period, pointing to improved economic conditions for the households of

workers in these important sectors of the economy.

In the bottom panel of Table 4.1, relative percentages or ratios are reported for the

manufacturing and service sector data on employment, establishments, and wages, as

compared with the overall county-level data for these respective variables. For 1988, the

manufacturing sector had 26.39% of the average county-level employees, compared to

22.44% for the service sector. This reversed by 1997, with the service sector increasing to

29.04% of the total employees (an increase of 6.6%), and the manufacturing sector

declining to 23.82% (a decrease of 2.57%). The percentage of establishments in

manufacturing increased slightly between 1988 and 1997, from 6.31% to 6.42% (an

increase of 0.11%), while the percentage of establishments in service increased by

4.31%, from 27.57% in 1988 to 31.88% in 1997. Finally, the proportion of the

manufacturing average wage versus the overall average wage increased slightly between

1988 and 1997, from 1.25 to 1.28. This means that on average, the county-level average

wage for manufacturing increased from 25% to 28% higher than the overall average

wage. The proportion of the service average wage versus the overall average wage also

increased between 1988 and 1997, going from 0.79 to 0.87. This means that on average,

the countylevel average wage for service ranged from 21% to 13% less than the overall

average wage. Thus while relative gains were made, service sector employees still fare

worse than the average American worker regarding wages, while manufacturing sector

employees fare substantially better, a point that those seeking to protect manufacturing

activity often note.

In Table 4.2, county-level toxic characteristics for 1988 and 1997 are described.

The toxic characteristics include TRI pounds, number of TRI facilities, TRI pounds per

facility, and Dirtiness Index scores. The average number of pounds of toxic releases per

county in 1988 was 1,927,459. This grew by 374,116 pounds (a 19.4% increase) between

1988 and 1997, to 2,301,575 pounds per county in 1997. The average number of TRI

facilities per county in 1988 was 6.54. This increased slightly in 1997 to 7.01, an increase

of .47 average facilities per county (a 7.2% increase). The average number of pounds of

toxic releases per facility per county in 1988 was 327,009. This increased to 368,575 in

1997, an increase of 41,566 pounds of toxic releases per facility per county, a growth of

12.7%. Finally, the average Dirtiness Index Score in 1988 was 0.14. This declined by

0.01 in 1997, to an average score of 0.13.

Table 4.2

Toxic Release Characteristics, 1988 and 1997

1988

1997

S.D.

TRI pounds

1927458.80

9758561.20

3107

2301574.70

10872244.00

3107

TRI facilities

6.54

22.38

3107

7.01

21.73

3107

TRI pounds/

facility

327009.02

1379643.00

2118

368574.66

1674638.70

2184

Dirtiness Index

Score

0.14

0.13

2801

0.13

2848

As relates to the relationship between manufacturing sector composition and

change and the level of toxic releases, thus far it would appear that at the aggregate level,

the growth in manufacturing facilities and the growth in toxic facilities and releases

correspond rather well. This could indicate that those new manufacturing facilities

became part of the TRI at roughly the same proportion as the existing manufacturing

facilities, though it is just as plausible that facilities which closed during this time were

significantly more or less likely to be TRI-reporting facilities.

In Table 4.3, frequencies are presented for census region and metropolitan area

status. The U.S. census regions in 1990 were as follows: Northeast (Connecticut, Maine,

Massachusetts, New Hampshire, New Jersey, New York, Pennsylvania, Rhode Island,

Vermont); Midwest (Indiana, Illinois, Iowa, Kansas, Michigan, Minnesota, Missouri,

Nebraska, North Dakota, Ohio, South Dakota, Wisconsin); West (Arizona, California,

Colorado, Idaho, Montana, Nevada, New Mexico, Oregon, Utah, Washington,

Wyoming); and South (Alabama, Arkansas, Delaware, Florida, Georgia, Kentucky,

Louisiana, Maryland, Mississippi, North Carolina, Oklahoma, South Carolina, Tennessee,

Texas, Virginia, West Virginia). Alaska and Hawaii are part of the West region, but are

excluded from consideration in this dissertation.

Table 4.3

Distribution of Counties by Region and Metropolitan Proximity

Frequency

Percent

Region, 1990

Northeast

217

7.0

Midwest

1055

34.0

West

413

13.3

South

1422

45.8

Metro Proximity, 1993

Metro

833

26.8

Adjacent

998

32.1

Not Adjacent

1276

41.1

Regarding metropolitan area status, the classification of counties in 1993 was

either: part of metropolitan area, adjacent to metropolitan area, or non-

metropolitan/notadjacent county. Subsequent analyses in this chapter will break down

changes in economic status and toxic releases by census region and metropolitan area. In

the next section, temporal and spatial changes in the manufacturing and service sectors,

as well as in toxic releases, will be described.

Temporal and Spatial Change in the Manufacturing and Service Sectors, and of

Toxic Releases, 1988-97

In this section, descriptive analyses will be presented which involve temporal and

spatial patterns of change between 1988 and 1997, both in the relative strength of the

manufacturing and service sectors, as well as the scale of toxic releases. In Table 4.4,

county-level means and standard deviations of percentage change between 1988 and

1997 for economic characteristics are reported for the manufacturing and service sectors,

as well as the overall economic conditions for the county. These economic characteristics

include number of employees, number of establishments, and average wage.

Table 4.4

Total Economic, Manufacturing and Service Sector Percent Change, 1988-1997

Percent change

S.D.

All employees

31.85%

63.68

3102

All establishments

17.43%

22.30

3107

Average wage

0.62%

13.06

3089

Manufacturing

employees

-2.60%

13.25

3043

Manufacturing

establishments

0.12%

1.68

3055

Manufacturing

average wage

4.53%

27.06

2845

Service employees

6.62%

8.31

3084

Service

establishments

4.32%

3.27

3100

Service average

wage

13.33%

22.66

3031

Comparisons for average wage are made between the 1988 inflation adjusted

figure and the 1997 amount. With respect to the average number of employees, there was

overall average growth of 31.85% per county between 1988 and 1997. Growth occurred

in the service sector, where the proportional percentage of service employees increased

by 6.62%. In the same time period, proportional employment in the manufacturing sector

declined by 2.60%. In terms of establishments, growth occurred overall, as well as

proportionally in the service and manufacturing sectors. The average of establishments

per county grew 17.43%. Average growth in the service sector again outpaced growth in

the manufacturing sector, as the average percentage of establishments in services

increased by 4.32% and the average percentage of establishments in manufacturing grew

only 0.12%. As regards wages, the average wage grew between 1988 and 1997 across

sectors and overall. County-level inflation-adjusted overall wages grew by 0.62%, while

proportional wages in the service sector increased 13.3%, and proportional wages in the

manufacturing sector grew 4.53%.

Table 4.5

Toxic Release Change, 1988-1997

Percent or

Proportion change

S.D.

TRI pounds

3585.98%

74709.22

2118

TRI facilities

16.25%

61.07

2218

TRI pounds/facility

3518.05%

76258.08

2017

Dirtiness Index

Score

-0.01

0.12

2702

In Table 4.5, the percentage change between 1988 and 1997 for county-level toxic

characteristics is described. The toxic characteristics include TRI pounds, number of TRI

facilities, TRI pounds per facility, and Dirtiness Index scores. The average number of

pounds of toxic releases per county increased 3585.98%. The average number of TRI

facilities per county also increased by 16.25%. The average number of pounds of toxic

releases per facility per county was up 3518.05%. Interestingly, the average Dirtiness

Index score decreased between 1988 and 1997 by 0.01 units.

Categorization of the Relationship between Changes in the Manufacturing and

Service Sectors, 1988-1997

Table 4.6 presents a characterization of the relationship between changes in the

manufacturing and the service sectors between 1988 and 1997 for employment, wages,

and establishments. The general argument underlying this cross-classification of

manufacturing and services is laid out in Chapters One and Two, though it is worthwhile

to go through this logic again. Essentially, this period of time has been characterized as

one in which localities (counties in these analyses) were likely to have experienced

deindustrialization (manufacturing decline) (Bluestone and Harrison 1982; Rowthorn and

Ramaswamy 1997), with many recovering from those negative economic impacts by

growing in the service sector. Others experienced decline in both sectors, growth in both

sectors or experienced growth in manufacturing and decline in services.

To construct the categories introduced in Table 4.6, thresholds of growth or

decline were chosen for employment, wages, and establishments. These thresholds were

based on sensitivity analyses, conducted to determine the optimal percentage of growth

or decline. In the case of employment and wages, 3% growth or decline was set as the

minimum threshold; i.e. if either manufacturing or services grew or declined by less than

3%, that county fell into the stability category. For establishments, any growth or decline

was utilized for the threshold due to the high number of counties which had less than 1%

growth or decline across the manufacturing and service sectors.

Table 4.6

Categorization of Changes in the Manufacturing and Service Sectors, 1988-1997

Manufacturing

Decline

Stability

Growth

Decline

Employment

Wages

Establishments

Local economic

decline

1.1%

5.5%

2.2%

Shift to manufacturing

economy

2.3%

6.5%

3.4%

Employment 47.9%

Wages 36.9%

Note: The values in each cell represent the percentage of counties which meet the

described cross-classification criteria of manufacturing and service sector changes

between 1988 and 1997.

As indicated in the table, the values in each cell represent the percentage of

counties which meet the described cross-classification criteria of manufacturing and

service sector changes between 1988 and 1997. The first set of values refer to the

percentage of counties which meet the cross-classification of manufacturing and service

Establishments

1.9%

Growth

Employment

Wages

Establishments

Shift to service

economy

40.7%

23.0%

42.9%

Diverse local

economic growth

8.0%

28.1%

49.6%

sector changes in employment during the 1988-1997 time frame. The possible outcomes

include: stability - a less than 3% increase or a less than 3% decrease in both the

manufacturing and service sector employment levels; local economic decline – a decline

of greater than 3% in both the manufacturing and service sector employment; shift to a

service economy – a decline of greater than 3% in manufacturing and an increase of more

than 3% in service employment; shift to a manufacturing economy – an increase of more

than 3% in manufacturing and a decrease of more than 3% in service employment; and

diverse local economic growth – an increase of more than 3% in employment in both

sectors. The largest category is stability, with 47.9% of the counties experiencing less

than +/-3% change in both manufacturing and service employment. The next largest

category is the shift to a service economy, as 40.7% of the counties experienced more

than 3% growth in relative service employment while also experiencing more than a 3%

decline in relative manufacturing employment. Diverse local economic growth in

employment was found in 8% of the counties, while 2.3% experienced a shift to a

manufacturing economy and 1.1% had local economic decline, with both manufacturing

and service employment declining by more than 3%.

Spatially, counties which fit the shift to a service economy category are

particularly found in the eastern part of the U.S. and in the Pacific Northwest, while most

of the counties that experienced diverse local economic growth or the shift to a

manufacturing economy are in the Great Plains or the Mountain West.

The next row of values in Table 4.6 refer to the percentage of counties which meet

the cross-classification of manufacturing and service sector changes in wages during the

1988-1997 time frame. Again, the possible outcomes include: stability - a less than 3%

increase or a less than 3% decrease in both manufacturing and service sector wages; local

economic decline – a decline of greater than 3% in both manufacturing and service sector

wages; shift to a service economy – a decline of greater than 3% in manufacturing and an

increase of more than 3% in service wages; shift to a manufacturing economy – an

increase of more than 3% in manufacturing and a decrease of more than 3% in service

wages; and diverse local economic growth – an increase of more than 3% in wages for

both sectors. The largest category is stability, with 36.9% of the counties experiencing

less than +/-3% change in both manufacturing and service wages. The next largest

category is the diverse local economic growth in wages was found in 28.1% of the

counties. The shift to a service economy is observed in 23% of the counties, meaning that

these areas experienced more than 3% growth in relative service wages while also

experiencing more than a 3% decline in relative manufacturing wages. 6.5% experienced

a shift to a manufacturing economy and 5.5% had local economic decline, with both

manufacturing and service wages declining by more than 3%.

The spatial patterning of counties is rather different for the classification of wages

than for employment. Diverse local economic growth, much more common for wages

than for employment, is found in clusters throughout the country, though particularly so

along the East Coast and in the Southeast. The shift to a service economy is particularly

found in the Great Plains and the Midwest, while most of the counties that experienced

the shift to a manufacturing economy or local economic decline are relatively dispersed

throughout the country.

The third and final row of values in Table 4.6 includes the percentage of counties

which meet the cross-classification of manufacturing and service sector changes in

establishments between 1988 and 1997. The possible outcomes here include: stability –

the absence of an increase or a decrease in both manufacturing and service sector

establishments; local economic decline – a decrease in both manufacturing and service

sector establishments; shift to a service economy – a decrease in manufacturing

establishments and an increase in service establishments; shift to a manufacturing

economy – an increase in manufacturing establishments and a decrease in service

establishments; and diverse local economic growth – an increase in establishments across

both sectors. The largest category is diverse local economic growth, which 49.6% of the

counties experienced. Close behind is the shift to a service economy, observed in 42.9%

of the counties. 3.4% experienced a shift to a manufacturing economy, 2.2% had local

economic decline and only 1.9% experienced stability in the establishment trend across

manufacturing and service sectors.

Again, the pattern is rather different for establishments than for the other

measures. Diverse local economic growth is observed in the interior Northeast, the

Southeast, the Great Plains, Mountain West, and in Southern California, while the shift to

a service economy is most commonly observed along the immediate East Coast and in

the

Pacific Northwest.

Cross-Tabulation of Manufacturing and Service Sector Changes with Toxic

Changes, 1988-1997

Tables 4.7 through 4.18 each contain the cross-tabulation of one of the

manufacturing and service sector change characterization variables (either for employees,

establishments, or wages) with one of the toxic change characterization variables (change

in TRI facilities, change in TRI pounds, change in TRI pounds/facility, and change in

Dirtiness Index).

Tables 4.7 through 4.9 present cross-tabulations of manufacturing and service

sector change with the characterization of change in TRI facilities, between 1988 and

1997. In Table 4.7, the cross tabulation of changes in manufacturing and service sector

employment and change in TRI facilities is presented. The categories for the

manufacturing and service sector changes are described in the above section. For each of

the toxic release variables, the characterization of change is based on the percentile

distribution of the percentage of change between 1988 and 1997 in the selected variable.

The categories include: bottom 25%, middle 50%, and top 25%. For the change in TRI

facilities variable, the bottom 25% of counties are those in which the percentage change

is -5.26% or less; the middle 50% of counties fall between -5.25% and 33.63%; and the

top 25% of counties experienced change at or above 33.64%. Using this approach allows

comparisons across categories of the manufacturing and service sector variable to be

made based on the deviation from the three percentiles used for the toxic change variable.

Table 4.7

Cross tabulations of Changes in Manufacturing (Mfg) and Service (Svc) Employment

and Change in TRI Facilities, 1988-1997

Characterization of % change, TRI facilities

Bottom 25% Middle 50% Top 25%

Stability 22.5% 51.4% 26.1%

Growth both 19.8% 50.5% 29.7% Growth mfg/ 16.7% 40% 43.3%

decline svc

Decline mfg/ 28.2% 48.6% 23.1% growth svc

Decline both 18.2% 81.8% 0%

For example, for the category “growth in both manufacturing and service sector

employment”, 19.8% of the counties in this category fall within the bottom 25% of

change in TRI facilities (i.e. the largest percentage decline in releases), 50.5% of the

counties fall within the middle 50% of change in TRI facilities, and 29.7% of the

counties fall within the top 25% of change in TRI facilities. This indicates that counties

with growth in both manufacturing and service sector employment between 1988 and

1997 are somewhat more likely to experience the highest percentage of increases in

releases. However, this relationship is much more pronounced for counties experiencing

growth in manufacturing and decline in service employment, as 43.3% of the counties in

this category experienced the highest level of growth in the percentage of TRI facilities

between 1988 and 1997. Finally, those counties experiencing a decline in manufacturing

employment and growth in service employment are slightly more likely to fall in the

bottom 25% of percentage change in TRI facilities.

In Table 4.8, the cross tabulation of changes in manufacturing and service sector

wages and change in TRI facilities is presented. There is a less clear pattern of difference

here, though for the growth in manufacturing and decline in service wages, the

percentages are slightly elevated for the extremes (bottom and top 25%) relative to the

middle 50%. The reverse is evident when decline occurs in the wages in both sectors,

with the percentage of those falling in the middle 50% of the percentage change in toxic

facilities being 57.1%, and the extremes falling to 22.3% and 20.5% respectively.

Table 4.8

Cross tabulations of Changes in Manufacturing (Mfg) and Service (Svc) Wages and

Change in TRI Facilities, 1988-1997

Characterization of % change, TRI facilities

Bottom 25% Middle 50% Top 25%

Stability 24.6% 50% 25.4%

Growth both 26.6% 48.6% 24.8%

Growth mfg/ 27.6% 45.7% 26.8% decline svc

Decline mfg/ 23.6% 51.3% 25.1% growth svc

Decline both 22.3% 57.1% 20.5%

In Table 4.9, the cross tabulation of changes in manufacturing and service sector

establishments and change in TRI facilities is presented. Of note here is that for the

growth in manufacturing and decline in service wages, 34% of the counties fall in the top

25% of percentage change in TRI facilities, while when both service and manufacturing

facilities experience declines, 34.6% of the counties are in the bottom 25% of percentage

change in TRI facilities.

Table 4.9

Cross tabulations of Changes in Manufacturing (Mfg) and Service (Svc) Establishments

and Change in TRI Facilities, 1988-1997

Characterization of % change, TRI facilities

Bottom 25% Middle 50% Top 25%

Stability 50% 50% 0%

Growth both 24.3% 49.5% 26.2% Growth mfg/ 23.4% 42.6% 34%

decline svc

Decline mfg/ 25.6% 50.9% 23.5% growth svc

Decline both 34.6% 50% 15.4%

Tables 4.10 through 4.12 present cross-tabulations of manufacturing and service

sector change with the characterization of change in TRI pounds between 1988 and 1997.

For the change in TRI pounds variable, the bottom 25% of counties are those in which

the percentage change is -45.75% or less; the middle 50% of counties fall between

45.74% and 148.18%; and the top 25% of counties experienced change at or above

148.19%.

Table 4.10

Cross tabulations of Changes in Manufacturing (Mfg) and Service (Svc) Employment

and Change in TRI Pounds, 1988-1997

Characterization of % change, TRI pounds

Bottom 25% Middle 50% Top 25%

Stability 24.5% 48% 27.5%

Growth both 27% 44.1% 28.8%

Growth mfg/ 33.3% 20% 46.7% decline svc

Decline mfg/ 24.6% 53.6% 21.8% growth svc

Decline both 54.5% 36.4% 9.1%

In Table 4.10, the cross tabulation of changes in manufacturing and service sector

employment and change in TRI pounds is presented. Most notable is the distribution for

the growth in manufacturing and decline in service category, in which 46.7% of the

counties fall within the top 25% of percentage change in TRI pounds, 33.3% of these

counties fall within the bottom 25% of percentage change in TRI pounds, and only 20%

of these counties fall in the middle 50%. Also of interest is that in those counties where

both manufacturing and service employment decreased by 3% or more, it was much more

likely that the county fell in the bottom 25% of percentage change in TRI pounds

(54.5%) and less likely that the county fell in the top 25% (9.1%).

Table 4.11

Cross tabulations of Changes in Manufacturing (Mfg) and Service (Svc) Wages and

Change in TRI Pounds, 1988-1997

Characterization of % change, TRI pounds

Bottom 25% Middle 50% Top 25%

Stability 22.9% 51.6% 25.5%

Growth both 26.1% 50.8% 23.1%

Growth mfg/ 30.7% 41.7% 27.6% decline svc

Decline mfg/ 24.8% 49.1% 26.1% growth svc

Decline both 26.8% 49.1% 24.1%

In Table 4.11, the cross tabulation of changes in manufacturing and service sector

wages and change in TRI pounds is presented. As was the case with wages in Table 4.8,

there is a less clear pattern of difference here. The most notable difference involves the

growth in manufacturing and decline in service wages category, where the percentages

are slightly elevated for the extremes (bottom and top 25%) relative to the middle 50%.

In Table 4.12, the cross tabulation of changes in manufacturing and service sector

establishments and change in TRI pounds is presented. The most apparent deviation here

involves the decline in both manufacturing and service wages category, where 57.7% of

Stability

Growth both

Growth mfg/

decline svc

Decline mfg/

growth svc

Decline both

54.5

18.2

27.3

the counties fell in the bottom 25% of percentage change in TRI pounds, and only 7.7%

fell in the top 25%.

Table 4.12

Cross tabulations of Changes in Manufacturing (Mfg) and Service (Svc) Establishments

and Change in TRI Pounds, 1988-1997

Characterization of % change, TRI pounds

Bottom 25% Middle 50% Top 25%

Stability 100% 0% 0%

Growth both 25.6% 48.7% 25.7%

Growth mfg/ 25.5% 48.9% 25.5% decline svc

Decline mfg/ 23.1% 52.1% 24.7% growth svc

Decline both 57.7% 34.6% 7.7%

Tables 4.13 through 4.15 present cross-tabulations of manufacturing and service

sector change with the characterization of change in TRI pounds per facilities between

1988 and 1997. For the change in TRI pounds per facility variable, the bottom 25% of

counties are those in which the percentage change is -44.82% or less; the middle 50% of

counties fall between -44.81% and 128.84%; and the top 25% of counties experienced

change at or above 128.85%.

Table 4.13

Cross tabulations of Changes in Manufacturing (Mfg) and Service (Svc) Employment

and Change in TRI Pounds per Facility, 1988-1997

Characterization of % change, TRI LBs per facility

Bottom 25% Middle 50% Top 25%

25.4% 48% 26.7%

23.8% 39.6% 36.6% 35.7% 17.9% 46.4%

24.2% 54.2% 21.6%

In Table 4.13, the cross tabulation of changes in manufacturing and service sector

employment and change in TRI pounds per facilities is presented. Several of the

categories present results of interest here. First, for counties with growth in both

manufacturing and service sector employment, the percentage change in TRI pounds per

facility is much more likely to fall in the top 25% (36.6%). However, for counties with

growth in manufacturing and declines in service employment, the percentage change in

TRI pounds per facility is even more likely to land in the top 25% (46.4%). When

considering counties with declines in manufacturing and increases in service

employment, the pattern shifts, with only 21.6% falling in the top 25% of change in TRI

pounds per facility. When declines in both manufacturing and service sector employment

occur, polarization to the extremes takes place, as 27.3% of the counties are in the top

25% of change in TRI pounds per facility and 54.5% of the counties are in the bottom

25%.

Table 4.14

Cross tabulations of Changes in Manufacturing (Mfg) and Service (Svc) Wages and

Change in TRI Pounds per Facility, 1988-1997

Characterization of % change, TRI Pounds per

facility

Bottom 25%

Middle 50

Top 25%

Stability

50.8

24.2

Growth both

25.3

51.9

22.8

Growth mfg/

decline svc

27.7

40.3

31.9

Decline mfg/

growth svc

23.3

49.5

27.2

Decline both

46.7

25.2

In Table 4.14, the cross tabulation of changes in manufacturing and service sector

wages and change in TRI pounds per facilities is presented. Again, as was the case with

wages above, there is a less clear pattern of difference here. The most notable difference

involves the growth in manufacturing and decline in service wages category, where the

percentages are slightly elevated for the extremes (bottom and top 25%) relative to the

middle 50%, particularly for the top 25% (31.9%). Also, for declines in manufacturing

and growth in service wages, there is a somewhat higher percentage in the top 25%

(27.2%) than in the bottom 25% (23.3%). In the decline in both manufacturing and

service wages category, 28% of the counties fall in the bottom 25% of percentage change

in TRI pounds per facility.

Table 4.15

Cross tabulations of Changes in Manufacturing (Mfg) and Service (Svc) Establishments

and Change in TRI Pounds per Facility, 1988-1997

In Table 4.15, the cross tabulation of changes in manufacturing and service sector

establishments and change in TRI pounds per facilities is presented. A notable difference

involves the growth in manufacturing and decline in service wages category, where the

Characterization of % change, TRI Pounds per

facility

ottom 25%

Middle 50%

Top 25%

Stability

100

Growth both

26.1

48.6

25.3

Growth mfg/

decline svc

18.6

58.1

23.3

Decline mfg/

growth svc

23.8

51.2

Decline both

38.1

47.6

14.3

percentages are elevated for the middle 50% of change in TRI pounds per facility

(58.1%) versus the extremes (18.6% for the bottom 25% and 23.3% for the top 25%).

Also, for the decline in both service and manufacturing establishments, 38.8% of the

counties fall in the bottom 25% of percentage change in TRI pounds per facility, and only

14.3% are in the top 25%.

Tables 4.16 through 4.18 present cross-tabulations of manufacturing and service

sector change with the characterization of change in the Dirtiness Index between 1988

and 1997. For the change in the Dirtiness Index variable, the bottom 25% of counties are

those in which the change is -0.0430 or less; the middle 50% of counties fall between -

0.0429 and 0.0250; and the top 25% of counties experienced change of 0.0251 or greater.

Table 4.16

Cross tabulations of Changes in Manufacturing (Mfg) and Service (Svc) Employment

and Change in Dirtiness Index, 1988-1997

Characterization of change, Dirtiness Index

Bottom 25% Middle 50% Top 25%

Stability 23.6% 51% 25.4%

Growth both 20.5% 51.5% 28% Growth mfg/ 20% 51.7% 28.3%

decline svc

Decline mfg/ 27.7% 48.5% 23.7% growth svc

Decline both 17.9% 64.3% 17.9%

In Table 4.16, the cross tabulation of changes in manufacturing and service sector

employment and change in the Dirtiness Index is presented. For both growth in both

manufacturing and service sector employment and growth in manufacturing and decline

in service, there is a slightly elevated likelihood of falling in the top 25% (28% and

28.3% respectively) than the bottom 25% (20.5% and 20% respectively). This reverses

for the decline in manufacturing and increase in service employment, with 27.7% of the

counties in this category falling in the bottom 25% of change in the Dirtiness Index.

Finally, when declines in both manufacturing and service employment occur, 64.3% of

the counties fall in the middle 50%, and 17.9% land in each extreme category.

Table 4.17

Cross tabulations of Changes in Manufacturing (Mfg) and Service (Svc) Wages and

Change in Dirtiness Index, 1988-1997

Characterization of change, Dirtiness Index

Bottom 25% Middle 50% Top 25%

Stability 24.3% 52.8% 22.9%

Growth both 23.5% 49.3% 27.2%

Growth mfg/ 21.3% 53.6% 25.1% decline svc

Decline mfg/ 28.2% 46.3% 25.5% growth svc

Decline both 28.7% 49.3% 22%

In Table 4.17, the cross tabulation of changes in manufacturing and service sector

wages and change in the Dirtiness Index is presented. Again, there is very little deviation

for the wages, though the general pattern holds, that counties which experience growth

are slightly more likely to fall in the top 25% than the bottom 25%, and vice versa.

In Table 4.18, the cross tabulation of changes in manufacturing and service sector

establishments and change in the Dirtiness Index is presented. Here, counties which

experience growth in manufacturing and declines in service establishments are more

likely to land in the top 25% of change in the Dirtiness Index (31.4%) than the bottom

25% (14.3%). Interestingly, counties with growth in both manufacturing and service

establishments are actually slightly more likely to fall in the bottom 25% of change in the

Dirtiness Index (28.3%) than in the top 25% (23.7%).

Table 4.18

Cross tabulations of Changes in Manufacturing (Mfg) and Service (Svc) Establishments

and Change in Dirtiness Index, 1988-1997

Characterization of change, Dirtiness Index

Bottom 25% Middle 50% Top 25%

Stability 33.3% 66.7% 0%

Growth both 28.3% 48.1% 23.7%

Growth mfg/ 14.3% 54.3% 31.4% decline svc

Decline mfg/ 22.2% 52% 25.9% growth svc

Decline both 18.9% 54.7% 26.4%

Cross-Tabulation of Manufacturing-Service Sector Changes and Toxic Changes

with Census Regions, 1988-1997

Tables 4.19 through 4.25 each contain the cross-tabulation of one of the

manufacturing and service sector change characterization variables (either for employees,

establishments, or wages) or one of the toxic change characterization variables (change in

TRI facilities, change in TRI pounds, change in TRI pounds/facility, and change in

Dirtiness Index) with the census region variable. The categories for the manufacturing

and service sector change variables, and the toxic change variables, are described above.

As noted earlier, the census regions in the U.S. include: the Northeast, Midwest, West,

and South. Table 4.3 describes the distribution of counties in each of these regions. Using

this approach allows comparisons across categories of the manufacturing and service

sector variable or toxic variable to be made based on the percentage differences across

Census regions.

Table 4.19

Cross tabulations of Changes in Manufacturing (Mfg) and Service (Svc) Employment,

1988-1997 and Census Region

In Table 4.19, the cross tabulation of changes in manufacturing and service sector

employment by census region is presented. In the Northeast, decline in manufacturing

and growth in service employment is the most common pattern, observed in 65.9% of the

counties. Stability (less than +/- 3% change in both manufacturing and service) is much

less common here than in the other regions (30.9%). In the Midwest, stability (56.3%)

and a decline in manufacturing and increase in service employment (30.5%) are the most

common occurrences, but growth in both sectors (9.4%) and growth in manufacturing

and a decline in service (2.7%) are more common here than in other regions of the

country. The West region distribution is very similar to the Midwest, with a slightly lower

Census Region

Northeast

Midwest

West

South

Stability

30.9

56.3

57.4

41.6

Growth both

2.3

9.4

6.3

8.4

Growth

mfg/decline

svc

0.5

2.7

2.2

Decline

mfg/growth

svc

65.9

30.5

32.9

46.7

Decline both

0.5

1.2

percentage of growth in both manufacturing and service employment (6.3%), and a

slightly higher percentage of growth in service and a decline in manufacturing (32.9%).

In the South, stability is less common than in the Midwest or the West (41.6%), with a

higher percentage decline in manufacturing and growth in service than in the Midwest or

the West (46.7%).

Table 4.20

Cross tabulations of Changes in Manufacturing (Mfg) and Service (Svc) Wages, 1988-

1997 and Census Region

In Table 4.20, the cross tabulation of changes in manufacturing and service sector

wages by census region is presented. In the Northeast, growth in wages across both

sectors is the most common pattern, observed in 38.7% of the counties, though stability is

close behind (35.9%). Growth in manufacturing and decline in service wages is more

common than in the Midwest or the West (7.4%). The decline in manufacturing wages

and growth in service wages (14.3%) and the decline across both sectors (3.7%) is less

common than in the other regions. In the Midwest, stability (38%) and a decline in

Census Region

Northeast

Midwest

West

South

Stability

35.9

34.1

36.9

Growth both

38.7

24.4

23.7

30.6

Growth

mfg/decline

svc

7.4

4.6

4.4

8.4

Decline

mfg/growth

svc

14.3

27.5

28.6

19.3

Decline both

3.7

5.5

9.2

4.8

manufacturing and increase in service wages (27.5%) are the most common occurrences,

though growth in wages in both sectors (24.4%) is close behind. The West region

distribution is very similar to the Midwest, with a slightly lower percentage of growth in

both manufacturing and service employment (23.7%) and stability (34.1%), and a slightly

higher percentage of growth in service and a decline in manufacturing (28.6%). In the

South, stability is the most common (36.9%) with growth in both slightly behind

(30.6%), with a higher percentage decline in manufacturing and growth in service than in

the other regions (8.4%).

Table 4.21

Cross tabulations of Changes in Manufacturing (Mfg) and Service (Svc) Establishments,

1988-1997 and Census Region

In Table 4.21, the cross tabulation of changes in manufacturing and service sector

establishments by census region is presented. In the Northeast, growth in establishments

across both sectors (49.8%), and decline in manufacturing and growth in service

establishments (50.2%) are the only two categories which occur. In the Midwest, 54.2%

Census Region

Northeast

Midwest

West

South

Stability

3.6

1.7

Growth both

49.8

54.2

44.8

47.5

Growth

mfg/decline

svc

3.6

4.6

3.4

Decline

mfg/growth

svc

50.2

37.8

44.1

45.1

Decline both

2.4

2.9

2.3

of the counties experienced growth of establishments across sectors, which was a higher

percentage than in other regions. 37.8% of the counties experienced a decline in

manufacturing and growth in service establishments, less common than in the other

regions. In the West, there is a lower percentage of growth in both manufacturing and

service employment (44.8%) than in the Midwest, and a higher percentage of counties

experiencing a decline in manufacturing and increase in service establishments versus the

Midwest (44.1%). The pattern in the South is similar to the other regions, with 47.5% of

the counties experiencing growth in establishments across both sectors, and 45.1%

having a decline in manufacturing and an increase in service establishments.

Table 4.22

Cross tabulations of Change in TRI Facilities, 1988-1997 and Census Region

In Table 4.22, the cross tabulation of changes in TRI facilities by census region is

presented. In the Northeast, almost half (49.3%) of the counties fell into the bottom 25%

of percentage change in TRI facilities, which is considerably higher than the other

regions. By comparison, only 18.9% of Midwest counties fell into this bottom category,

while 28.3% of the counties were in the top 25% of TRI facility percentage change. In

the

Census Region

Northeast

Midwest

West

South

Bottom

49.3

18.9

32.2

Middle

36.9

52.7

47.8

51.2

Top 25%

13.8

28.3

25.9

West, 32.2% of the counties fell in the bottom 25%, and only 20% were in the top 25% of

TRI facility change. The South fell relatively close to the percentage distribution, with a

slightly smaller percentage (23%) being in the bottom 25% category, and a slightly

higher percentage (25.9%) falling in the top 25% of percentage change for TRI facilities.

Table 4.23

Cross tabulations of Change in TRI Pounds, 1988-1997 and Census Region

In Table 4.23, the cross tabulation of changes in TRI pounds by census region is

presented. In the Northeast, 57.1% of the counties fell within the middle 50% of

percentage change in TRI pounds, while only 18.7% were in the top quarter. In the

Midwest, 27.5% of the counties fell into the top quarter of TRI pound percentage change,

and 22.2% were in the bottom quarter. In the West, 30.7% of the counties fell in the

bottom 25%, and only 21% were in the top 25% of TRI facility change. The South fell

relatively close to the percentage distribution, with a slightly higher percentage (26%)

being in the bottom 25% category than in the top 25% of percentage change for TRI

facilities (25.3%).

In Table 4.24, the cross tabulation of changes in TRI pounds per facility by census

region is presented. Given that this variable is linked to the prior two variables (TRI

Census Region

Northeast

Midwest

West

South

Bottom

24.1

22.2

30.7

Middle

50.4

48.3

48.7

Top 25%

18.7

27.5

25.3

pounds and TRI facilities), it is not surprising that the distribution is virtually identical to

the above results.

Table 4.24

Cross tabulations of Change in TRI Pounds per Facility, 1988-1997 and Census Region

In the Northeast, 53.2% of the counties fell within the middle 50% of percentage change

in TRI pounds per facility, while 22.4% were in the top quarter and 24.4% were in the

bottom quarter. In the Midwest, 26.1% of the counties fell into the top quarter of TRI

pound per facility percentage change, and 22.8% were in the bottom quarter. In the West,

29.7% of the counties fell in the bottom 25%, and only 22.2% were in the top 25% of

TRI facility change. In the South, a slightly higher percentage of counties (25.9%) were

in the bottom 25% category than in the top 25% of percentage change for TRI facilities

(25.3%).

Table 4.25

Cross tabulations of Change in Dirtiness Index, 1988-1997 and Census Region

Census Region

Northeast

Midwest

West

South

Bottom

24.4

22.8

29.7

25.9

Middle

53.2

51.1

48.1

48.8

Top 25%

22.4

26.1

22.2

25.3

In Table 4.25, the cross tabulation of changes in the Dirtiness Index by census

region is presented. In the Northeast, 58.1% of the counties fell within the middle 50% of

change in the Dirtiness Index, while 27.6% were in the bottom quarter and only 14.3%

were in the top quarter. In the Midwest, 25.8% of the counties fell into the top quarter of

Dirtiness Index change, and 25.4% were in the bottom quarter. In the West, 59.3% of the

counties were in the middle 50%, 27.2% of the counties were in the bottom 25%, and

only 13.6% of the counties were in the top quarter of the Dirtiness Index change

distribution. In the South, a higher percentage of counties (28.9%) were in the top 25%

category than in the bottom quarter of counties for change in the Dirtiness Index (23.7%).

Cross-Tabulation of Manufacturing-Service Sector Changes and Toxic Changes

with Metropolitan Status, 1988-1997

Tables 4.26 through 4.32 each contain the cross-tabulation of one of the

manufacturing and service sector change characterization variables (either for employees,

establishments, or wages) or one of the toxic change characterization variables (change in

TRI facilities, change in TRI pounds, change in TRI pounds/facility, and change in

Dirtiness Index) with the metropolitan status variable. The categories for the

manufacturing and service sector change variables, and the toxic change variables, are

Census Region

Northeast

Midwest

West

South

Bottom

27.6

25.4

27.2

23.7

Middle

58.1

48.8

59.3

47.3

Top 25%

14.3

25.8

13.6

28.9

described above. As noted earlier, the metropolitan status variable includes three

categories for each county: metropolitan area, adjacent to metropolitan area, or not

adjacent to a metropolitan area. Table 4.3 describes the distribution of counties in each of

these regions. Using this approach allows comparisons across categories of the

manufacturing and service sector variable or toxic variable to be made based on the

percentage differences across metropolitan category types.

Table 4.26

Cross tabulations of Changes in Manufacturing (Mfg) and Service (Svc) Employment,

1988-1997 and Metro Proximity

In Table 4.26, the cross tabulation of changes in manufacturing and service sector

employment by metropolitan status is presented. In metropolitan areas, decline in

manufacturing and growth in service employment is the most common pattern, observed

in 51.5% of the counties. Stability is also relatively common (44.4%). In counties which

are adjacent to metropolitan areas, stability (45.7%) and a decline in manufacturing and

increase in service employment (41.5%) are the most common occurrences, but growth in

Metro Proximity

Metro

Adjacent to

Metro

Not Adjacent

to Metro

Stability

44.4

45.7

Growth both

2.8

8.4

11.1

Growth

mfg/decline svc

0.8

3.4

2.3

Decline

mfg/growth svc

51.5

41.5

33.1

Decline both

1.0

1.6

both sectors (8.4%) and growth in manufacturing and a decline in service (3.4%) are

more common here than in metropolitan areas. In non-adjacent counties, stability (52%)

is most frequent, with a decline in manufacturing and increase in service employment

being next at 33.1%, though this is substantially lower than in the other metropolitan

area-types. Growth in both manufacturing and service employment is more common here

than in the other areas (11.1%).

In Table 4.27, the cross tabulation of changes in manufacturing and service sector

wages by metropolitan status is presented. Across areas, stability is the most common

occurrence, ranging from 36.1% in non-adjacent areas to 37.6% in metropolitan counties.

In metropolitan areas, growth in wages across both sectors follows at 33.7%. Decline in

manufacturing and increase in service wages follows at 17.4%, which is lower than in the

other types. In counties which are adjacent to metropolitan areas, growth in both sectors

(26.9%) and a decline in manufacturing and increase in service wages (23.7%) are

relatively close. In non-adjacent counties, a decline in manufacturing and increase in

service wages (26%) is slightly more common than growth in both (25.5%).

Table 4.27

Cross tabulations of Changes in Manufacturing (Mfg) and Service (Svc) Wages, 1988-

1997 and Metro Proximity

In Table 4.28, the cross tabulation of changes in manufacturing and service sector

establishments by metropolitan status is presented. In metropolitan counties, a decline in

manufacturing and growth in service establishments is most common (56.8%), with

growth in wages across both sectors following at 41.4%. In counties which are adjacent

to metropolitan areas, growth in both sectors (52.4%) is most common, followed by a

decline in manufacturing and increase in service establishments (40.4%). Compared to

metropolitan counties, growth in manufacturing and a decline in service establishments

(3.7%), and a decline in both (2.2%) are more common. In non-adjacent counties, growth

in both manufacturing and service establishments is the highest (52.7%), followed by a

decline in manufacturing and increase in service establishments (35.7%). Growth in

manufacturing and decline in service (4.6%), stability (3.6%), and a decline in both

(3.4%) are each more frequent than in the other metropolitan types.

Table 4.28

Cross tabulations of Changes in Manufacturing (Mfg) and Service (Svc) Establishments,

1988-1997 and Metro Proximity

Metro Proximity

Metro

Adjacent to

Metro

Not Adjacent

to Metro

Stability

37.

37.3

36.1

Growth both

33.7

26.9

25.5

Growth

mfg/decline svc

6.5

6.9

6.2

Decline

mfg/growth svc

17.4

23.7

Decline both

4.8

5.2

6.3

In Table 4.29, the cross tabulation of changes in TRI facilities by metropolitan

status is presented. 28.9% of the metropolitan counties fell into the bottom 25% of

percentage change in TRI facilities, which is considerably higher than the other regions.

In adjacent to metro counties, 27.5% were in the top 25% of TRI facility change. Finally,

in non-metropolitan counties, 52.6% were in the middle 50%, and 25.2% fell in the top

quarter of percentage change for TRI facilities.

Table 4.29

Cross tabulations of Change in TRI Facilities, 1988-1997 and Metro Proximity

In Table 4.30, the cross tabulation of changes in TRI pounds by metropolitan

status is presented. 58.7% of the metropolitan counties were in the middle 50% of

percentage change in TRI pounds, which is considerably higher than the other regions. In

Metro Proximity

Metro

Adjacent to

Metro

Not Adjacent

to Metro

Stability

0.1

1.3

3.6

Growth both

41.4

52.4

52.7

Growth mfg/decline

svc

1.2

4.6

Decline mfg/growth

svc

56.8

40.4

35.7

Decline both

0.5

2.2

3.4

Metro Proximity

Metro

Adjacent to

Metro

Not Adjacent

to Metro

Bottom 25%

28.9

23.3

22.2

Middle 50%

48.6

49.2

52.6

Top 25%

22.5

27.5

25.2

adjacent to metro counties, 27.2% fell in the bottom 25%, and 27.1% were in the top 25%

of TRI pounds change. In non-metropolitan counties, 28.3% were in the bottom quarter,

and 27.2% fell in the top 25% of percentage change for TRI facilities.

Table 4.30

Cross tabulations of Change in TRI Pounds, 1988-1997 and Metro Proximity

In Table 4.31, the cross tabulation of changes in TRI pounds per facility by

metropolitan status is presented. As was the case when comparing TRI pounds per

facility by other variables, the distribution is somewhat similar to that for TRI pounds

and TRI facilities. Here, 54.8% of the metropolitan counties were in the middle 50% of

percentage change in TRI pounds per facility, which is still considerably higher than the

other regions. In adjacent to metro counties, 26.6% fell in the bottom 25%, and 26% were

in the top 25% of TRI pounds per facility change. In non-metropolitan counties, 27.9%

were in the top 25%, and 25.6% fell in the bottom quarter of percentage change for TRI

facilities.

Table 4.31

Cross tabulations of Change in TRI Pounds per Facility, 1988-1997 and Metro Proximity

Metro Proximity

Metro

Adjace

nt to

Metro

Not Adjacent

to Metro

Bottom 25%

20.2

27.2

28.3

Middle 50%

58.7

45.6

44.5

Top 25%

21.2

27.1

27.2

Finally, in Table 4.32, the cross tabulation of changes in the Dirtiness Index by

metropolitan status is presented. For metropolitan counties, 55.3% fell in the middle 50%

of Dirtiness Index change, while 25.7% were in the bottom quarter. In the adjacent to

metro category, 29.3% of the counties were in the top 25%, and only 45.1% were in the

middle 50%. For non-metropolitan counties, 23.9% fell in the bottom quarter of the

distribution of change for the Dirtiness Index. In the final section of this chapter, the

analysis of spatial clustering of changes in the manufacturing and service sectors, and of

toxic releases, will be discussed.

Table 4.32

Cross tabulations of Change in Dirtiness Index, 1988-1997 and Metro Proximity

Spatial Clustering of Changes in the Manufacturing and Service Sectors, and of

Toxic Releases

In this section, the results of univariate and bivariate Local Indicators of Spatial

Metro Proximity

Metro

Adjacent

Metro

Not Adjacent

to Metro

Bottom 25%

23.3

26.6

25.6

Middle 50%

54.8

47.5

46.5

Top 25%

21.9

27.9

Metro Proximity

Metro

Adjacent to

Metro

Not Adjacent

to Metro

Bottom 25%

25.7

25.6

23.9

Middle 50%

55.3

45.1

50.4

Top 25%

29.3

25.7

Analysis (LISA) will be presented. The general logic underlying LISA is laid out in

Chapter Three. It is important to note that there is a substantive interpretation difference

for univariate and bivariate LISA results. In examining the figures, it is important to note

that different colors represent the possible values for the analysis: White indicates a

county in which a non-significant relationship exists at the p < 0.05 level; the darker red

counties are those which have a high value and also border neighbors with high values on

either the same variable OR in bivariate LISA, on another variable of interest (for

example, having a high value on the change in manufacturing establishments and having

neighbors with high values on the change in TRI facilities variable); the deeper blue

counties are those which have a low value and also border neighbors with low values on

the same variable OR in bivariate LISA, on the other variable of interest; the lighter blue

counties include those which have a low value yet border counties with high values on

the same variable OR again in bivariate LISA, on the other variable of interest; and the

lighter red counties are those which have a high value but border counties with low

values on the same variable OR for the other variable of interest in bivariate LISA.

There are twenty-eight univariate and bivariate LISA figures included in this

chapter. Only eight of these are discussed in this chapter; however, these and all of the

figures can be found in the appendix, beginning on page 220. Upon careful examination

of the options, the choice was made to focus on those figures which utilize manufacturing

and/or service establishment data. This decision was made because the figures for both

manufacturing and service (for employment, wages, and establishments) show fairly

similar patterns (for example, a pattern of metropolitan areas falling into the high-high

category most often). Since there is such similarity in the patterns for all three industry

measures, and given that the toxic data includes a measure of the number of TRI

facilities, focusing exclusively on the establishment data makes the most sense. Again,

however, the additional figures are included in the appendix.

Figure A.5 displays a bivariate LISA of the changes in number of county-level

manufacturing establishments between 1988 and 1997. There are several large clusters of

high-high, including around the northeastern megapolis (Boston-New York-

PhiladelphiaWashington), Seattle, Portland Oregon, southern California, San Francisco,

Dallas and Houston Texas, south Florida, Detroit, Chicago, Minneapolis, Charlotte, and

in a few individual counties around the country. The largest cluster of low-low counties is

found in the Great Plains from the Canadian to Mexican borders, where many counties

either have zero facilities or did not report data in one or both years. Other low-low

clusters are found in large swaths of the Mountain West (Nevada, Utah, Wyoming,

Colorado), in Appalachia (Kentucky and West Virginia particularly), and in parts of the

Black Belt (Georgia, and in the Mississippi Delta region). Many of these areas are among

the poorest in the country and have been long-standing targets of broad-based

development efforts, apparently to little success as of 1997 (if manufacturing facilities

equate with economic success in any meaningful way). A handful of low-high counties

exist, mostly in areas near metropolitan areas (California, Arizona, Washington, Texas,

Florida, etc.), bordering high-high counties. These are likely rural counties just outside of

metropolitan areas with little to no manufacturing as of 1997. Finally, a handful of high-

low counties exist as well, mostly in the Mountain West or the Great Plains. These

represent counties which are „oases‟ of manufacturing, surrounded generally by little to

no population or

manufacturing development.

Figure A.6 displays a bivariate LISA of the changes in number of county-level

service establishments between 1988 and 1997. This figure is rather similar to the

previous one for manufacturing establishments, with a few notable differences. First,

there is a larger high-high cluster in south Florida, particularly around Orlando (likely

due to the growth of the tourist industry in that region). Also, the large low-low cluster in

the Midwest and Great Plains is still visible, but is a bit more sporadic. This is likely the

result of the general pattern of greater overall growth in services than manufacturing

around the country during the late 1980s and 1990s. Finally, there does appear to be more

low-low counties in the South, most notably in Georgia, Alabama, Mississippi, and

Arkansas.

In Figure A.7, the bivariate LISA of the changes in number of county-level TRI

facilities between 1988 and 1997 is shown. In this figure, high-high counties represent

those counties which had a significantly high number of TRI facilities in 1988 and are

bordered by counties which had a significantly high number of TRI facilities in 1997.

The patterns observed are again similar to those seen in Figures A.5 and A.6. However,

some interesting differences do emerge. For example, in the Northeast, where there are

many high-high counties in metropolitan areas, there are some low-high counties that

appear in neighboring counties just outside of the central metropolitan area. Also, there

are high-high clusters which appear in North Carolina‟s furniture manufacturing belt, and

also some high-high counties near Atlanta.

Figure A.13 displays the bivariate LISA of the percent change in manufacturing

establishments versus the percent change in service establishments. Thus in this figure,

for example, high-high counties are those counties which had a significantly high

positive change in the percentage of manufacturing establishments between 1988 and

1997, and had neighbors which had a significantly high positive change in the percentage

of service establishments between 1988 and 1997. This figure is noticeably different than

the previous figures, which should not be surprising since two distinctive variables are

involved, as opposed to the prior figures which each focused on the (typically correlated)

relationship of one variable to itself over time. Another difference is that the largely

metropolitan effect observed in the previous figures is also largely absent here. Large

clusters of high-high counties are observed in Arkansas, Iowa, and in east Mississippi.

Clusters of low-low counties are found in much of the Southeast, pockets of the West

Coast states, and in Upper Michigan. Low-high clusters (i.e. larger declines in

manufacturing and greater growth in service) are observed in the Mountain West, the

Upper Midwest, and in the Appalachians. High-low clusters (i.e. higher percentage of

growth in manufacturing and larger declines in service) are found mostly in the Mountain

West, with some pockets in New York, Virginia, Georgia, Florida, Alabama, and

Mississippi.

In Figure A.14, a univariate LISA is presented for the change in TRI facilities

between 1988 and 1997. High-high counties in this figure indicate that a county and its

neighbors each had significantly high levels of positive change in TRI facilities between

1988 and 1997. Notable high-high clusters are observed in western Oregon, southern

Arizona along the Mexican border, Minnesota, Indiana, Ohio, Tennessee, Alabama,

North Carolina and in South Carolina. Low-low clusters are found in Vermont, New

York, Virginia, Massachusetts, Florida, Texas, and in the Mountain West region. Highlow

clusters are found in Maine, New Hampshire, and Washington, with some other

individual counties spread out around the country. Low-high clusters are observed

throughout the Northwest, in Minnesota and much of the Midwest, in Alabama, and in

the Carolinas, though in virtually each case these counties border high-high counties.

Figure A.20 displays the univariate LISA of the manufacturing-service

establishments change characterization variable. High-high counties in this figure

indicate that a county and its neighbors each had significantly high levels of positive

change in both manufacturing and service facilities between 1988 and 1997. The first

thing observable in this figure is the relatively high presence of colors, indicating

significant findings virtually across the country. Another point of interest is that an initial

glance would lead the viewer to conclude that counties and their neighbors grew in both

services and manufacturing establishments between 1988 and 1997, or counties and their

neighbors experienced declines in manufacturing establishments and growth in service

establishments, and that these findings were largely independent of space. However, a

more careful examination indicates that while those counties (and their neighbors) in

which both manufacturing and service establishments grew are many, they do not

generally appear in the coastal (and frequently urban) areas of the East. They are also less

visible in Michigan and in coastal Oregon and Washington. In these areas, there is a

much greater presence of manufacturing decline and service growth. There are very few

counties which experienced a decline in both manufacturing and service along with their

neighbors, though a few small clusters and isolated counties can be observed in the

Mountain West. There are also few counties in which manufacturing grew and services

declined, though these are most frequent in the Mountain West and are generally not

clustered around other similar counties.

Figure A.21 displays the bivariate LISA of the percent change in TRI facilities

versus the percent change in manufacturing establishments. Thus in this figure, for

example, high-high counties are those counties which had a significantly high positive

change in the percentage of TRI facilities between 1988 and 1997, and had neighbors

which had a significantly high positive change in the percentage of manufacturing

establishments during this same time period. Counties which were high-high are mostly

located in rural and relatively isolated regions, except for some clusters found in the

Southeast, and particularly in Alabama. Low-low clusters are observed in parts of New

York and Massachusetts, along with Virginia, Florida, Colorado and Washington.

Lowhigh clusters are seen in southwestern Oregon, in Alabama, the Carolinas, and

around Orlando. High-low clusters are observable particularly in Colorado, Washington,

and northern New England (northern New York, Vermont, New Hampshire, and Maine).

In Figure A.22, a bivariate LISA of the percent change in TRI facilities versus the

percent change in service establishments is displayed. In this figure, for example,

highhigh counties are those counties which had a significantly high positive change in the

percentage of TRI facilities between 1988 and 1997, and had neighbors which had a

significantly high positive change in the percentage of service establishments during this

same time period. Many similarities to the previous figure are observed, though some

interesting differences can also be seen. For example, the high-high clusters appear in

Oregon, along the Mexican border in Arizona, in Minnesota, North Carolina, Alabama

and Florida. Low-low clusters are seen in Colorado, Texas, Louisiana, Washington and

Montana. Low-high clusters appear in Oregon, Minnesota, Idaho, Indiana, Ohio,

Alabama and Kentucky. There are also high-low clusters in New England, Virginia,

Florida, around Dallas, Texas, and in Washington.

Conclusion

The numerous analyses offered above were designed to address four primary

issues, phrased as questions, which were stated at the beginning of this chapter. Taking

each of these questions in turn then, they include:

1. What is the composition and character of the manufacturing and service sectors

and of manufacturing’s toxic releases in 1988 and 1997 across U.S. counties?

As was discussed earlier in this dissertation, the “deindustrialization” of America does

appear to be going on during the time period under consideration, as the number of

manufacturing employees declined at the county-level, while the number of service

employees increased. However, this interpretation must be tempered somewhat by the

fact that manufacturing establishments grew slightly during the late 1980s and early to

mid 1990s, though this growth was far outpaced by the growth in service establishments.

Additionally, the wages paid to manufacturing employees continued to outpace the

average overall wage and the average service wage, though service wages gained traction

relative to both manufacturing wages and especially overall wages at the county-level. As

relates to the composition of toxic output, increases in the number of facilities reporting

to the TRI, the number of pounds of toxic releases reported by these facilities, and the

number of TRI pounds per facility each indicate that environmental toxicity worsened

somewhat during this time period.

2. Did changes across manufacturing and/or service sectors affect levels of toxic

releases across U.S. counties between 1988 and 1997?

There are some linkages of note between industrial sector changes and changes in the

measures of toxicity. Those counties in which declines in manufacturing employment or

establishments occurred were more likely to have declines in TRI facilities, pounds of

TRI releases, TRI pounds per facility, and the Dirtiness Index. Also, TRI facilities,

pounds of TRI releases, TRI pounds per facility, and the Dirtiness Index all increased in

those counties where manufacturing employment or establishments grew and service

employment or establishments declined. Shifts in wages across sectors had little or no

relationship to changes in toxic measures.

3. Was there a spatial pattern involved in changes which took place for industry

and/or toxic releases between 1988 and 1997? Specifically, are there regional or

metropolitan status differences in industry and/or toxic release changes between

1988 and 1997?

For certain measures it does appear that there was a spatial composition in manufacturing

and service sector changes between 1988 and 1997. For example, in the Northeast and to

a lesser extent the South, county-level declines in manufacturing employment were much

more commonly observed than in the other Census regions. This carried over to declines

in manufacturing establishments in the Northeast as well, which were somewhat more

common than in the other regions of the country. Of further interest is that the Northeast

also experienced more counties with growth in wages across manufacturing and service

sectors and fewer counties with decline in manufacturing wages and growth in service

wages, particularly as compared with the Midwest and the West. This indicates that even

as more manufacturing facilities were leaving the Northeast, those that were staying or

opening their doors during this time period were paying higher wages at the end of the

time period than in the beginning, as compared to the other regions of the country.

Regarding toxic pollution and polluting facilities, Northeast counties were much

more likely to experience declines in the number of TRI facilities and a reduction in the

Dirtiness Index, while counties in the West appear slightly more likely to have

experienced declines in number of TRI pounds during the time period.

Relating metropolitan status and industrial sector changes, not adjacent to metro

counties were more likely to experience employment growth across sectors and less

likely to face the decline of manufacturing employment than metro or adjacent to metro

counties. Metro counties were more likely to experience wage growth across sectors and

less likely to experience declines in manufacturing wages than adjacent or not adjacent

counties. Metro counties were also more likely to experience declines in manufacturing

establishments and less likely to experience growth in establishments across sectors than

adjacent to metro or not adjacent to metro counties.

Finally, there are relatively minor differences regarding metropolitan status and

toxicity, as metro areas were more likely to have declines in TRI facilities but less likely

to have declines in TRI pounds than adjacent to metro or non adjacent to metro counties.

Metro areas were also generally less likely to have gains in any of the toxicity measures

as compared to adjacent to metro or not adjacent to metro counties.

4. Are there significant spatial concentrations in the composition of manufacturing

and service sectors, or of changes across these sectors between 1988 and 1997?

Also, are there significant spatial concentrations of TRI facilities, releases, and

“dirty” facilities which overlap with significant clusters of manufacturing or

service change?

In examining the figures displaying the univariate and bivariate LISA analyses, it appears

that there are significant spatial concentrations within both the manufacturing and service

sectors. There are relatively clear clusters of both high and low numbers of

manufacturing establishments and positive and negative change within manufacturing

establishments across the period 1988-1997. As noted above, it appears that

manufacturing establishments are more likely to be found in metropolitan and adjacent to

metropolitan areas, and are also less likely to be found in clusters in the poorest or most

remote parts of the country (Appalachia, Mississippi Delta, parts of the Great Plains,

etc.).

When examining the link between toxicity and industrial change, several clusters

of interest emerged, given that they represented potentially problematic situations from

either the economic or environmental standpoint. First, locales with high levels of growth

in the percentage of TRI facilities and neighbors with a high percentage of growth in

manufacturing establishments were found in rural and relatively isolated regions. These

areas represent potential sub-regions of environmental risk, with both existing and

potential sources of pollution spread across multiple counties. Locales with high levels of

growth in the percentage of TRI facilities and neighbors with declines in service

establishments were found in areas of New England, Virginia, Florida, Texas, and

Washington. These are negative from both the environmental and economic standpoints,

as the general decline of manufacturing is not being countered in these locales with

growth in the service sector, and yet the percentage of TRI facilities is also increasing in

these areas, indicating elevated environmental risk.

In summarizing the findings of this chapter, it appears there are at least three key

points to consider: 1) there appears to be something to the deindustrialization argument,

as counties are generally moving away from manufacturing and are certainly moving

towards the service sector; 2) space matters, either as conceptualized by Census region or

metropolitan status, or as observed empirically through the presence of significant

clusters of similar characteristics; and 3) it appears that there is a category of places

which experienced both potentially negative economic and environmental conditions

simultaneously. The first issue will be explored in the next two chapters through

modeling of economic change in regression analyses. Spatial considerations will be

accounted for in the modeling in both subsequent analysis chapters, including through the

use of innovative spatial regression techniques in Chapter Six. This final issue will be

explored in Chapter Six through the development and testing of the Domestic Pollution

Havens concept, adapted from the efforts of ecological economists to examine those

areas around the world which take on polluting industries as a source of economic

development.

CHAPTER V

IS POLLUTION THE COST OF DOING BUSINESS IN AMERICAN

COMMUNITIES?

Introduction: Linking Pollution and Economic Change In this chapter,

attention is focused on the pivotal issue of whether or not pollution is the “cost of doing

business” for U.S. communities; in other words, is pollution a necessary trade-off for jobs

or higher wages in the local area? This is the core principle implied in the rhetorical “jobs

versus the environment” duality debated in the public arena and even in the academic

realm (Diamond 2005; Kazis and Grossman 1983).

The central premise of this divide is that those communities which seek economic growth

(either through an increase in facilities, jobs, or wages) must be willing to sacrifice

environmental quality to do so. The other side in this debate argues that the natural

environment can, and indeed has, become so damaged and degraded that human activity,

including that which is economic in nature, can and should no longer be carried out as

usual.

Those involved in this debate have largely relied on rhetoric or anecdotal case

studies to support their position. In this dissertation, a central goal is to advance this

debate through the use of more detailed national-level data. In the previous chapter,

univariate descriptive evidence was presented which points to both an increase in

manufacturing establishments and a concomitant increase in toxic facilities and pollution

levels between 1988 and 1997. Bivariate relationships were also observed between

changes in both manufacturing establishments and employees and changes in several

measures of toxicity (number of TRI facilities, pounds of TRI releases, pounds of TRI

releases per facility, and the Dirtiness Index) during the same time period.

To further accomplish the goal of advancing the “jobs versus the environment”

debate, there are two primary questions which are addressed here through the use of

cross-sectional comparisons and regression-based analyses: 1) is pollution part of the cost

of doing business associated with job or wage growth in U.S. communities; and 2) has

the “great slide” or decline of manufacturing affected local pollution levels?

As related to the theoretical perspectives described earlier in this dissertation, the

key hypotheses for this chapter include the following:

Hypothesis 5.1. Drawing from the Ecological Modernization and Environmental

Kuznets Curve perspectives (EM/EKC), counties which are more ―economically

advanced‖ or which experience economic growth over time will be expected to

experience growth in pollution output up to a point (the form of the relationship

will be curvilinear, first positive and then negative as the ecological state emerges

or a shift from manufacturing to services occurs), net of other factors. The

variables used to assess this hypothesis are median family income and median

family income squared.

Hypothesis 5.2. Drawing from the Environmental Conflict perspective (EC),

growth in pollution output will be expected in counties with greater economic

growth in manufacturing, net of other factors. The variables used to assess this

hypothesis are median family income and median family income squared.

Hypothesis 5.3. Drawing from the Environmental Justice perspective (EJ),

counties with higher percentages of social minorities will be expected to

experience growth in pollution output regardless of level of economic

―advancement,‖ net of other factors. The variables used to assess this hypothesis

include county-level log percent in poverty, log percent elderly, log percent

African-American, the Black-White median income ratio, and the Black-White

dissimilarity index.

Hypothesis 5.4. Drawing from theories of industrial concentration, counties with

positive growth in manufacturing activity or relative growth in manufacturing

versus services will be expected to experience growth in pollution output, net of

other factors. The variables used to assess this hypothesis include change in

manufacturing employees, change in manufacturing wages, change in

manufacturing establishments, service-manufacturing establishments

manufacturing growth dummy, service-manufacturing establishments service

growth dummy, service-manufacturing establishments decline dummy,

servicemanufacturing wages manufacturing growth dummy, service-

manufacturing wages service growth dummy, service-manufacturing wages

decline dummy, service-manufacturing employees manufacturing growth dummy,

servicemanufacturing employees service growth dummy, and service-

manufacturing employees decline dummy.

Hypothesis 5.5. Drawing from demographic theories, growth in pollution output

will be expected in counties as the population density increases. The variable used

to assess this hypothesis is the log population density.

Hypothesis 5.6. Drawing from quality of life theories, declines in pollution

output will be expected in counties as the level of in-migration increases. The variable

used to assess this hypothesis is the log percentage of county in-migration. To test the

hypotheses, regression modeling is utilized, including zero-inflated negative binomial

regression and ordinary least squares regression. The analyses in the next section are

performed to address the question “is pollution part of the cost of doing business?” In the

subsequent section, analyses are performed to address the question “has the „great slide‟

of manufacturing affected local pollution levels?” The chapter is concluded with a

summary of findings and results of the hypothesis testing.

Is Pollution Part of the Cost of Doing Business?

In Table 5.1, five environmental toxicity variables for 1988 are regressed on

economic standing, environmental justice, and other factors listed above. In panel one,

zero-inflated negative binomial regression is used to regress the number of TRI facilities

in the county during 1988 on the variables listed above. In this panel, a positive

coefficient indicates a higher number of TRI facilities. All of the variables are highly

statistically significant and many conform to expectations in the literature and the

hypotheses stated above. For example, median family income and median family income

squared are significant yet in the opposite direction, pointing to a curvilinear relationship

(i.e. increased median family income is associated with more TRI facilities up to a

certain threshold point, at which time higher median family income is associated with

fewer TRI facilities). This finding conforms to the EM/EKC hypothesis. However,

several of the environmental justice variables, particularly those linked to racial

inequality, point towards worsened environmental conditions (as measured by number of

facilities) as the percent black or the log black-white median income ratio or the

blackwhite dissimilarity index grows. To illustrate, the coefficient for the black-white

dissimilarity index, 0.832, indicates that a one unit increase in the dissimilarity index

(reflecting greater racial imbalance) increases the number of TRI facilities by 0.832.

The log percent poverty and the log percent elderly are highly significant but

negative in sign, as opposed to the environmental justice hypothesis expectations. The

log population density is highly significant and positive, indicating that as population

density grows, more TRI facilities are likely to be found in a county. Finally, the log in-

migration measure is highly significant and negative, indicating that those counties which

had more in-migration were likely to have fewer TRI facilities.

In Table 5.1, panels two through five, ordinary least squares regression is utilized

to regress the log number of TRI pounds, the log number of TRI pounds per facility, the

log median risk, and the dirtiness index in the county during 1988 (respectively) on the

variables listed above. OLS regression was determined to be appropriate, following

initial analyses utilizing Heckman regression, which did not produce significant selection

effects in the independent variables for counties that had a TRI facility versus those that

did not.

Table 5.1, panel two, presents the OLS regression of the log number of TRI

pounds on the variables related to economic standing, environmental justice, and other

factors noted above. In this panel, a positive coefficient indicates a higher number of log

pounds of TRI releases, reflecting worsened environmental conditions in the county. The

findings generally correspond to the expectations in the literature and the hypotheses

stated above. Median family income, log percent black, the black-white dissimilarity

index, and log population density are significant and positive, while the median family

income squared, log percent poverty, log percent elderly, and log in-migration are

significant and negative. The R2 for this model is .34, indicating that 34% of the variation

in log TRI pounds is explained by the independent variables in the model.

Of significance is the finding that median family income and median family

income squared again are highly significant and pointed in the opposite direction,

indicating a curvilinear relationship; i.e., increased median family income is associated

with more log TRI pounds up to a certain threshold point, at which time higher median

family income is associated with reduced log TRI pounds. As before, this finding is

supportive of the EM/EKC hypothesis. Certain environmental justice variables,

particularly those linked to racial inequality, are indicative of worsened environmental

conditions as the log percent black or the black-white dissimilarity index grows. The log

percent poverty and the log percent elderly are again significant but negative, as opposed

to the environmental justice hypothesis expectations. Also, the log population density is

highly significant and positive, indicating that as the population becomes denser,

countylevel log TRI pounds should increase. Finally, the log in-migration measure is

highly significant and negative, which means that those counties which had more in-

migration were likely to have fewer log TRI pounds.

Table 5.1

Pollution Measures Regressed on Economic Standing, Environmental Justice, and Other

Factors, 1988

TRI

Facilities

Log TRI

Pounds

Log TRI

Pounds per

Facility

Log Median

Risk

Dirtiness

Index

Median Family

Income

0.000***

0.000*

0.000***

0.000

Median Family

Income Squared

-0.000***

0.000*

-0.000***

Log Percent Poverty

-0.399***

-0.522*

-0.420*

-0.872***

-0.084***

Log Percent Elderly

-0.518***

-1.764***

-1.352***

-1.385***

-0.058***

Log Percent Black

0.088***

0.131*

0.073

0.236***

0.008**

Log Black-White

Median Income

Ratio

0.360***

-0.063

-0.100

-0.112

0.017

Black-White

Dissimilarity Index

0.832***

1.407***

0.693*

-0.275

0.053**

Log Population

Density

0.703***

0.796***

0.341***

0.380***

0.008**

-0.610***

-1.119***

-0.614***

-0.026

-0.018**

Constant

0.297

15.089***

14.974***

32.596***

0.536***

INFLATE D_Man88

-6.348***

---

INFLATE

Constant

1.797

---

Log Alpha

-0.568***

---

Alpha

0.566

---

Vuong test z-score

2.09*

---

Log Likelihood

-6506.897

---

0.337

0.117

0.073

Note: Model one (TRI facilities) was calculated using zero-inflated negative binomial regression;

models two through five use Ordinary Least Squares regression.

* p < 0.05; ** p < .01; *** p< .001 (two-tailed tests).

In Table 5.1, panel three, results are displayed for the OLS regression of the log

number of TRI pounds per facility on the variables related to economic standing,

environmental justice, and other factors noted above. In this panel, a positive coefficient

indicates a higher number of log pounds of TRI releases per facility. The findings

generally correspond to the expectations in the literature and the hypotheses stated above.

Median family income, the black-white dissimilarity index, and log population density

are significant and positive, while the median family income squared, log percent

poverty, log percent elderly, and log in-migration are significant and negative. R2 for this

model equals .12, which is much lower than that for log TRI pounds in the previous

model.

Of significance is the finding that median family income and median family

income squared again are both significant and pointed in the opposite direction,

indicating a curvilinear relationship; i.e., increased median family income is associated

with more log TRI pounds per facility up to a certain threshold point, at which time

higher median family income is associated with reduced log TRI pounds per facility. This

finding again conforms to the EM/EKC hypothesis. Here however only one measure of

environmental justice, the black-white dissimilarity index, points towards worsened

environmental conditions along racial lines. The log percent poverty and the log percent

elderly are again significant but negative, as opposed to the environmental justice

hypothesis expectations. Also, the log population density is highly significant and

positive, and the log in-migration measure is highly significant and negative.

In Table 5.1, panel four, results are displayed for the OLS regression of the log

median risk score on the variables related to economic standing, environmental justice,

and other factors noted above. In this panel, a positive coefficient indicates a higher log

median risk score. Median family income, median family income squared, log percent

black, and log population density are significant and positive, while the log percent

poverty and log percent elderly are significant and negative. The R2 for this model is .07,

again much lower than for the previous dependent variables.

It should be noted that for the log median toxic risk score, both median family

income and median family income squared point in the same direction, as opposed to the

previous models. Thus, increased median family income here is associated with an

increased log median risk score, or worsened environmental quality in the county. This

finding runs counter to the EM/EKC hypothesis, and points to worsened median risk as

income grows. This is significant because median risk is the most refined toxicity

measure, since it provides a modeled estimate of actual exposure and risk, not just

potential generalized harm. Only one measure of environmental justice, log percent

black, points towards worsened environmental conditions along racial lines. The log

percent poverty and the log percent elderly are once again significant but negative.

Finally, the log population density is highly significant and positive, indicating that as

population becomes denser, the log median risk score is likely to be higher.

In Table 5.1, panel five, results are displayed for the OLS regression of the

dirtiness index score on the variables related to economic standing, environmental

justice, and other factors noted above. In this panel, a positive coefficient indicates a

higher dirtiness index score. Log percent black, black-white dissimilarity index and log

population density are significant and positive, while the median family income squared,

log percent poverty, log percent elderly, and log in-migration are significant and

negative. The R2 for this model equals .07, which is the same explained variance for log

median risk.

Two of the environmental justice measures, log percent black and the black-white

dissimilarity index, point towards worsened environmental conditions along racial lines,

such that an increase in either is linked to an increased dirtiness index score. The log

percent poverty and the log percent elderly are again significant but negative, the log

population density is significant and positive again, and the log in-migration measure is

significant and negative.

The results in Table 5.1 provide relatively strong and consistent support for the

EM/EKC hypothesis (H5.1); limited support for the EC hypothesis (H5.2); mixed

findings regarding the EJ hypothesis (H5.3), with relatively strong support for the race

component but opposite to expected findings for poverty and the elderly; and strong

support for the hypotheses linked to the control variables, population density (H5.5) and

in-migration (H5.6).

In Table 5.2, five environmental toxicity variables for 1997 are regressed on

economic standing, environmental justice, and other factors listed above. As was the case

for the models displayed in Table 5.1, zero negative binomial regression and ordinary

least squares regression were used.

In panel one, zero negative binomial regression is used to regress the number of

TRI facilities in the county during 1997 on the variables listed above, as was done with

the same variable from 1988. In this panel, a positive coefficient indicates a higher

number of TRI facilities. As with the 1988 data, all of the variables are statistically

significant and generally conform to expectations in the literature and the hypotheses

stated above. Median family income and median family income squared are significant

yet in the opposite direction, pointing to a curvilinear relationship as described above.

This inverted u-curve finding conforms to the EM/EKC hypotheses. Several of the

environmental justice variables, particularly those linked to racial inequality, point

towards worsened environmental conditions (as measured by number of facilities) as the

percent black, the log black-white median income ratio or the black-white dissimilarity

index grows. The log percent poverty and the log percent elderly are highly significant

but negative, as opposed to the environmental justice hypothesis expectations. Log

population density is highly significant and positive, meaning that as population density

grows, a higher number of TRI facilities are likely to be found in a county. Finally, the

had more in-migration were likely to have fewer TRI facilities.

In Table 5.2, panels two through five, ordinary least squares regression is utilized

to regress the log number of TRI pounds, the log number of TRI pounds per facility, the

log median risk, and the dirtiness index in the county during 1997 (respectively) on the

variables listed above. As was described above for the 1988 environmental toxicity

variables, OLS regression was determined to be appropriate following modeling using

Heckman regression did not yield significant selection effects across counties with TRI

facilities.

Table 5.2, panel two, presents the OLS regression of the log number of TRI

pounds in 1997 on the variables related to economic standing, environmental justice, and

other factors noted above. In this panel, a positive coefficient indicates a higher number

of log pounds of TRI releases. Median family income, log percent black, the black-white

dissimilarity index, and log population density are significant and positive, while the

median family income squared, log percent poverty, log percent elderly, and log

inmigration are significant and negative. R2 for this model equals .31, a slight decrease

from the 1988 model.

Of significance is the finding that median family income and median family

income squared again are highly significant and pointed in the opposite direction,

indicating a curvilinear relationship. Once again, this means that increased median family

income is associated with more log TRI pounds up to a certain threshold point, at which

time higher median family income is associated with reduced log TRI pounds. This

finding supports the EM/EKC hypothesis and runs counter to the EC hypothesis.

Certain environmental justice variables, particularly those linked to racial

inequality, point towards worsened environmental conditions as the log percent black or

the black-white dissimilarity index grows. The log percent poverty and the log percent

elderly are again highly significant yet negative, as opposed to the environmental justice

hypothesis expectations. Also, the log population density is highly significant and

positive, indicating that as population becomes denser, more log TRI pounds are likely to

be found in a county. Finally, the log in-migration measure is highly significant and

negative, indicating that those counties which had more in-migration were likely to have

fewer log TRI pounds.

Table 5.2

Pollution Measures Regressed on Economic Standing, Environmental Justice, and Other

Factors, 1997

TRI

Facilities

Log TRI

Pounds

Log TRI

Pounds per

Facility

Log Median

Risk

Dirtiness

Index

Median Family

Income

0.000***

0.000**

0.000*

0.000

Median Family

Income Squared

-0.000***

Log Percent Poverty

-0.445***

-0.844***

-0.559**

-5.841***

-0.070***

Log Percent Elderly

-0.563***

-1.841***

-1.470***

-4.736***

-0.410***

Log Percent Black

0.068**

0.176**

0.109*

1.003***

0.010***

Log Black-White

Median Income

Ratio

0.253**

-0.251

-0.252

3.143***

0.009

Black-White

Dissimilarity Index

0.746***

1.582***

0.820**

4.877***

0.001

Log Population

Density

0.722***

0.740***

0.271***

3.404***

0.007**

-0.572***

-1.121***

-0.600***

-1.862***

-0.013*

Constant

0.816

16.755***

15.934***

33.950***

0.459***

INFLATE D_Man88

-4.984***

---

INFLATE

Constant

0.697

---

Log Alpha

-0.553***

---

Alpha

0.575

---

Vuong test z-score

1.69*

---

Log Likelihood

-6859.877

---

0.305

0.103

0.349

0.051

Note: Model one (TRI facilities) was calculated using zero-inflated negative binomial regression;

models two through five use Ordinary Least Squares regression. * p < 0.05; ** p < .01; *** p<

.001 (two-tailed tests).

In Table 5.2, panel three, results are displayed for the OLS regression of the log

number of TRI pounds per facility in 1997 on the variables related to economic standing,

environmental justice, and other factors noted above. In this panel, a positive coefficient

indicates a higher number of log pounds of TRI releases per facility. The findings

generally correspond to the expectations in the literature and the hypotheses stated above.

Median family income, log percent black, the black-white dissimilarity index, and log

population density are significant and positive, while the median family income squared,

log percent poverty, log percent elderly, and log in-migration are significant and negative.

R2 for this model equals .10.

A notable finding is that median family income and median family income

squared again are both significant and pointed in the opposite direction, again

conforming to the EM/EKC hypothesis. For log pounds per facility, both the log percent

black and the black-white dissimilarity index conform to the expectation that growth in

either variable, indicating greater racial inequality is linked with worsened environmental

conditions. The log percent poverty and the log percent elderly are again significant but

negative, the log population density is highly significant and positive, and the log

inmigration measure is highly significant and negative.

In Table 5.2, panel four, results are displayed for the OLS regression of the log

median risk score in 1997 on the variables related to economic standing, environmental

justice, and other factors noted above. Median family income, log percent black, log

black-white median income ratio, black white dissimilarity index, and log population

density are significant and positive, while the median family income squared, log percent

poverty, log percent elderly, and log in-migration rate are significant and negative. R2 for

this model equals .35, which should be noted represents a remarkably improved model fit

as opposed to the log median risk score in 1988, which was .07.

For the log median risk score, both median family income and median family

income squared point in opposite directions, again representing a curvilinear relationship

between median family income and environmental quality, which conforms to the

EM/EKC hypothesis. For this measure, all three of the race-based measure of

environmental justices, log percent black, log black-white median income ratio, and the

black white dissimilarity index, point towards worsened environmental conditions for

minorities. The log percent poverty and the log percent elderly are again significant but

negative. Also, the log population density is highly significant and positive, and the log

in-migration rate is highly significant and negative.

In Table 5.2, panel five, results are displayed for the OLS regression of the

dirtiness index score on the variables related to economic standing, environmental

justice, and other factors noted above. In this panel, a positive coefficient indicates a

higher dirtiness index score. Log percent black and log population density are significant

and positive, while the median family income squared, log percent poverty, log percent

elderly, and log in-migration are significant and negative. R2 for this model equals .05,

somewhat lower than the R2 of .07 for the dirtiness index in 1988.

One of the environmental justice measures, log percent black, points towards

worsened environmental conditions for African-Americans, such that a one unit increase

in the log percent black is linked to a 0.01 increase in the dirtiness index score. The log

percent poverty and the log percent elderly are again significant but negative, the log

population density is significant and positive, and the log in-migration measure is

significant and negative.

The results in Table 5.2 once again provide relatively strong support for the

EM/EKC hypothesis (H5.1), and thus relatively opposite to expected results for the EC

hypothesis (H5.2). There are again mixed findings regarding the EJ hypothesis (H5.3),

with relatively strong support for the race component and opposite to expected findings

for poverty and the elderly. As in 1988, there is strong support for the hypotheses linked

to the control variables, population density (H5.5) and in-migration (H5.6).

In comparing 1988 and 1997, it would appear that the economic boom of the

1990s led to somewhat mixed differences in environmental quality across space. As the

median income of counties improves, there does appear to be a salutatory positive effect

on environmental quality, once a certain threshold is reached (the inverted U-curve

posited by those in the Environmental Kuznets Curve and Ecological Modernization

camps). However, this result is not uniform across the dependent variables or over time.

As related to environmental inequality, particularly along racial lines, there is consistent

inequality for African-American populations, a result which is consistent with the

Environmental Justice perspectives.

In Table 5.3, change between 1988 and 1997 in five environmental toxicity

variables is regressed on economic standing, environmental justice, economic change,

and other factors of interest. Economic change variables are included here to address the

effects of changes across manufacturing and service establishments, wages, and

employees between 1988 and 1997. In panels one through five, ordinary least squares

regression is utilized to regress the change in TRI facilities, the log number of TRI

pounds, the log number of TRI pounds per facility, the log median risk, and the dirtiness

index in the county between 1988 and 1997 on the independent variables.

In panel one, results are displayed for the OLS regression of the change in TRI

facilities on the variables of interest. In this panel, a positive coefficient indicates a higher

number of TRI facilities. Several variables are statistically significant and most conform

to expectations in the literature and the hypotheses stated above. Median family income,

log population density, change in the number of manufacturing employees, and the

number of TRI facilities in 1988 are each statistically significant and positive, while

median family income squared, log percent poverty, log percent elderly, change in

number of manufacturing establishments, and change in manufacturing wages are each

statistically significant and negative. R2 is .27 for this model.

As was observed in the cross-sectional analyses, median family income and

median family income squared are significant but in the opposite direction, pointing to a

curvilinear relationship as described above, a finding which is consistent with the

EM/EKC hypothesis. No environmental justice findings of note are observed, except

again that log percent elderly and log percent poverty are both significant yet in the

opposite direction from that expected (i.e. as each increases, change in the TRI facilities

is negative). Log population density is significant and positive, indicating that as log

population density increases, there is an increase in the number of TRI facilities between

1988 and 1997. Also significant and positive is the number of TRI facilities in 1988,

which means that as the number of facilities in 1988 increases, the change in the number

of TRI facilities between 1988 and 1997 is greater. Finally, there are mixed results for the

change in manufacturing variables, as change in the number of manufacturing employees

is significant and positive, yet change in the number of manufacturing establishments and

change in manufacturing wages are both significant and negative. For employees, this

means that as the number of employees in manufacturing increases, the change in the

number of TRI facilities is higher. For establishments and wages, this means that as each

increases, the change in the number of TRI facilities is negative. This is an interesting

and somewhat unexpected finding that may indicate improved environmental efficiency

in the production process, a sign of ecological modernization of the production process.

In panel two, results are displayed for the OLS regression of the change in log

TRI pounds on the independent variables described above. In this panel, a positive

coefficient indicates a greater change in log TRI pounds between 1988 and 1997. Six

variables are statistically significant in this model, including log population density,

which is statistically significant and positive, while median family income squared, log

percent poverty, log percent elderly, log in-migration, and log TRI pounds in 1988 are

each statistically significant and negative. R2 is .04 for this model.

Median family income squared is highly significant and negative, indicating that

as median family income squared increases, log TRI pounds decrease, a finding that

contrasts the ones seen earlier, and is indicative of overall environmental improvement

for areas with higher median incomes. This could represent the exportation of highly

polluting facilities to other poorer locales, a topic explored in great depth in Chapter Six.

Log percent elderly and log percent poverty are both significant yet are again in the

opposite direction from that expected from the environmental justice perspective. Log

population density is significant and positive. Also significant and negative is the number

of log TRI pounds in 1988, which means that as the number of log TRI pounds in 1988

increases, the change in the number of log TRI pounds between 1988 and 1997 is

negative. In effect, counties with more log TRI pounds in 1988 saw a reduction in log

TRI pounds between 1988 and 1997, an effect that grows along with the number of log

TRI pounds in 1988. Finally, none of the change in manufacturing or service variables is

statistically significant.

Table 5.3

Change in Pollution Measures Regressed on Economic Change, Economic Standing,

Environmental Justice, and Other Factors, 1988-1997

Change in

TRI

Facilities

Change in

Log TRI

Pounds

Change in

Log TRI

Pounds per

Facility

Change in

Log

Median

Risk

Change in

Dirtiness

Index

Median Family

Income

0.000**

0.000

14.604

0.000

Median Family

Income Squared

-0.000***

0.000

-0.000***

-0.000**

Log Percent Poverty

-0.586*

-0.769**

2133.56

-1.715**

-0.029**

Log Percent Elderly

-0.801**

-0.963**

19263.672

-1.497*

-0.010

Log Percent Black

0.109

0.043

5595.065

0.234

0.007**

Log Black-White

Median Income Ratio

-0.375

-0.329

-75612.8

-0.284

-0.004

Black-White

Dissimilarity Index

0.715

0.211

-3062.585

0.995

-0.016

Log Population Density

0.290***

0.324***

-1582.833

0.683***

0.001

-0.277

-0.346*

-73911.5

-0.265

-0.008

Change in # of

Manufacturing

Establishments

-0.003*

0.001

138.150

0.001

-0.000

Change in # of

Manufacturing

Employees

0.000***

0.000

0.380

0.000

Change in

Manufacturing Wages

-0.000*

0.000

-8.706

0.000

0.000*

Change in # of Service

Establishments

0.000

34.867

0.000

-0.000

Change in # of Service

Establishments

-0.000

0.000

-0.476

-0.000

0.000

Change in Service

Wages

-0.000

1.166

-0.000

0.000

TRI Facilities 1988

0.012*

---

Log TRI Pounds 1988

---

-0.129***

---

Log TRI Pounds per

Facility 1988

---

-85890.3***

---

Log Median Risk 1988

---

-0.327***

Dirtiness Index 1988

---

-0.457***

Constant

2.928

5.390**

997161.9

11.199**

0.190**

0.272

0.040

0.030

0.059

0.255

Note: * p < 0.05; ** p < .01; *** p< .001 (two-tailed tests).

In panel three of Table 5.3, results are displayed for the OLS regression of the

change in log TRI pounds per facility on the independent variables described above. R2 is

.03 for this model. In this panel, a positive coefficient indicates a greater change in log

TRI pounds per facility between 1988 and 1997. Only one of the variables, log TRI

pounds per facility in 1988, is statistically significant. This variable indicates that for a

one unit increase in log TRI pounds per facility in 1988, there was a reduction in log TRI

pounds per facility of 85,890.3.

In panel four, results are displayed for the OLS regression of the change in log

median risk on the independent variables described above. In this panel, a positive

coefficient indicates a greater change in log median risk between 1988 and 1997. Five

variables are statistically significant and most of these conform to expectations in the

literature and the hypotheses described in the previous section. Log population density is

statistically significant and positive, while median family income squared, log percent

poverty, log percent elderly, and log median risk score in 1988 are each statistically

significant and negative. R2 is a rather anemic .04 for this model.

As with change in log TRI pounds, median family income squared is highly

significant and negative, indicating that as median family income squared increases, the

log median risk score decreases. Log percent elderly and log percent poverty are both

significant and negative and log population density is significant and positive. The log

median risk score in 1988 is also significant and negative, which means that as the log

median risk score in 1988 increases, the change in the log median risk score between

1988 and 1997 is negative. Finally, none of the change in manufacturing or service

variables is statistically significant.

In panel five, results are displayed for the OLS regression of the change in the

dirtiness index on the variables of interest. In this panel, a positive coefficient indicates a

greater increase in the dirtiness index between 1988 and 1997. Five variables are

statistically significant, with most of these conforming to expectations in the literature

and the hypotheses stated above. Log percent black and change in manufacturing wages

are each statistically significant and positive, while median family income squared, log

percent poverty, and dirtiness index score in 1988 are each statistically significant and

negative. R2 is .26 for this model, substantially better than for all variables in this table

except change in TRI facilities.

Median family income squared is significant and negative, indicating that as

median family income squared increases, the log median risk score decreases. Log

percent black is significant and positive, conforming to environmental justice

expectations that increases in African Americans are associated with an increased change

in the dirtiness index score. Log percent poverty is significant yet in the opposite

direction from that expected. Also significant and negative is the dirtiness index score in

1988, which means that as the dirtiness index score in 1988 increases, the change in the

dirtiness index score between 1988 and 1997 is negative. Finally, change in

manufacturing wages is significant and positive, which means that as manufacturing

wages increase, the change in the dirtiness index score is greater.

Table 5.3 presents some rather different results than those in Tables 5.1 and 5.2,

indicating that changes between 1988 and 1997 were somewhat distinctive in character

than the compositional results for those individual years. There is a relative lack of

support for both the ecological modernization/Environmental Kuznets Curve hypothesis

(H5.1), and the EC hypothesis (H5.2); relatively negative findings regarding the EJ

hypothesis (H5.3), with little support for the race component and opposite to expected

findings for poverty and the elderly; mixed findings regarding the economic composition

hypothesis (H5.4); and limited support for the hypotheses linked to the control variables,

population density (H5.5) and in-migration (H5.6).

In Table 5.4 panels one through three, change in TRI facilities between 1988 and

1997 is regressed on economic standing, environmental justice, economic change, and

other factors of interest. Economic change is measured here with three dummy variables

each representing the change in service-manufacturing establishments, wages, and

employment. These are the variables discussed in Chapter Four, which involve attempt to

characterize changes across both the manufacturing and service sectors between 1988

and 1997. These changes could include: growth in both sectors; decline in both sectors;

growth in manufacturing and decline in service; growth in service and decline in

manufacturing; or relative stability across one or both sectors. The dummy variables

incorporate changes in manufacturing and service establishments, wages, and

employment.

In panel one, results are displayed for the OLS regression of the change in TRI

facilities on the variables of interest, including the shift in service and manufacturing

establishments between 1988 and 1997. In this panel, a positive coefficient indicates a

higher number of TRI facilities. Several variables are statistically significant and several

conform to expectations in the literature and the hypotheses stated above. Median family

income, the black-white dissimilarity index, and log population density are each

statistically significant and positive, while number of TRI facilities in 1988, median

family income squared, log percent poverty, log percent elderly, and log in-migration are

each statistically significant and negative. R2 is .14 for this model, which means that 14%

of the variance in change in TRI facilities is explained by the independent variables

included in the model.

Again, median family income and median family income squared are significant

yet in the opposite direction, pointing to a curvilinear relationship. This finding conforms

to the EM/EKC hypothesis. The black-white dissimilarity index is significant and

positive, which is an environmental justice finding of interest. Log percent elderly and

log percent poverty are both significant yet in the opposite direction from that expected

(i.e. as each increases, change in the TRI facilities is negative). Log population density is

significant and positive, indicating that as log population density increases, the change in

the number of TRI facilities between 1988 and 1997 grows. Log in-migration is again

significant and negative, thus as log in-migration grows, the change in the number of TRI

facilities is negative. Also significant and positive is the number of TRI facilities in 1988,

which means that as the number of facilities in 1988 increases, the change in the number

of TRI facilities between 1988 and 1997 is negative. Finally, there are no significant

results for the dummy variables related to the compositional shift in manufacturing and

service between 1988 and 1997.

Table 5.4

Change in TRI Facilities Regressed on Economic Change, Economic Standing,

Environmental Justice, and Other Factors, 1988-1997

Model 1

Model 2

Model 3

TRI Facilities 1988

-0.049***

Median Family Income

0.000**

0.000***

Median Family Income

Squared

-0.000***

Log Percent Poverty

-0.800***

-0.791***

-0.810***

Log Percent Elderly

-1.339***

-1.331***

-1.337***

Log Percent Black

0.098

0.099

0.107

Log Black-White Median

Income Ratio

-0.381

-0.388

-0.382

Black-White Dissimilarity

Index

0.881*

0.897*

0.882*

Log Population Density

0.309***

0.311***

0.316***

-0.298*

-0.303*

-0.304*

Economic Change

Service-Manufacturing

Establishments Manufacturing

Growth Dummy

-0.053

---

Service-Manufacturing

Establishments Service

Growth Dummy

0.036

---

Service-Manufacturing

Establishments Decline

Dummy

-0.201

---

Service-Manufacturing

Wages Manufacturing Growth

Dummy

---

-0.399

---

Service-Manufacturing

Wages Service Growth

Dummy

---

-0.216

---

Service-Manufacturing Wages

Decline Dummy

---

-0.101

---

Service-Manufacturing

Employees Manufacturing

Growth Dummy

---

-0.041

Service-Manufacturing

Employees Service Growth

Dummy

---

-0.101

Service-Manufacturing

Employees Decline Dummy

---

-0.211

Constant

4.659**

4.676**

4.701**

0.139

0.140

0.139

Note: * p < 0.05; ** p < .01; *** p< .001 (two-tailed tests).

In panel two of Table 5.4, results are displayed for the OLS regression of the

change in TRI facilities on the variables of interest, including the shift in service and

manufacturing wages between 1988 and 1997. The same variables are statistically

significant and point in the same direction as in panel one of Table 5.4. Median family

income, the black-white dissimilarity index, and log population density are each

statistically significant and positive, while number of TRI facilities in 1988, median

family income squared, log percent poverty, log percent elderly, and log in-migration are

each statistically significant and negative. Substantive interpretation for the significant

findings is the same here as is described for Table 5.4 panel one. As with the previous

model, R2 is .14.

In Table 5.4, panel three, results are displayed for the OLS regression of the

change in TRI facilities on the variables of interest, including the shift in service and

manufacturing employees between 1988 and 1997. The same variables are statistically

significant and point in the same direction as in panel one and panel two of Table 5.4.

Median family income, the black-white dissimilarity index, and log population density

are each statistically significant and positive, while number of TRI facilities in 1988,

median family income squared, log percent poverty, log percent elderly, and log

inmigration are each statistically significant and negative. R2 is again .14 for this model.

The results in Table 5.4 provide relatively strong support for the ecological

modernization/Environmental Kuznets Curve hypothesis (H5.1), and thus opposite to

expected results for the EC hypothesis (H5.2); negative findings regarding the EJ

hypothesis (H5.3), with no support for the race component and opposite to expected

findings for poverty and the elderly; no significant findings for the economic composition

hypothesis (H5.4); and consistent support for the hypotheses linked to population density

(H5.5) and in-migration (H5.6).

In Table 5.5 panels one through three, change in log TRI pounds between 1988

and 1997 is regressed on economic standing, environmental justice, economic change,

and other factors of interest. Economic change is measured here with three dummy

variables each representing the change in service-manufacturing establishments, wages,

and employment.

In panel one, results are displayed for the OLS regression of the change in log

TRI pounds on the variables of interest, including the shift in service and manufacturing

establishments between 1988 and 1997. In this panel, a positive coefficient indicates a

positive change in log TRI pounds. Seven variables are statistically significant and

several conform to expectations in the literature and the hypotheses stated above. Log

population density is statistically significant and positive, while log TRI pounds in 1988,

median family income squared, log percent poverty, log percent elderly, log in-migration,

and the service-manufacturing establishments decline dummy variable are each

statistically significant and negative. R2 is a rather low .05 for this model.

Median family income squared is significant and negative, a finding somewhat

distinctive from both the EM/EKC and EC hypotheses discussed earlier in the chapter.

Log percent elderly and log percent poverty are both significant yet negative, log

population density is significant and positive, and log in-migration is significant and

negative. Also significant and negative is the log TRI pounds in 1988, which means that

as the number of log TRI pounds in 1988 increases. Finally, the service-manufacturing

establishments decline dummy variable is significant and negative, which means that

counties which experienced an overall decline in both manufacturing and service

establishments between 1988 and 1997 experienced a 2.213 unit decline in log TRI

pounds during the same time period.

In panel two of Table 5.5, results are displayed for the OLS regression of the

change in log TRI pounds on the variables of interest, including the shift in service and

manufacturing wages between 1988 and 1997. Six variables are statistically significant

and several conform to expectations in the literature and the hypotheses stated above.

Log population density is statistically significant and positive, while log TRI pounds in

1988, median family income squared, log percent poverty, log percent elderly and log

inmigration are each statistically significant and negative. R2 for this model is .04.

Substantive interpretation for the significant findings is the same here as is described for

Table 5.5 panel one.

In Table 5.5, panel three, results are displayed for the OLS regression of the

change in log TRI pounds on the variables of interest, including the shift in service and

manufacturing employees between 1988 and 1997. The same variables are statistically

significant and point in the same direction as in panel two of Table 5.5. Substantive

interpretation for the significant findings is again the same as is described for Table 5.5

panel one. R2 is again a low .04 for this model.

Table 5.5

Change in Log TRI Pounds Regressed on Economic Change, Economic Standing,

Environmental Justice, and Other Factors, 1988-1997

Model 1

Model 2

Model 3

Log TRI Pounds 1988

-0.119***

-0.114***

Median Family Income

0.000

Median Family Income

Squared

-0.000**

Log Percent Poverty

-0.785**

-0.784**

-0.809**

Log Percent Elderly

-1.115***

-1.104***

-1.098***

Log Percent Black

0.030

0.040

0.050

Log Black-White Median

Income Ratio

-0.226

-0.232

-0.238

Black-White Dissimilarity

Index

0.445

0.443

0.432

Log Population Density

0.361***

0.372***

0.382***

-0.379**

-0.381**

-0.387**

Economic Change

Service-Manufacturing

Establishments Manufacturing

Growth Dummy

-0.177

---

Service-Manufacturing

Establishments Service

Growth Dummy

0.083

---

Service-Manufacturing

Establishments Decline

Dummy

-2.213***

---

Service-Manufacturing

Wages Manufacturing Growth

Dummy

---

-0.192

---

Service-Manufacturing

Wages Service Growth

Dummy

---

-0.012

---

Service-Manufacturing Wages

Decline Dummy

---

0.142

---

Service-Manufacturing

Employees Manufacturing

Growth Dummy

---

-0.015

Service-Manufacturing

Employees Service Growth

Dummy

---

-0.215

Service-Manufacturing

Employees Decline Dummy

---

-0.538

Constant

5.965***

5.773***

5.872***

0.051

0.042

0.043

Note: * p < 0.05; ** p < .01; *** p< .001 (two-tailed tests).

The results in Table 5.5 provide no significant results regarding the ecological

modernization/Environmental Kuznets Curve hypothesis (H5.1), or the EC hypothesis

(H5.2); negative findings regarding the EJ hypothesis (H5.3), with no support for the race

component and opposite to expected findings for poverty and the elderly; limited findings

for the economic composition hypothesis (H5.4); and consistent support for the

hypotheses related to population density (H5.5) and in-migration (H5.6).

In Table 5.6 panels one through three, change in log TRI pounds per facility

between 1988 and 1997 is regressed on economic standing, environmental justice,

economic change, and other factors of interest. Economic change is measured here with

three dummy variables each representing the change in service-manufacturing

establishments, wages, and employment.

In panel one, results are displayed for the OLS regression of the change in log

TRI pounds per facility on the variables of interest, including the shift in service and

manufacturing establishments between 1988 and 1997. In this panel, a positive

coefficient indicates a positive change in log TRI pounds per facility. Five variables are

statistically significant and most conform to expectations in the literature and the

hypotheses stated above. Log TRI pounds per facility in 1988, median family income

squared, log percent elderly, log in-migration, and the service-manufacturing

establishments decline dummy variable are each statistically significant and negative. R2

is .17 for this model, a substantial improvement as compared to the last set of models.

Median family income squared is significant and negative. Log percent elderly is

significant yet in the opposite direction from that expected: as each increases, change in

the log TRI pounds per facility is negative. Log in-migration is significant and negative;

i.e. as log in-migration grows, the change in log TRI pounds per facility is negative. Also

significant and negative is the log TRI pounds per facility in 1988, meaning that as the

number of log TRI pounds per facility in 1988 increases, the change in log TRI pounds

per facility between 1988 and 1997 is negative. Finally, the service-manufacturing

establishments decline dummy variable is significant and negative. Counties which

experienced an overall decline in both manufacturing and service establishments between

1988 and 1997 experienced a 1.074 unit decline in log TRI pounds per facility during the

same time period.

In panel two and panel three of Table 5.6, results are displayed for the OLS

regression of the change in log TRI pounds per facility on the variables of interest. In

panel two, this includes the variables described above, along with the shift in service and

manufacturing wages between 1988 and 1997. In panel three, this includes the variables

described above and the shift in service and manufacturing employees between 1988 and

1997. In both models, four variables are statistically significant and several conform to

expectations in the literature and the hypotheses stated above. Log TRI pounds per

facility in 1988, median family income squared, log percent elderly and log in-migration

are each statistically significant and negative. R2 is .17 for both models.

Table 5.6

Change in Log TRI Pounds per Facility Regressed on Economic Change, Economic

Standing, Environmental Justice, and Other Factors, 1988-1997

Model 1

Model 2

Model 3

Log TRI Pounds per Facility

1988

-0.351***

-0.350***

Median Family Income

0.000

Median Family Income Squared

-0.000**

Log Percent Poverty

-0.144

-0.151

-0.163

Log Percent Elderly

-0.522**

-0.518**

-0.519**

Log Percent Black

0.023

0.028

0.027

Log Black-White Median Income

Ratio

-0.173

-0.174

-0.188

Black-White Dissimilarity Index

0.105

0.096

0.084

Log Population Density

0.058

0.065

0.064

-0.229**

-0.231**

Economic Change

Service-Manufacturing

Establishments Manufacturing

Growth Dummy

0.179

Service-Manufacturing

Establishments Service Growth

Dummy

0.020

---

Service-Manufacturing

Establishments Decline Dummy

-1.074**

---

Service-Manufacturing Wages

Manufacturing Growth Dummy

---

0.019

---

Service-Manufacturing Wages Service

Growth Dummy

---

0.119

---

Service-Manufacturing Wages Decline

Dummy

---

0.050

---

Service-Manufacturing

Employees Manufacturing

Growth Dummy

---

-0.177

Service-Manufacturing

Employees Service Growth

Dummy

---

-0.063

Service-Manufacturing

Employees Decline Dummy

---

-0.860

Constant

5.905***

5.830***

5.953***

0.172

0.168

0.169

Note: * p < 0.05; ** p < .01; *** p< .001 (two-tailed tests).

The results in Table 5.6 provide mixed results for both the ecological

modernization/Environmental Kuznets Curve hypothesis (H5.1), and the EC hypothesis

(H5.2); weak findings regarding the EJ hypothesis (H5.3), with no support for the race or

poverty components, and opposite to expected findings for the elderly; limited findings

for the economic composition hypothesis (H5.4); no support for the population density

hypothesis (H5.5); and consistent support for the in-migration hypothesis (H5.6).

In Table 5.7 panels one through three, change in the log median risk score

between 1988 and 1997 is regressed on economic standing, environmental justice,

economic change, and other factors of interest.

In panel one, results are displayed for the OLS regression of the change in the log

median risk score on the variables of interest, including the shift in service and

manufacturing establishments between 1988 and 1997. In this panel, a positive

coefficient indicates a positive change in the log median risk score. Six variables are

statistically significant and most conform to expectations in the literature and the

hypotheses stated above. Log population density is statistically significant and positive,

while the log median risk score in 1988, median family income squared, log percent

elderly, log in-migration, and the service-manufacturing establishments decline dummy

variable are each statistically significant and negative. R2 is a rather paltry .07 for this

model.

Median family income squared is significant and negative. Log percent elderly is

significant yet in the opposite direction from that expected (i.e. as each increases, change

in the log median risk score is negative). Log in-migration is significant and negative,

thus as log in-migration grows, the change in the log median risk score is negative. Also

significant and negative is the log median risk score in 1988, which means that as the

number of the log median risk score in 1988 increases, the change in the log median risk

score between1988 and 1997 is negative. Finally, the service-manufacturing

establishments decline dummy variable is significant and negative, which means that

counties which experienced an overall decline in both manufacturing and service

establishments between 1988 and 1997 experienced a 5.264 unit decline in the log

median risk score during the same time period.

In panel two and panel three of Table 5.7, results are displayed for the OLS

regression of the change in the log median risk score on the variables of interest. In panel

two, this includes the variables described above, along with the shift in service and

manufacturing wages between 1988 and 1997. In panel three, this includes the variables

described above and the shift in service and manufacturing employees between 1988 and

1997.

In both models, five variables are statistically significant and several conform to

expectations in the literature and the hypotheses stated above. Log population density is

significant and positive, while the log median risk score in 1988, median family income

squared, log percent elderly and log in-migration are each statistically significant and

negative. In both models, R2 equals .06.

Table 5.7

Change in Log Median Risk Regressed on Economic Change, Economic Standing,

Environmental Justice, and Other Factors, 1988-1997

Model 1

Model 2

Model 3

Log Median Risk 1988

-0.321***

-0.317***

-0.316***

Median Family Income

0.000

Median Family Income

Squared

-0.000**

Log Percent Poverty

-1.788**

-1.787**

-1.770**

Log Percent Elderly

-1.875**

-1.851**

-1.853**

Log Percent Black

0.201

0.225

0.238

Log Black-White Median

Income Ratio

-0.226

-0.214

-0.221

Black-White Dissimilarity

Index

1.401

1.430

1.466

Log Population Density

0.768***

0.799***

0.815***

-0.389

-0.399

-0.405

Economic Change

Service-Manufacturing

Establishments Manufacturing

Growth Dummy

-0.739

---

Service-Manufacturing

Establishments Service

Growth Dummy

0.099

---

Service-Manufacturing

Establishments Decline

Dummy

-5.264***

---

Service-Manufacturing

Wages Manufacturing Growth

Dummy

---

0.183

---

Service-Manufacturing

Wages Service Growth

Dummy

---

0.038

---

Service-Manufacturing Wages

Decline Dummy

---

0.764

---

Service-Manufacturing

Employees Manufacturing

Growth Dummy

---

0.901

Service-Manufacturing

Employees Service Growth

Dummy

---

-0.287

Service-Manufacturing

Employees Decline Dummy

---

1.416

Constant

13.068**

12.530**

12.373**

0.070

0.061

Note: * p < 0.05; ** p < .01; *** p< .001 (two-tailed tests).

Table 5.7 provides mixed support for both the ecological modernization/

Environmental Kuznets Curve hypothesis (H5.1), and the EC hypothesis (H5.2);

relatively negative findings regarding the EJ hypothesis (H5.3), with no support for the

race component, and opposite to expected findings for poverty and the elderly; mixed

findings for the economic composition hypothesis (H5.4); support for the population

density hypothesis (H5.5); and non-significant results for the in-migration hypothesis

(H5.6).

In Table 5.8 panels one through three, change in the dirtiness index between 1988

and 1997 is regressed on economic standing, environmental justice, economic change,

and other factors of interest.

In panel one, results are displayed for the OLS regression of the change in the

dirtiness index on the variables of interest, including the shift in service and

manufacturing establishments between 1988 and 1997. In this panel, a positive

coefficient indicates a positive change in the dirtiness index score. Five variables are

statistically significant and most conform to expectation. Log percent black and the

service-manufacturing establishments service growth dummy variable are each

statistically significant and positive, while the dirtiness index score in 1988, median

family income squared, and log percent poverty are all statistically significant and

negative. R2 is a relatively robust .24 for this model.

Median family income squared is significant and negative, again somewhat

conforming to the EM/EKC hypothesis. Log percent black is significant and positive,

which conforms to the environmental justice hypothesis (i.e. as the log percent black

increases, negative environmental conditions follow, here in the form of a positive

change in the county-level dirtiness index score). Log percent poverty is significant yet in

the opposite direction from that expected (i.e. as the log percent poverty increases,

change in the log median risk score is negative). The dirtiness index score in 1988 is also

significant and negative, which means that as the dirtiness index score in 1988 increases,

the change in the dirtiness index score between 1988 and 1997 is negative. Thus those

areas with higher dirtiness index scores in 1988 are likely to have declined somewhat

between 1988 and 1997, meaning that the “worst of the worst” counties improved

somewhat on a proportional basis. This could be attributed to an increase in non-polluting

manufacturing, a decrease in polluting facilities, or a combination of both options.

Finally, the service-manufacturing establishments service growth dummy variable is

significant and positive, so counties that experienced an overall increase of greater than

3% in service establishments between 1988 and 1997 experienced a 0.011 unit increase

in the dirtiness index score during the same time period.

In panels two and three of Table 5.8, results are displayed for the OLS regression

of the change in the dirtiness index score on the variables of interest, including the shift

in service and manufacturing wages between 1988 and 1997. R2 is .23 in panel two and

.24 in panel three.

Table 5.8

Change in Dirtiness Index Regressed on Economic Change, Economic Standing,

Environmental Justice, and Other Factors, 1988-1997

Model 1

Model 2

Model 3

Dirtiness Index 1988

-0.447***

-0.449***

Median Family Income

0.000

Median Family Income Squared

-0.000**

Log Percent Poverty

-0.028**

-0.030**

Log Percent Elderly

-0.013

Log Percent Black

0.006**

0.007**

0.008***

Log Black-White Median Income

Ratio

-0.002

-0.003

-0.002

Black-White Dissimilarity Index

-0.021

-0.020

Log Population Density

-0.000

0.000

0.001

-0.007

-0.006

-0.008

Economic Change

Service-Manufacturing

Establishments Manufacturing

Growth Dummy

0.025

Service-Manufacturing

Establishments Service Growth

Dummy

0.011**

---

Service-Manufacturing

Establishments Decline Dummy

-0.015

---

Service-Manufacturing Wages

Manufacturing Growth Dummy

---

-0.007

---

Service-Manufacturing Wages Service

Growth Dummy

---

-0.005

---

Service-Manufacturing Wages Decline

Dummy

---

-0.006

---

Service-Manufacturing

Employees Manufacturing

Growth Dummy

---

-0.017

Service-Manufacturing

Employees Service Growth

Dummy

---

-0.014**

Service-Manufacturing

Employees Decline Dummy

---

-0.005

Constant

0.195***

0.200***

0.214***

0.235

0.232

0.235

Note: * p < 0.05; ** p < .01; *** p< .001 (two-tailed tests).

In panel two, four variables are statistically significant, while in panel three, five

variables are statistically significant. In each case, most conform to the previous literature

and the hypotheses stated earlier. In both panels, log percent black is statistically

significant and positive, while the dirtiness index score in 1988, median family income

squared, and log percent poverty are each statistically significant and negative. In panel

three, the service-manufacturing employees service growth dummy variable is significant

and negative. This finding means that counties which experienced an overall increase of

greater than 3% in service employees between 1988 and 1997 experienced a 0.014 unit

decrease in the dirtiness index score during the same time period.

The results in Table 5.8 again provide mixed results for both the EM/EKC

hypothesis (H5.1), and thus the EC hypothesis (H5.2); mixed findings regarding the EJ

hypothesis (H5.3), with some support for the race component, and opposite to expected

findings for poverty; mixed findings for the economic composition hypothesis (H5.4);

and non-significant findings for the population density hypothesis (H5.5) and the

inmigration hypothesis (H5.6).

Overall in this section, there have been clearly mixed results observed. Generally

speaking there appears to be more support for the EM/EKC hypothesis (H5.1) than the

EC hypothesis (H5.2), a sign that some sort of inverted U-curve relationship exists

between median income and environmental quality (first it worsens, then it gets better at

some threshold point). This is not a uniform finding though, as in some cases there is

simply improving environmental quality as median income increases. As was noted

earlier in the chapter, this could represent the exportation of highly polluting facilities to

other poorer locales, either internationally or within the United States. This important

issue will be examined in the next chapter on Domestic Pollution Havens. Findings

related to the environmental justice hypotheses (H5.3) were somewhat mixed, as

increased racial diversity was often associated with lowered environmental quality

(expected), but percent elderly and percent in poverty were each often significant and

pointed in the opposite to expected direction. There were not very many significant

findings for the economic composition or economic change hypotheses (H5.4). However,

population density (H5.5) and level of in-migration (H5.6) were each typically significant

and in the expected direction (positive for population density and negative for

inmigration). In the next section, attention is turned to the second question posed in the

chapter introduction: “has the „great slide‟ of manufacturing affected local pollution

levels?”

Has the Alleged “Great Slide” of Manufacturing affected Local Pollution Levels?

The second question this chapter is designed to address involves the alleged

“great slide” of manufacturing (i.e. the de-industrialization hypothesis), whether or not

this compositional change has impacted environmental quality across local space, and if

so, how it has done so. To begin the section and the process of addressing this question,

Table 5.9 presents the results of the regression of changes in manufacturing

establishments, employment, and wages on economic standing, environmental justice,

and other variables of interest.

In panel one, change in manufacturing establishments between 1988 and 1997 is

regressed on the variables of interest. In this panel, a positive coefficient indicates a

positive change in the manufacturing establishments. Four variables are statistically

significant, including log in-migration, which is statistically significant and positive,

while log percent poverty, log percent elderly, the black-white dissimilarity index, and

log population density are each statistically significant and negative. R2 for this model is

a low .04.

Log percent poverty, log percent elderly, and the black-white dissimilarity index

each point in the direction which would be expected in relation to social justice

hypotheses; i.e. as each increases, the change in manufacturing establishments is

negative. Log in-migration is statistically significant and positive, indicating that an

increase is linked to a positive change in manufacturing establishments. Log population

density is statistically significant and negative; increases in log population density are

linked to declines in manufacturing establishments.

In panel two of Table 5.9, change in manufacturing employment between 1988

and 1997 is regressed on the variables of interest. In this panel, a positive coefficient

indicates a positive change in the manufacturing employment. Six variables are

statistically significant, including log in-migration, which is statistically significant and

positive, while median family income, log percent poverty, log percent elderly, the

blackwhite dissimilarity index, and log population density are each statistically

significant and negative. R2 is .09 for this model.

Table 5.9

Change in Manufacturing Regressed on Economic Standing, Environmental Justice and

Other Factors, 1988-1997

Change in

Establishments

Change in

Employment

Change in Wages

Median Family

Income

-0.001

-0.352**

-0.193

Median Family

Income Squared

-0.000

Log Percent Poverty

-26.807***

-3012.567***

-898.741*

Log Percent Elderly

-40.686***

-3548.566***

-141.541

Log Percent Black

-2.053

-150.291

302.822**

Log Black-White

Median Income

Ratio

4.873

415.364

222.137

Black-White

Dissimilarity Index

-15.716*

-1373.491*

-2305.964**

Log Population Density

-2.504*

-515.875***

1.580

6.017*

1456.050***

-79.698

Constant

210.646***

21375.588***

0.214***

0.046

0.091

0.025

Note: * p < 0.05; ** p < .01; *** p< .001 (two-tailed tests).

Median family income is significant and negative, which means that as median

family income increases, there is a negative change in manufacturing employment. Log

percent poverty, log percent elderly, and the black-white dissimilarity index again are

each significant and negative, meaning that as each increases, the change in

manufacturing employees is negative. This again is a finding consistent with social

justice expectations; i.e. as socially disadvantaged groups increase, manufacturing

employment decreases. Log in-migration is statistically significant and positive,

indicating that an increase is linked to a positive change in manufacturing establishments.

Log population density is statistically significant and negative, which indicates that

increases are linked to negative change in manufacturing establishments.

In Table 5.9 panel three, change in manufacturing wages between 1988 and 1997

is regressed on the variables of interest. In this panel, a positive coefficient indicates a

positive change in manufacturing wages. Three variables are statistically significant,

including log percent black, which is statistically significant and positive, while log

percent poverty and the black-white dissimilarity index are each statistically significant

and negative. R2 is very low in this model, equaling .03.

The finding related to log percent black runs counter to social justice

expectations: i.e. as log percent black increases, the change in manufacturing wages is

positive. Of course the jobs in manufacturing could be going to other groups, so this link

in itself does not necessarily counter the other social inequality findings. On the other

hand, log percent poverty and the black-white dissimilarity index again are each

significant and negative, which means that as each increases, the change in

manufacturing wages is negative. This finding does conform to the social justice

expectations.

In Table 5.10 panels one through five, ordinary least squares regression is used to

regress the change in five environmental toxicity variables between 1988 and 1997 on

dirtiness index change, environmental conditions in 1988, economic standing,

environmental justice, economic change, and other factors of interest. In panel one, the

change in TRI facilities between 1988 and 1997 is regressed on the variables of interest.

In this panel, a positive coefficient indicates a positive change in TRI facilities. Several

variables are statistically significant and most conform to expectations in the literature

and the hypotheses stated above. Change in the dirtiness index score between 1988 and

1997, change in the number of manufacturing employees, the black-white dissimilarity

index, and log population density are statistically significant and positive, while change

in the number of manufacturing establishments, change in manufacturing wages, median

family income squared, log percent poverty, and log percent elderly are significant and

negative. R2 is a relatively robust .31 for this model.

Change in the dirtiness index score is significant and positive, indicating that as

the dirtiness index score experiences a positive change, the change in TRI facilities is

also positive. The median family income squared finding is again significant and

negative. Regarding the environmental justice hypothesis, findings are again mixed, as

the black-white dissimilarity index is significant and positive (expected), but the log

percent poverty and log percent elderly findings are significant and negative. Log

population density is significant and positive, indicating that as log population density

increases, the change in TRI facilities is positive. Finally, there are mixed findings for the

relationship between changes in manufacturing and TRI facilities, as changes in

manufacturing employees are significant and positive, while changes in manufacturing

establishments and wages are significant and negative.

In panel two of Table 5.10, the change in TRI facilities between 1988 and 1997 is

again regressed on the variables of interest, though the number of TRI facilities in 1988

replaces the log TRI pounds in 1988. Ten variables are statistically significant, including

number of TRI facilities in 1988. The other nine variables were significant in the

previous model, and in each case again point in the same direction. For the number of

TRI facilities in 1988, the finding is significant and positive, which indicates that an

Table 5.10

Change in Pollution Regressed on Economic Change, Dirtiness Index Change,

Environmental Conditions 1988, Economic Standing, Environmental Justice,

and Other Factors, 1988-1997

Change in

TRI

Facilities

Change in

TRI

Facilities

Change in

Log TRI

Pounds

Change in

Log TRI

Pounds per

Facility

Change in

Log

Median

Risk

Dirtiness Index Change

6.435***

6.493***

5.398***

0.125

11.244***

Log TRI Pounds 1988

0.062

---

-0.137***

---

TRI Facilities 1988

---

0.013*

---

Log TRI Pounds per

Facility 1988

---

-0.377***

---

Log Median Risk 1988

---

-0.404***

Change in # of

Manufacturing

Establishments

-0.003*

-0.004*

0.001

0.000

0.001

Change in # of

Manufacturing

Employees

0.000***

0.000

-0.000

Change in # of

Manufacturing Wages

-0.000**

0.000

Change in # of Service

Establishments

-0.000

0.000

Change in # of Service

Establishments

0.000

-0.000

0.000

-0.000

Change in # of Service

Wages

-0.000

Median Family Income

0.000

0.000*

Median Family Income

Squared

-0.000***

-0.000**

-0.000***

Log Percent Poverty

-0.922**

-0.812*

-0.613*

-0.084

-1.329*

Log Percent Elderly

-1.309**

-1.249**

-1.032***

-0.537**

-1.764**

Log Percent Black

0.087

0.086

0.044

0.025

0.240

Log Black-White

Median Income Ratio

-0.302

-0.294

-0.309

-0.302

-0.131

Black-White

Dissimilarity Index

1.145*

1.107*

0.190

0.100

1.011

Log Population Density

0.349***

0.339***

0.198**

0.032

0.376**

-0.284

-0.268

-0.319*

-0.258**

-0.299

Constant

4.750*

4.372*

5.543***

6.012***

13.738***

0.311

0.313

0.088

0.181

0.096

Note: * p < 0.05; ** p < .01; *** p< .001 (two-tailed tests).

increase in the number of facilities in 1988 is associated with a positive change in TRI

facilities between 1988 and 1997. Again, R2 is .31 for this model.

In panel three, ordinary least squares regression is used to regress the change in

log TRI pounds between 1988 and 1997 on the variables of interest. Eight variables are

statistically significant, with most of these conforming to expectations. Change in the

dirtiness index score between 1988 and 1997, median family income, and log population

density are statistically significant and positive, while log TRI pounds in 1988, median

family income squared, log percent poverty, log percent elderly, and log in-migration are

significant and negative. R2 is a rather low .09 for this model.

Change in the dirtiness index score is again significant and positive, indicating

that as the dirtiness index score experiences a positive change, the change in log TRI

pounds is also positive. The median family income and median family income squared

findings conform to the EM/EKC hypothesis; i.e. increased median family income is

associated with a positive change in log TRI pounds up to a certain threshold point, at

which time higher median family income is associated with a negative change in log TRI

pounds. Regarding the environmental justice hypothesis, findings are again opposite to

that expected, since both the log percent poverty and log percent elderly findings are

significant and negative. Log population density is significant and positive, indicating

that as log population density increases, the change in log TRI pounds is positive. Finally,

increases, the change in log TRI pounds is negative.

In panel four, ordinary least squares regression is used to regress the change in log

TRI pounds per facility between 1988 and 1997 on the variables of interest. Five

variables are statistically significant and some conform to expectations in the literature

and the hypotheses stated above. Median family income is statistically significant and

positive, while log TRI pounds per facility in 1988, median family income squared, log

percent elderly, and log in-migration are significant and negative. R2 is better than in the

previous model, equaling .18.

Log TRI pounds per facility in 1988 is highly significant and negative, which

means that as log TRI pounds per facility in 1988 increases, the change in log TRI

pounds per facility is negative. The median family income and median family income

squared findings again conform to the EM/EKC hypothesis; i.e. increased median family

income is associated with a positive change in log TRI pounds per facility up to a certain

threshold point, at which time higher median family income is associated with a negative

change in log TRI pounds per facility. Regarding the environmental justice hypothesis,

findings are again opposite to that expected, since only log percent elderly is significant,

and it is negative. Finally, log in-migration is again significant and negative, which

means that as log in-migration increases, the change in log TRI pounds per facility is

negative.

In Table 5.10 panel five, ordinary least squares regression is used to regress the

change in log median risk score between 1988 and 1997 on the variables of interest.

Seven variables are statistically significant and most conform to expectations in the

literature and the hypotheses stated above. Dirtiness index score change, median family

income, and log population density are statistically significant and positive, while log

median risk score in 1988, median family income squared, log percent poverty, and log

percent elderly are significant and negative. R2 equals .10 for this model.

Log median risk score in 1988 is highly significant and negative, which means

that as the log median risk score in 1988 increases, the change in the log median risk

score is negative. The median family income and median family income squared findings

once again conform to the EM/EKC hypothesis. Regarding the environmental justice

hypothesis, findings are again opposite to that expected, since log percent poverty and

log percent elderly findings are significant and negative. Finally, log population density is

again significant and positive.

The results in Table 5.10 provide relatively strong support for the ecological

modernization/Environmental Kuznets Curve hypothesis (H5.1), and thus strong

opposition to the EC hypothesis (H5.2). There are relatively negative findings regarding

the EJ hypothesis (H5.3), with weak support for the race component, and consistently

opposite to expected findings for poverty and the elderly. Mixed findings are observed

for the economic composition hypothesis (H5.4). Mixed support also exists for the

population density hypothesis (H5.5) and the in-migration hypothesis (H5.6).

In Table 5.11 panels one through three, change in TRI facilities between 1988 and

1997 is regressed on change in the dirtiness index score, TRI facilities in 1988, economic

change, economic standing, environmental justice, and other factors of interest. As earlier

in the chapter, economic change is measured here with three dummy variables each

representing the change in service-manufacturing establishments, wages, and

employment.

In panel one, results are displayed for the OLS regression of the change in TRI

facilities on the variables of interest. In this panel, a positive coefficient indicates a

positive change in TRI facilities between 1988 and 1997. Several variables are

statistically significant and some conform to expectations in the literature and the

hypotheses stated above. Change in the dirtiness index, median family income, the

blackwhite dissimilarity index, and log population density are each statistically

significant and positive, while number of TRI facilities in 1988, median family income

squared, log percent poverty, log percent elderly, and log in-migration are each

statistically significant and negative. R2 equals .16 for this model.

Change in the dirtiness index between 1988 and 1997 is significant and positive,

which means that as positive change in the dirtiness index occurs, positive change in TRI

facilities also is observed. Also, as has been observed in many of the models in this

chapter, median family income and median family income squared are significant yet in

the opposite direction, pointing to a curvilinear relationship. This finding conforms to the

EM/EKC hypothesis. The black-white dissimilarity index is significant and positive,

which is an environmental justice finding of interest. Log percent elderly and log percent

poverty are both significant yet in the opposite direction from that expected (i.e. as each

increases, change in the TRI facilities is negative). Log population density is significant

and positive, indicating that as log population density increases, the change in the number

of TRI facilities between 1988 and 1997 is positive. Log in-migration is again significant

and negative, thus as log in-migration grows, the change in the number of TRI facilities

is negative. Also significant and positive is the number of TRI facilities in 1988, which

means that as the number of facilities in 1988 increases, the change in the number of TRI

facilities between 1988 and 1997 is negative.

Table 5.11

Change in TRI Facilities Regressed on Economic Change, Dirtiness Index Change, TRI

Facilities 1988, Economic Standing, Environmental Justice, and Other Factors,

1988-1997

Model 1

Model 2

Model 3

Dirtiness Index Change19881997

3.668***

3.625***

3.636***

TRI Facilities 1988

-0.049***

-0.050***

-0.049***

Median Family Income

0.000**

0.000***

Median Family Income Squared

-0.000***

Log Percent Poverty

-1.090***

-1.068***

-1.090***

Log Percent Elderly

-1.697***

-1.690***

-1.695***

Log Percent Black

0.117

0.115

0.122

Log Black-White Median Income

Ratio

-0.361

-0.358

Black-White Dissimilarity Index

1.297**

1.300**

Log Population Density

0.351***

0.349***

0.356***

-0.333*

-0.336*

Economic Change

Service-Manufacturing

Establishments Manufacturing

Growth Dummy

-0.220

Service-Manufacturing

Establishments Service Growth

Dummy

-0.028

---

Service-Manufacturing

Establishments Decline Dummy

-0.285

---

Service-Manufacturing Wages

Manufacturing Growth Dummy

---

-0.380

---

Service-Manufacturing Wages Service

Growth Dummy

---

-0.183

---

Service-Manufacturing Wages Decline

Dummy

---

-0.033

---

Service-Manufacturing

Employees Manufacturing

Growth Dummy

---

-0.032

Service-Manufacturing

Employees Service Growth

Dummy

---

-0.073

Service-Manufacturing

Employees Decline Dummy

---

-0.339

Constant

6.544***

6.479***

6.510***

0.158

0.159

0.158

Note: * p < 0.05; ** p < .01; *** p< .001 (two-tailed tests).

In panel two of Table 5.11, results are displayed for the OLS regression of the

change in TRI facilities on the variables of interest, including the shift in service and

manufacturing wages between 1988 and 1997. The same variables are statistically

significant and point in the same direction as in panel one of Table 5.11. Change in the

dirtiness index, median family income, the black-white dissimilarity index, and log

population density are each statistically significant and positive, while number of TRI

facilities in 1988, median family income squared, log percent poverty, log percent

elderly, and log in-migration are each statistically significant and negative. R2 is again

equal to .16 for this model.

In Table 5.11, panel three, results are displayed for the OLS regression of the

change in TRI facilities on the variables of interest, including the shift in service and

manufacturing employees between 1988 and 1997. The same variables are statistically

significant and point in the same direction as in panel one and panel two of Table 5.11.

Change in the dirtiness index, median family income, the black-white dissimilarity index,

and log population density are each statistically significant and positive, while number of

TRI facilities in 1988, median family income squared, log percent poverty, log percent

elderly, and log in-migration are each statistically significant and negative. Once again, it

should be noted that there are no significant findings for the dummy variables related to

the compositional shift in manufacturing and service employees between 1988 and 1997.

Finally, R2 is again .16 for this model.

The results in Table 5.11 provide support for the ecological modernization/

Environmental Kuznets Curve hypothesis (H5.1), and thus oppose the EC hypothesis

(H5.2). There are relatively negative findings regarding the EJ hypothesis (H5.3), with

weak support for the race component, and opposite to expected findings for poverty and

the elderly. Non-significant results emerged for the economic composition hypothesis

(H5.4). There is support for both the population density hypothesis (H5.5) and the

inmigration hypothesis (H5.6).

In Table 5.12 panels one through three, change in log TRI pounds between 1988

and 1997 is regressed on change in the dirtiness index score, TRI facilities in 1988,

economic change, economic standing, environmental justice, and other factors of interest.

In panel one, results are displayed for the OLS regression of the change in log

TRI pounds on the variables of interest, including the shift in service and manufacturing

establishments between 1988 and 1997. Seven variables are statistically significant and

several conform to expectations in the literature and the hypotheses stated above.

Dirtiness index change and log population density are statistically significant and

positive, while log TRI pounds in 1988, log percent poverty, log percent elderly, log

inmigration, and the service-manufacturing establishments decline dummy variable are

each statistically significant and negative. R2 equals .09 for this model.

Change in the dirtiness index between 1988 and 1997 is significant and positive, which

means that as positive change in the dirtiness index occurs, positive change in log TRI

pounds also is observed. Log TRI pounds in 1988 is significant and negative, which

means that as the number of log TRI pounds in 1988 increases, the change in log TRI

pounds between 1988 and 1997 is negative. Log percent elderly and log percent poverty

are both significant yet in the opposite direction from that expected (i.e. as each

increases, change in the log TRI pounds is negative). Log population density is

significant and positive, indicating that as log population density increases, the change in

log TRI pounds between 1988 and 1997 is positive. Log in-migration is significant and

negative, thus as log in-migration grows, the change in log TRI pounds is negative.

Finally, the service-manufacturing establishments decline dummy variable is significant

and negative, which means that counties which experienced an overall decline in both

manufacturing and service establishments between 1988 and 1997 experienced a 2.021

unit decline in log TRI pounds during the same time period.

In panel two of Table 5.12, results are displayed for the OLS regression of the

change in log TRI pounds on the variables of interest, including the shift in service and

manufacturing wages between 1988 and 1997. Eight variables are statistically significant

in this model. Dirtiness index change, median family income, and log population density

are statistically significant and positive, while log TRI pounds in 1988, median family

income squared, log percent poverty, log percent elderly, log in-migration, and the

service-manufacturing establishments decline dummy variable are each statistically

significant and negative. Again, R2 is .09 for this model.

Median family income and median family income squared are significant yet in

the opposite direction, pointing again to a curvilinear relationship which conforms to the

EM/EKC hypothesis. Substantive interpretation for the remaining significant findings is

the same here as is described for Table 5.12 panel one.

Table 5.12

Change in Log TRI Pounds Regressed on Economic Change, Manufacturing/Service

Shift, Dirtiness Index Change, Log TRI Pounds 1988, Economic Standing,

Environmental Justice, and Other Factors, 1988-1997

Model 1

Model 2

Model 3

Dirtiness Index Change19881997

5.204***

5.244***

5.180***

Log TRI Pounds 1988

-0.132***

-0.126***

Median Family Income

0.000

0.000*

0.000

Median Family Income Squared

-0.000

-0.000***

Log Percent Poverty

-0.575*

-0.577*

-0.602**

Log Percent Elderly

-1.106***

-1.119***

-1.106***

Log Percent Black

0.044

0.050

0.061

Log Black-White Median Income

Ratio

-0.204

-0.197

-0.206

Black-White Dissimilarity Index

0.339

0.345

-0.329

Log Population Density

0.205**

0.211**

0.223***

-0.389**

-0.390**

-0.394**

Economic Change

Service-Manufacturing

Establishments Manufacturing

Growth Dummy

-0.241

Service-Manufacturing

Establishments Service Growth

Dummy

0.011

---

Service-Manufacturing

Establishments Decline Dummy

-2.021***

---

Service-Manufacturing Wages

Manufacturing Growth Dummy

---

-0.238

---

Service-Manufacturing Wages Service

Growth Dummy

---

-0.068

---

Service-Manufacturing Wages Decline

Dummy

---

0.213

---

Service-Manufacturing

Employees Manufacturing

Growth Dummy

---

-0.028

Service-Manufacturing

Employees Service Growth

Dummy

---

-0.208

Service-Manufacturing

Employees Decline Dummy

---

-0.560

Constant

5.769***

5.692***

5.747***

0.091

0.085

0.086

Note: * p < 0.05; ** p < .01; *** p< .001 (two-tailed tests).

In Table 5.12, panel three, results are displayed for the OLS regression of the

change in log TRI pounds on the variables of interest, including the shift in service and

manufacturing employees between 1988 and 1997. In this panel, a positive coefficient

indicates a positive change in log TRI pounds. The same variables are statistically

significant and point in the same direction as in panel two of Table 5.12, except that

median family income is not significant. Dirtiness index change and log population

density are statistically significant and positive, while log TRI pounds in 1988, median

family income squared, log percent poverty, log percent elderly, log in-migration, and the

service-manufacturing establishments decline dummy variable are each statistically

significant and negative. Once again, R2 is .09 for this model.

The results in Table 5.12 provide mixed support for the ecological modernization/

Environmental Kuznets Curve hypothesis (H5.1), and thus the EC hypothesis (H5.2);

negative findings regarding the EJ hypothesis (H5.3), with non-significant findings for

the race component, and opposite to expected findings for poverty and the elderly; mixed

results for the economic composition hypothesis (H5.4); and support for the population

density hypothesis (H5.5) and the in-migration hypothesis (H5.6).

In Table 5.13 panels one through three, change in log TRI pounds per facility

between 1988 and 1997 is regressed on the change in the dirtiness index score, TRI

facilities in 1988, economic change, economic standing, environmental justice, and other

factors of interest.

In panel one, results are displayed for the OLS regression of the change in log

TRI pounds per facility on the variables of interest, including the shift in service and

manufacturing establishments between 1988 and 1997. Six variables are statistically

significant and most conform to expectations in the literature and the hypotheses stated

above. Median family income is statistically significant and positive, while log TRI

pounds per facility in 1988, median family income squared, log percent elderly, log

inmigration, and the service-manufacturing establishments decline dummy variable are

each statistically significant and negative. R2 equals .18 for this model.

Median family income and median family income squared are significant yet in

the opposite direction, pointing again to a curvilinear relationship which conforms to the

EM/EKC hypothesis. Log percent elderly is significant yet in the opposite direction from

that expected (i.e. as each increases, change in the log TRI pounds per facility is

negative). Log in-migration is significant and negative. This suggests that localities

attracting in-migrants reduced their per-plant toxic emissions over the period. Also

significant and negative is the log TRI pounds per facility in 1988. This indicates that, net

of other factors in the model, those counties with more log TRI pounds per facility in

1988 reduced their output per plant between 1988 and 1997. Finally, the

servicemanufacturing establishments decline dummy variable is significant and negative,

which means that counties which experienced an overall decline in both manufacturing

and service establishments between 1988 and 1997 experienced a 1.014 unit decline in

log TRI pounds per facility during the same time period.

In panel two of Table 5.13, results are displayed for the OLS regression of the

change in log TRI pounds per facility on the variables of interest, including the shift in

service and manufacturing wages between 1988 and 1997. Five variables are statistically

significant and several conform to expectations in the literature and the hypotheses stated

above. Median family income is statistically significant and positive, while log TRI

pounds per facility in 1988, median family income squared, log percent elderly, and log

in-migration are statistically significant and negative. Substantive interpretation for the

significant findings is the same here as is described for Table 5.13 panel one. R2 is .18 for

this model.

In Table 5.13, panel three, results are displayed for the OLS regression of the

change in log TRI pounds per facility on the variables of interest, including the shift in

service and manufacturing employees between 1988 and 1997. The same variables are

statistically significant and point in the same direction as in panel two of Table 5.13.

Median family income is statistically significant and positive, while log TRI pounds per

facility in 1988, median family income squared, log percent elderly, and log in-migration

are statistically significant and negative. Substantive interpretation for the significant

findings is again the same as is described for Table 5.13 panel one. R2 is again .18 for this

model.

The results in Table 5.13 provide support for the ecological modernization/

Environmental Kuznets Curve hypothesis (H5.1), and thus are opposite to expectations

for the EC hypothesis (H5.2); relatively weak findings regarding the EJ hypothesis

(H5.3), with non-significant findings for the race and poverty components, and opposite

to expected findings for the elderly; mixed results for the economic composition

hypothesis (H5.4); non-significant findings for the population density hypothesis (H5.5);

and support for the in-migration hypothesis (H5.6).

Table 5.13

Change in Log TRI Pounds per Facility Regressed on Economic Change, Manufacturing/

Service Shift, Dirtiness Index Change, Log TRI Pounds 1988, Economic Standing,

Environmental Justice, and Other Factors, 1988-1997

Model 1

Model 2

Model 3

Dirtiness Index Change19881997

0.033

0.041

-0.008

Log TRI Pounds per Facility

1988

-0.371***

-0.370***

Median Family Income

0.000*

Median Family Income Squared

-0.000**

Log Percent Poverty

-0.104

-0.119

-0.133

Log Percent Elderly

-0.543**

-0.549**

-0.551**

Log Percent Black

0.032

0.036

0.037

Log Black-White Median Income

Ratio

-0.247

-0.248

-0.260

Black-White Dissimilarity Index

0.101

0.105

0.098

Log Population Density

0.031

0.037

0.038

-0.266**

-0.265**

-0.268**

Economic Change

Service-Manufacturing

Establishments Manufacturing

Growth Dummy

0.306

Service-Manufacturing

Establishments Service Growth

Dummy

0.024

---

Service-Manufacturing

Establishments Decline Dummy

-1.014**

---

Service-Manufacturing Wages

Manufacturing Growth Dummy

---

0.005

---

Service-Manufacturing Wages Service

Growth Dummy

---

0.098

---

Service-Manufacturing Wages Decline

Dummy

---

0.052

---

Service-Manufacturing

Employees Manufacturing

Growth Dummy

---

-0.272

Service-Manufacturing

Employees Service Growth

Dummy

---

-0.078

Service-Manufacturing

Employees Decline Dummy

---

-0.854

Constant

6.067***

6.071***

6.212***

0.182

0.178

0.180

Note: * p < 0.05; ** p < .01; *** p< .001 (two-tailed tests).

Turning to Table 5.14, panels one through three, change in the log median risk

score between 1988 and 1997 is regressed on the change in the dirtiness index score, TRI

facilities in 1988, economic change, economic standing, environmental justice, and other

factors of interest. Economic change is again measured here with three dummy variables

each representing the change in service-manufacturing establishments, wages, and

employment.

In panel one, results are displayed for the OLS regression of the change in the log

median risk score on the variables of interest, including the shift in service and

manufacturing establishments between 1988 and 1997. In this panel, a positive

coefficient indicates a positive change in the log median risk score. Eight variables are

statistically significant and most conform to expectations in the literature and the

hypotheses stated above. Dirtiness index change, median family income, and log

population density are statistically significant and positive, while the log median risk

score in 1988, median family income squared, log percent poverty, log percent elderly,

and the service-manufacturing establishments decline dummy variable are each

statistically significant and negative. R2 is .10 for this model.

Change in the dirtiness index between 1988 and 1997 is significant and positive,

which means that as positive change in the dirtiness index occurs, positive change in the

log median risk score also happens. This finding is interesting, in that it points to a

relationship between proportionality and toxicity: those counties where a higher

proportion of total manufacturing facilities were part of the TRI were ones in which the

toxic exposure risk increased for residents of the county, net of other factors. Also

significant and negative is the log median risk score in 1988, which means that as the log

median risk score in 1988 increases, the change in the log median risk score between

1988 and 1997 is negative. In combination with the previous finding, these two together

indicate that counties that were increasing their proportional profile of TRI facilities were

becoming more toxic, while those which were highly toxic in the 1980s were generally

shifting to less toxic industries in the 1990s. This again is indicative of an exportation of

risk, which will be addressed explicitly in Chapter Six.

Median family income and median family income squared are significant but in

the opposite direction, pointing again to a curvilinear relationship which conforms to the

EM/EKC hypothesis. Log percent poverty and log percent elderly are significant but in

the opposite direction from that expected (i.e. as each increases, change in the log median

risk score is negative). Log population density is significant and positive, indicating that

as log population density increases, positive change occurs in the log median risk score.

Finally, the service-manufacturing establishments decline dummy variable is significant

and negative, which means that counties which experienced an overall decline in both

manufacturing and service establishments between 1988 and 1997 experienced a 5.488

unit decline in the log median risk score during the same time period.

In panel two of Table 5.14, results are displayed for the OLS regression of the

change in the log median risk score on the variables of interest, including the shift in

service and manufacturing wages between 1988 and 1997. Seven variables are

statistically significant and several conform to expectations in the literature and the

hypotheses stated above. Dirtiness index change, median family income, and log

population density are statistically significant and positive, while the log median risk

score in 1988, median family income squared, log percent poverty, and log percent

elderly are each statistically significant and negative. R2 is .10 for this model.

Table 5.14

Change in Log Median Risk Regressed on Economic Change, Manufacturing/Service

Shift, Dirtiness Index Change, TRI Pounds 1988, Economic Standing,

Environmental Justice, and Other Factors, 1988-1997

Model 1

Model 2

Model 3

Dirtiness Index Change 19881997

10.952***

10.984***

11.002***

Log Median Risk Score 1988

-0.390***

-0.386***

Median Family Income

0.000*

Median Family Income Squared

-0.000**

-0.000***

Log Percent Poverty

-1.345*

-1.351*

-1.320*

Log Percent Elderly

-1.941**

-1.973**

-1.959**

Log Percent Black

0.247

0.263

0.272

Log Black-White Median Income

Ratio

-0.014

0.044

0.014

Black-White Dissimilarity Index

1.205

1.258

1.278

Log Population Density

0.387**

0.411**

0.424**

-0.448

-0.455

-0.454

Economic Change

Service-Manufacturing

Establishments Manufacturing

Growth Dummy

-1.140

Service-Manufacturing

Establishments Service Growth

Dummy

-0.119

---

Service-Manufacturing

Establishments Decline Dummy

-5.488***

---

Service-Manufacturing Wages

Manufacturing Growth Dummy

---

0.044

---

Service-Manufacturing Wages Service

Growth Dummy

---

-0.028

---

Service-Manufacturing Wages Decline

Dummy

---

0.883

---

Service-Manufacturing

Employees Manufacturing

Growth Dummy

---

1.116

Service-Manufacturing

Employees Service Growth

Dummy

---

-0.215

Service-Manufacturing

Employees Decline Dummy

---

1.591

Constant

14.612***

14.052***

13.942***

0.104

0.095

Note: * p < 0.05; ** p < .01; *** p< .001 (two-tailed tests).

In Table 5.14, panel three, results are displayed for the OLS regression of the

change in the log median risk score on the variables of interest, including the shift in

service and manufacturing employees between 1988 and 1997. The same variables are

statistically significant and point in the same direction as in panel two of Table 5.14.

Dirtiness index change, median family income, and log population density are

statistically significant and positive, while the log median risk score in 1988, median

family income squared, log percent poverty, and log percent elderly are each statistically

significant and negative. Once again, R2 is .10 for this model.

The results in Table 5.14 provide support for the ecological modernization/

Environmental Kuznets Curve hypothesis (H5.1), and thus opposite to expected findings

for the EC hypothesis (H5.2); mixed findings regarding the EJ hypothesis (H5.3), with

non-significant findings for the race component, and opposite to expected findings for

poverty and the elderly; mixed results for the economic composition hypothesis (H5.4);

support for the population density hypothesis (H5.5); and non-significant findings for the

in-migration hypothesis (H5.6).

As was noted with the previous section, there have been clearly mixed results

observed here. There does again appear to be more support for the EM/EKC hypothesis

(H5.1) than the EC hypothesis (H5.2), though the findings are clearly mixed, and in some

cases point to a different result than expected in either hypothesis (declines in

environmental degradation as income goes up in a linear fashion). The implications of

this result will be explored in further detail in Chapter Six. Findings related to the

environmental justice hypotheses (H5.3) were weaker than in the previous section, as

percent elderly and percent poverty were typically significant and in the opposite to

expected direction, and racial findings were pretty much non-existent. There were few

significant findings of note for the economic composition or economic change

hypotheses (H5.4). Once again though, population density (H5.5) and level of

inmigration (H5.6) were each typically significant and in the expected direction (positive

for population density and negative for in-migration). In the next section, the findings

will be summarized in tabular form and discussed, particularly as they relate to Chapter

Six, the final analysis chapter.

Conclusion

To further accomplish the goal of advancing the “jobs versus the environment”

debate, there are two primary questions which were addressed in this chapter: 1) is

pollution part of the cost of doing business associated with job or wage growth in U.S.

communities; and 2) has the “great slide” or decline of manufacturing affected local

pollution levels? Table 5.15 summarizes the results of the hypothesis testing performed in

this chapter. To restate, the hypotheses were as follows:

Hypothesis 5.1. Drawing from the Ecological Modernization and Environmental

Kuznets Curve perspectives (EM/EKC), counties which are more ―economically

advanced‖ or which experience economic growth over time will be expected to

experience growth in pollution output up to a point (the form of the relationship

will be curvilinear, first positive and then negative as the ecological state emerges

or a shift from manufacturing to services occurs), net of other factors.

Hypothesis 5.2. Drawing from the Environmental Conflict perspective (EC),

growth in pollution output will be expected in counties with greater economic

growth in manufacturing, net of other factors.

Hypothesis 5.3. Drawing from the Environmental Justice perspective (EJ),

counties with higher percentages of social minorities will be expected to

experience growth in pollution output regardless of level of economic

―advancement,‖ net of other factors.

Hypothesis 5.4. Drawing from theories of industrial concentration, counties with

positive growth in manufacturing activity or relative growth in manufacturing

versus services will be expected to experience growth in pollution output, net of

other factors.

Hypothesis 5.5. Drawing from demographic theories, growth in pollution output

will be expected in counties as the population density increases.

Hypothesis 5.6. Drawing from quality of life theories, declines in pollution output

will be expected in counties as the level of in-migration increases.

Taken as a whole, there appears to be somewhat mixed support for the hypotheses tested

in this chapter. First, the hypotheses drawn from the EM/EKC and EC perspectives can

be discussed in tandem, since they each utilize the same measure and are generally set in

opposition to each other. As was seen in the chapter, both over the individual years (1988

and 1997), and in terms of the change over time, the EM/EKC hypothesis enjoyed

stronger support than the EC hypothesis. More often than not, environmental quality first

declined, and then improved as economic growth occurred. This inverted u-shaped curve

is identical to Kuznets‟ classic income inequality curve, and also meets the expectations

of the ecological modernization camp that as economic growth reaches a certain point,

environmental improvements will occur. However, as impressive as these results were,

there was also another set of findings which indicated that environmental quality

improved as median family income increased from the very beginning. This finding is

indicative of overall environmental improvement for areas with higher median incomes,

pointing perhaps to the exportation of highly polluting facilities to other poorer locales

either internationally or within the United States. This important issue will be examined

in Chapter Six.

Second, findings for the environmental justice hypothesis are inconsistent.

Generally, the racial variables met expectations; i.e. as percent black, black-white income

inequality, or the dissimilarity index grew, counties faced higher levels of environmental

pollution. This finding is consistent with the work of Bullard, Mohai and Bryant, and

others in which environmental racism is found to be operating in American

neighborhoods and communities. This is particularly significant during this time period,

as the United States went through tremendous economic growth, and yet even during

these years race-based environmental inequality persisted. The question can legitimately

be raised “if not then, when can we get past environmental inequality?” This question

would be one posed by both environmental justice researchers and also those who focus

on issues of power and social inequality more generally, such as James Boyce. For

poverty and the elderly, when significant the findings ran counter to expectations and

point to improved conditions as these two groups grow proportionally in the population.

There are reasonable explanations for each of these contradictory findings. For the

elderly, higher proportions may be related to retirement or migration to lower polluting

areas. In terms of poverty, this could be somewhat mitigated by having more

manufacturing jobs, which are often higher paying, relatively speaking, yet which are

also generally linked to higher pollution levels. Each of these variables will be included

in the final analysis chapter to see if these somewhat inconsistent results continue.

Table 5.15

Summary of Hypotheses and Results in Chapter Five

HYPOTHESIS (5.x)

TRI

Facilities

Log TRI

Pounds

Log TRI

Pounds

per

Facility

Log

Median

Risk

Dirtiness

Index

EM/EKC (1)

EC (2)

EJ (3)

Poverty –

Elderly –

Black –

B-W Income Ratio –

Dissimilarity Index –

+ N.S.

* N.S.

N.S.

Population Density (5) In-

migration (6)

+ *

HYPOTHESIS (5.x)

Change in

TRI

Facilities

Change in

Log TRI

Pounds

Change in

Log TRI

Pounds

per

Facility

Change in

Log

Median

Risk

Change

Dirtiness

Index

EM/EKC (1)

EC (2)

EJ (3)

Poverty –

Elderly –

Black –

B-W Income Ratio –

Dissimilarity Index –

N.S.

N.S. N.S.

N.S.

- *

N.S.

Economic Composition and

Change (4)

Population Density (5) In-

migration (6)

+ *

N.S.

Note: N.S. (Not significant); + Significant and in expected direction; - Significant yet in opposite

direction; *Mixed/inconsistent.

Third, findings related to economic composition and economic change are largely

non-significant, and when they are significant, are generally inconclusive. Given that this

chapter is largely designed to assess how economic changes (generally during this period

characterized either as de-industrialization combined with service growth, or as overall

economic decline), this is certainly disappointing. However, when taken in tandem with

the mixed results regarding the EM/EKC hypothesis and the generally consistent

relationship between the Dirtiness Index and negative environmental conditions, the

findings do point towards the possibility that there is a certain category of counties or

sub-regions which are the recipients of highly polluting facilities. Again, this issue is at

the center of the analyses in Chapter Six.

Regarding the population density hypothesis, it was generally found to be

significant and pointed in the expected direction; i.e. as population density increased,

negative environmental conditions were more likely to be experienced. In Chapter Four,

it was observed that metropolitan areas (which generally have greater population density)

were most likely to have declines in negative environmental conditions between 1988

and 1997; however, the earlier analyses were bivariate, and more importantly, these areas

often began the period of this study (1988) with significantly higher levels of pollution

and negative environmental conditions.

Finally, level of in-migration was significant in almost every model and was

pointed in the expected direction each time; i.e. as in-migration increased, negative

environmental conditions were less likely to be experienced. This is expected because

inmigrants generally seek a higher quality of life whenever possible, and one significant

indicator of local quality of life is the condition of the natural environment (Jones et al.

2003).

In sum, it seems that the answers to the two questions which frame this chapter

are as follows: to a certain extent, pollution is part of the cost of doing business

associated with job or wage growth in U.S. communities, especially where socially

disadvantaged groups reside; and 2) the “great slide” or decline of manufacturing has

affected local pollution levels somewhat, though not always in the expected direction,

particularly when communities cling to whatever industrial base remains for economic

survival, regardless of its environmental quality record. It is to these areas and regions

where we next turn. In Chapter Six, the final analysis chapter, the Domestic Pollution

Havens concept is introduced, developed and tested. This concept focuses on locales

which could be characterized as “the worst of the worst” in relation to both

environmental quality and economic rewards. Thus the Domestic Pollution Havens

concept will be utilized as a way to move past the “jobs versus the environment” rhetoric,

since these areas will be conceptualized as having little success in terms of either jobs (or

economic success more generally) or environmental quality.

CHAPTER VI

SPATIAL LOCATION AND GOVERNMENT CONTROLS: ARE THERE

DOMESTIC POLLUTION HAVENS IN THE UNITED STATES?

Introduction: Integrating Economic and Environmental Components within the

Domestic Pollution Havens concept

Results in the previous two chapters pointed out the need to examine in more

depth locales which could be characterized as “the worst of the worst” in relation to both

environmental quality and economic rewards. In this chapter, the Domestic Pollution

Havens concept is utilized as an integrative concept in an empirically-driven effort

designed to move beyond the rhetorical „environment or economy‟ duality so often

posited by politicians, environmentalists and many academics. As described in Chapter

Three, Domestic Pollution Havens have higher levels of environmental pollution per unit

of economic reward and also lower levels of social controls, specifically environmental

regulations and enforcement actions. It is argued here that the existence of a Domestic

Pollution Haven is influenced by several factors, including local political culture,

sociopolitical organization, dirty industry concentration, socio-demographic

characteristics, and other variables. These hypothesized relationships will be formally

described in this section, and then tested empirically within several different regression

models later in the chapter.

The two predominant approaches in the environmental social sciences

(Environmental Kuznets Curve/Ecological Modernization and Environmental Conflict

theories), which were tested in the previous chapter, have informed the construction of

the hypotheses in this section where appropriate. However, both approaches are centrally

focused on the manner in which economic growth either positively or negatively

influences environmental quality, not the intersection of economic growth and

environmental quality as something to be predicted in its own right. Thus there are not

explicit hypotheses derived from those two perspectives regarding the existence of

Domestic Pollution Havens. The hypotheses to be tested in this chapter include:

Hypothesis 6.1. Drawing from the Power/Inequality perspective, counties will be

more likely to be characterized as Domestic Pollution Havens as the degree of

power imbalance between social groups increases. The variable used to assess the

degree of county-level power imbalance is the log black-white median income

ratio, since it captures the degree to which one central form of power, economic

power, is equally or unequally distributed within a county.

Hypothesis 6.2. Also drawing from the Power/Inequality perspective, counties

will be less likely to be characterized as Domestic Pollution Havens as the level

of involvement in the local political culture increases, net of other factors. The

variables used to measure the local political culture include voter turnout, League

of Conservation Voters state-level congressional delegation score, and the

presence of environmental 501c3 organizations in the county. Each of these assess

different aspects of the political process, from formal electoral participation

through legislative agendas and to social movement organization and mobilization

within a locality.

Hypothesis 6.3. Drawing from the Environmental Justice perspective, counties

with higher percentages of social minorities will be more likely to be

characterized as Domestic Pollution Havens, net of other factors. The variables

used to assess environmental justice include county-level log percent in poverty,

log percent elderly, log percent African-American, and the Black-White

dissimilarity index. As was seen in the previous chapters, these are commonly

used measures of environmental justice.

Hypothesis 6.4. Drawing from theories of industrial concentration and social

equity, counties will be more likely to be characterized as Domestic Pollution

Havens as the average dirtiness index score increases, net of other factors. The

variable used to assess the effect of county-level dirty industry concentration is

the average dirtiness index score. This measure examines the disproportionality of

“toxic” manufacturing industries across counties (see for example Freudenburg

2005).

Hypothesis 6.5. Drawing from quality of life theories, counties will be less likely

to be characterized as Domestic Pollution Havens as the level of in-migration

increases, net of other factors. The variable used to assess the overall community

quality of life is the log percentage of county in-migration. In-migrants are drawn

to amenities; in recent years higher environmental quality has become a

significant draw for many people locating in new communities (Jones et al. 2003).

To test these hypotheses, several regression analyses are utilized. First, the Domestic

Pollution Havens concept itself is explored in two different regressions where: 1) DPH is

dichotomized as “DPH/not DPH” based on a summing of the nine characterizations and

tested utilizing logistic regression; and 2) DPH is treated as a count variable based on a

summing of the nine characterizations (score range = 0-9) and tested utilizing negative

binomial regression. The final three regression analyses utilize separate elements of the

Domestic Pollution Havens concept, including Log Pounds of TRI Releases per

Manufacturing Job, Log Pounds of TRI Releases per Average Manufacturing Wage, and

Log Median Risk per Average Manufacturing Wage. Spatial Ordinary Least Squares

regression is used on these pollution output intensity variables.

In the next two sections, the Domestic Pollution Havens concept is first discussed

theoretically and then developed through descriptive statistics and spatial visualization.

In the subsequent sections, evidence is provided to test the hypotheses 6.1-6.5. The

chapter is concluded with a summary of findings and results of the hypothesis testing.

Understanding Domestic Pollution Havens

The Domestic Pollution Havens concept is proposed to serve as an integrative

theoretical and empirical concept that brings economic and ecological issues into a

common framework. The pollution havens concept is drawn from the ecological

economics literature, and provides researchers with a new conceptual tool to get at the

economic growth and environmental quality duality in a more comprehensive way than

has been done to date.

The underlying premise of a pollution haven is that, in attempting to improve

their economic condition, relatively disadvantaged areas may seek out dirty industries.

These areas are believed to have a comparative advantage in producing “dirty goods”

because they generally impose less regulation and because their workers require lower

wages to take on generally low-skill tasks. In the U.S. and other advanced industrial

nations, this would mean an increase in the importation of “dirty” manufacturing goods

from poorer

nations, while the internal production of such goods would decline.

However, the evidence on the presence of international pollution havens has been

relatively unimpressive to this point, with mixed findings, inconsistent

operationalization, data problems, and other issues plaguing the research (Kahn 2004).

The mixed findings in particular have led some to argue that there is something else at

play.

Ecological economist Matthew Kahn (2004) has argued that Domestic Pollution

Havens may substitute for international pollution havens. He notes that the U.S. has not

increased its imports of “dirty” manufacturing goods from poorer nations. Thus what he

believes is going on is the relocation of “dirty” production activity to areas where

environmental, safety and labor regulations are less stringent or relatively non-existent.

In his research, Kahn has found some evidence of Domestic Pollution Havens,

particularly at certain border areas, and at the boundaries of high and low regulation

states. Thus this concept offers some exciting promise, particularly as it explores the

interplay of economic conditions and environmental quality.

However, Kahn‟s (2004) work leaps to an outcome---mortality---that skips

centrally important production outcomes: amounts and toxicity of pollution outputs and

their risks to human populations. While a very interesting and insightful concept, the

Domestic Pollution Havens concept, as currently applied, suffers from at least three

problems:

1) Domestic Pollution Havens have been operationalized in a very limited

manner. Kahn uses environmentally-linked cancer rates as his dependent variable

representing the presence of a Domestic Pollution Haven. A broader-based measure that

captures (relatively) high levels of environmental pollution and also (relatively) low

levels of regulation is called for. This is what is proposed and tested in this study.

2) In addition to issues of regulation, what about other social and political

factors which may facilitate or inhibit the growth of Domestic Pollution Havens? The

demographic makeup of the community, particularly reflecting the distribution of groups

that face patterns of discrimination such as African Americans, the poor, or the elderly,

may serve to increase the likelihood of “dirty” industries locating in a county and also the

likelihood that regulatory controls will be put into place and implemented within that

area. The presence of environmental organizations would be expected to minimize “bad”

environmental conditions such as the emergence of a Domestic Pollution Haven. Local

political culture would also be expected to play an important role, as a more engaged

citizenry or a community with more environmentally-friendly politicians should be able

to work to minimize the likelihood of being in a Domestic Pollution Haven.

3) The previous work on Domestic Pollution Havens has utilized largely a-

spatial techniques, something which must be overcome, as we are increasingly

recognizing the importance of localities and spatial dependence in current theorizing and

research. Thus spatial visualization and spatial statistical techniques are needed to assess

the manner in which “place matters” for this very space-bounded concept.

As conceived here, a Domestic Pollution Haven is a county in which there is a

high level of pollution, a low level of economic rewards, and low regulatory controls.

However, since the link between economic and environmental conditions is so critical,

this relationship must be further clarified. Counties with high levels of pollution or

toxicity yet high aggregate levels of economic rewards are not characterized as Domestic

Pollution Havens. This follows the logic of those who argue that the environmental

problems are just the “cost of doing business.” Further, counties where there is a high

level of pollution or toxicity and a low level of economic rewards yet also have high

levels of regulatory controls are not considered Domestic Pollution Havens. Thus,

Domestic Pollution Havens are conceptualized as counties with high levels of pollution

per unit of economic reward, high toxicity per unit of economic reward, and low

regulatory controls.

An additional factor of importance involves the impact of local political culture as

an inhibiting or facilitating factor in the presence of Domestic Pollution Havens. Those

counties located in states which are relatively environmentally friendly, and in which the

state‟s congressional delegation has a relatively strong environmental record, are likely

to have lower levels of federal regulatory actions, an important component of a Domestic

Pollution Haven. It is expected that locales that have relatively “green” policies will elect

environmentally friendly politicians, who in turn will further assist the enhancement of

state-level “green” policies, and vice versa. Each of these aspects of local political culture

are expected to impact the level of federal regulatory activity, where pressure can be put

to bear on federal agencies such as the EPA to either aggressively enforce or generally

overlook environmental conditions, depending on the condition of local political culture

regarding the environment. In the next section, the construction of the Domestic Pollution

Havens concept is described.

Descriptive Characterization and Spatial Visualization of Domestic Pollution

Havens

In this section, the Domestic Pollution Havens concept is examined through the

use of descriptive statistics and spatial visualization. Given that this concept involves an

explicit attempt to empirically address the environment-economy ideological duality, it is

imperative that multiple indicators are utilized. Specifically, indicators are presented in

this section which integrate economic and environmental characteristics, as well as those

which reflect local social and political conditions. These indicators include: Log Pounds

of TRI Releases per Manufacturing Job, Log Pounds of TRI Releases per Average

Manufacturing Wage, Log Median Risk per Average Manufacturing Wage, EPA

DOCKET Enforcement Actions, EPA Enforcement Litigation, and the Green Policies

Environmental Policies score. The first three indicators reflect what is characterized by

economists as pollution output intensity (POI) measures (Hettige, Martin, Singh and

Wheeler 1995). The POI measures are considered to be representative of the tradeoff

between pollution levels and economic growth. In Chapter 3, the DPH operationalization

process was described in detail; Table 6.1 below summarizes this process.

Figure A.29 is a visualization of POI Pounds per Job, which is measured as the

mean of TRI Pounds per Manufacturing Job, 1988-1997. As was the case with the figures

presented in Chapter Four, the figures referred to in this chapter can be found in the

appendix. The counties shaded in dark red in Figure A.29 are those areas in which the

average TRI Pounds of releases per Manufacturing job is high, meaning that the

economic rewards per unit of pollution are low, and thus the putative “reward” in

manufacturing jobs per pound of toxic pollution is smaller for those counties. These

DPH-characterized counties are those in which the average TRI pounds of releases per

manufacturing job are above the 75th percentile, or 245.7988 pounds per job.

Table 6.1

Domestic Pollution Havens Operationalization Process

Used a classification of pollution-per-economic reward in

manufacturing by the level of social controls. These indicators include:

TRI Pounds per Job, TRI Pounds per Average Manufacturing Wage,

Median Risk per Average Manufacturing Wage <these three represent

pollution-per-economic reward>; EPA Docket Enforcement Actions,

EPA Enforcement Litigation, and Toughness of State

Environmental Regulations <these three represent social controls>.

The six potential indicators of Domestic Pollution Havens (DPHs)

were used in nine cross-tabulations and examined to see which

counties fit within both parameters of what is theoretically expected to

be a DPH. The results of each cross-tabulation were recoded 1 if

“high” (i.e. both meeting DPH conceptualization) or 0 if “medium” or

“low.”

The results of these nine classifications were summed to obtain the

number of DPH conditions met in each county.

This sum, ranging from 0-9, was then subjected to an Exploratory

Spatial Data Analysis using LISA with permutation tests for the

presence of significant spatial clusters of contiguous counties

representing potential DPH sub-regions.

Visual inspection of Figure A.29 indicates that there are patterns of spatial

dependence throughout different areas of the country, including many sections of the

Southeast (including the Gulf Coast region of Louisiana and Texas, southern and central

sections of Alabama and Mississippi, and much of South Carolina), the Rust Belt regions

of the Northeast and Midwest (western Pennsylvania, Ohio, Indiana and Michigan

especially), and parts of the West. Exploratory Spatial Data Analysis (ESDA) techniques

will be presented later in the chapter to determine whether the characterization of these

patterns based on the author‟s visual inspection is accurate.

Figure A.30 is a visualization of POI Pounds per Average Wage, measured as the

mean of TRI Pounds per Average Manufacturing Wage, 1988-1997. The counties shaded

in dark red are those areas in which the average TRI pounds per average manufacturing

wage is high, meaning that the economic rewards per unit of pollution are low, and thus

this is a characteristic of a Domestic Pollution Haven. These DPH-characterized counties

are those in which the average TRI pounds per average manufacturing wage are above

the 75th percentile, or 42.1569 pounds per hundred dollars of wage. Visual inspection of

Figure A.30 indicates that there are similar patterns of spatial dependence throughout

different areas of the country as was observed in Figure A.29, including many sections of

the Southeast (including the Gulf Coast region of Louisiana and Texas, southern and

central sections of Alabama and Mississippi, and much of South Carolina), the Rust Belt

regions of the Northeast and Midwest (western Pennsylvania, Ohio, Indiana and

Michigan especially), and parts of the West. However, additional clusters of high counties

are observed around in the megalopolis counties of the Northeast, including the areas

around Boston, New York, Philadelphia, and Baltimore, as well as most of the counties in

Southern California (the Los Angeles and San Diego areas).

Figure A.31 is a visualization of POI Risk by Wage, measured as the mean of

Median Facility Risk per Average Manufacturing Wage, 1988-1997. The counties shaded

in dark red are those areas in which the mean of median facility risk per average

manufacturing wage is high, meaning that the economic rewards per unit of pollution are

low, and thus this is a characteristic of a Domestic Pollution Haven. These

DPHcharacterized counties are those in which the mean of median facility risk per

average manufacturing wage is above the 75th percentile, or 21,736,149 risk score per

hundred dollars of wage. Visual inspection of Figure A.31 indicates that there are again

similar patterns of spatial dependence throughout different areas of the country as was

observed in Figures A.29 and A.30, including many sections of the Southeast (including

the Gulf Coast region of Louisiana and Texas, southern and central sections of Alabama

and

Mississippi, and much of South Carolina), the Rust Belt regions of the Northeast and

Midwest (western Pennsylvania, Ohio, Indiana and Michigan especially), and parts of the

West. However, the clusters of high counties observed in Figure A.30 located in the more

urban centers of the Northeast and Southwest are not apparent here.

Shifting to the social control variables, Figure A.32 is a visualization of EPA

DOCKET Enforcement Actions, measured as the number of Docket actions through Year

2000. This is indicative of the number of federal enforcement actions taken in the county

during the period of modern federal environmental protection legislation. These actions

can result in a variety of types of outcomes and penalty settlements, ranging from

litigation with or without penalties to administrative action to the establishment of a

consent instrument, to withdrawal or dismissal. The white indicates those areas in which

the number of EPA DOCKET enforcement actions is low, meaning that the regulatory

controls are low, and thus this is a characteristic of a Domestic Pollution Haven. These

DPH-characterized counties are those in which the EPA DOCKET enforcement actions

are below the 50th percentile (2 or fewer cases). Visual inspection of Figure A.32

indicates noticeable differences in the patterns of spatial dependence observed for the

POI variables. DPH-characterized counties appear to cluster in most of the Southeast,

except for the Gulf Coast counties of Texas and Louisiana and around certain

metropolitan areas in the region. Additionally, low numbers of cases appear in most of

the Great Plains region of the Midwest; which is perhaps not surprising since, unlike in

much of the Southeast, the POI indicators were very low there as well. In the Northeast,

the upper Midwest, and much of the West, there were relatively high numbers of cases

filed, though often this number is in the single digits over a three decade period.

Figure A.33 is a visualization of the EPA Enforcement Litigation, which is

measured as the percent of all DOCKET enforcement actions resulting in court litigation.

This differs from the previous indicator in which a county-level count of DOCKET

actions was taken, regardless of outcome. Litigation represents the most serious form of

action that can be undertaken by the EPA, and in Figure A.33, those enforcement actions

which end in litigation, regardless of outcome, are shown. The white indicates those areas

where the percentage of DOCKET enforcement actions resulting in court litigation is

low, meaning that the regulatory controls are low, and thus this is a characteristic of a

Domestic Pollution Haven. These DPH-characterized counties are those in which the

percentage of DOCKET enforcement actions resulting in court litigation is below the 50th

percentile (0% of cases). Visual inspection of Figure A.33 overall indicates similar

patterns to the previous figure, though with one important difference. There are some

clusters in the Southeast where the POI variables were high, and the number of EPA

enforcement actions was low, but the percentage of cases which were adjudicated in court

was high (particularly in much of Alabama, South Carolina, and southern Louisiana).

However, it is worth noting that many of these locales are places where one or two

actions were filed in a thirty year period, with that action resulting in a court case. This

does not necessarily imply stringent social control, particularly if the case was egregious.

Again, this is the reason that multiple indicators are utilized to construct the Domestic

Pollution Haven variable.

The final DPH indicator map (Figure A.34) is a visualization of Toughness of

State Environmental Regulations, measured as the Green Policies Environmental Policies

score for each state. Here, Green indicates relatively environmentally friendly state

policies, and Brown indicates the reverse. The counties located in brown states thus have

low regulatory controls, and thus this is a characteristic of a Domestic Pollution Haven.

These DPH-characterized counties are those in which the Green Policies Environmental

Policies score falls below the 25th percentile (below a score of 1865). Visual inspection of

Figure A.34 overall indicates a very different pattern, due to the state-level aggregation.

However it is apparent that the states where environmental regulations are most lax fall in

the Deep South (including Alabama, Mississippi, Tennessee and Arkansas), and most of

the Mountain West, while the states with the most stringent regulations are generally in

the Northeast, the upper Midwest, and the west coast.

Next, these six indicators are combined to create the Domestic Pollution Havens

indicator variable. This was done by performing six cross-tabulations of the economic

reward and environmental regulation variables, scoring a “yes” if the county has been

characterized as a Domestic Pollution Haven for both indicators. Figure A.35 presents a

characterization of the number of Domestic Pollution Haven “yes‟s” from these

crosstabulations. “No DPH” represents 0 characterizations, while “Some DPH”

represents 1-4 characterizations, and “High DPH” represents either 6 or 9

characterizations. While it is apparent that most counties have little or no

characterizations, there does appear to be spatial dependence in that the counties

characterized as “High DPH” tend to fall primarily in parts of Alabama, Mississippi and

Tennessee. This analysis, while instructive, is based upon visual inspection. In the next

section, a LISA visualization will be examined, in a final effort to build the Domestic

Pollution Havens variable.

LISA Cluster of DPH Characterizations

Figure A.36 is the LISA cluster map of the Domestic Pollution Havens

characterization variable. In Chapter Three, the LISA statistic is described in detail. In

sum, the LISA statistic measures spatial autocorrelation, or the presence of local spatial

dependence in the Domestic Pollution Havens variable (see Anselin 2002). If there is

significant spatial autocorrelation (at the p < .05 level), it is either characterized as

High/High (you and your neighbor(s) have a high score on the variable of interest: in this

case, the variable of interest is DPH Level), Low/Low, Low/High, High/Low, or

statistically non-significant (Anselin 2003, 2005).

Examining Figure A.36 brings the picture into much sharper focus compared to

Figure A.35, as there is clear spatial autocorrelation in certain places. Of most substantive

interest here are High/High clusters, which are found in several pockets primarily in the

South and Midwest. The most noticeable High/High cluster is relatively large, centered in

Alabama and stretching through parts of Tennessee, Arkansas, and Mississippi. There are

some gaps in this cluster, but it does appear to hold together well, particularly in that it cuts

through both urban and rural counties. Other High/High pockets can be found in Oklahoma,

Illinois, Ohio, Kentucky, West Virginia, Maryland, and South Carolina, with some isolated

counties found in these states and also Georgia. On the other hand, there are three Low/Low

pockets observed, in relatively rural counties in Nebraska, New Hampshire and Georgia.

The High/High clusters are of most interest, obviously, in that they reflect the

counties which are characterized as Domestic Pollution Havens in the subsequent

regression analyses. The argument can be made, particularly in the largest DPH cluster,

anchored in the Deep South, that what is at work is what James Cobb (1993) called “the

Selling of the South,” a process whereby industrial development is emphasized at any

cost. Combined with relatively low levels of environmental concern, such development

could lead to pollution and contamination which would threaten the very physical

existence of these areas. Further, many of these DPH clusters have experienced little

“bang for the buck” in terms of economic growth that leads to higher wages or lowered

poverty rates. However, this is supposition at this point; in the next section, several

regression analyses are utilized to better assess the relationship between Domestic

Pollution Havens and local political culture, social organization, socio-demographic and

control variables, which are theorized to impact DPHs.

What Factors Influence the Presence of Domestic Pollution Havens?

In this section, the results of five regression analyses are presented. These

regression analyses were designed to assess the relationship between Domestic Pollution

Havens and measures of local political culture, environmental equity, the concentration

of “dirty” industries in the county, and other predictors.

Table 6.2 contains a logistic regression model of the Domestic Pollution Haven

dummy variable (1=Yes), as determined by High/High counties in the LISA cluster

described in the previous section. In this table, a positive coefficient indicates an

increased likelihood of being in a DPH county. Most of the findings conform to

expectations based on the previous literature. The dissimilarity index for blacks, the log

percent black, and the log average dirtiness index score are all significant and positive,

while the log in-migration rate, the average voter turnout percent, and the League of

Conservation Voters score are all significant and negative.

There are four key findings in Table 6.2. First, the log percent black and

dissimilarity index, both measures of the relationship between racial composition and

Domestic Pollution Haven status, are both significant and positive. Thus, as more African

Americans live in a county, it is more likely to be a Domestic Pollution Haven. The

coefficient for the log percent black, 0.434, indicates that a one percent increase in the

log percent black increases the likelihood of being located in a Domestic Pollution Haven

county by 54.3 percent (Exp (0.434)). The coefficient for the dissimilarity index, 1.584,

indicates that a one unit increase in the dissimilarity index (reflecting greater racial

imbalance) increases the likelihood of being located in a Domestic Pollution Haven

county by 387.4%. These are notable environmental justice findings consistent with prior

literature on the relationship between race and environmental quality (Been 1994;

Bullard

1990; Mohai and Bryant 1992; Stretsky and Lynch 2002; Taylor 2000).

Table 6.2

Logistic Regression of Domestic Pollution Havens on Potential Predictors in the

Contiguous United States

Domestic Pollution Haven (1=Yes)

Intercept 3.504*

Log Ratio Black-White Median Income 1990 0.386

Average Voter Turnout Percent

-0.038*

LCV score

-0.057***

Any Environmental 501c3s (1=Yes)

-0.282

Log Percent Poverty 1990

-0.469

Log Percent Elderly 1990

0.161

Log Percent Black 1990

0.434***

Dissimilarity Index for Blacks 1990

1.584**

Log Average Dirtiness Index Score

0.691***

-0.882***

0.23

-2 Log Likelihood

902.55

Log pseudo-likelihood

---

Wald

1215.10***

* p < 0.05; ** p < .01; *** p< .001 (two-tailed tests).

Second, the log average dirtiness index score is highly significant and positive.

Again, this is a measure of the proportion of manufacturing industries in a county which

report to the Toxic Release Inventory. When as the score increases, there is a higher

proportion of “dirty” industries in that county, and a greater likelihood of being in a

Domestic Pollution Haven. The coefficient for the log average dirtiness index score,

0.691, indicates that a one unit increase in the log average dirtiness index score increases

the likelihood of being located in a Domestic Pollution Haven county by 99.6 percent.

Third, the average voter turnout and League of Conservation Voters score are both

highly significant and negative. As either average voter turnout or the “proenvironment”

score of a state‟s Congressional delegation increases, the likelihood of being in a

Domestic Pollution Haven decreases. The coefficient for the average voter turnout, -

0.038, indicates that a one percent increase in average voter turnout decreases the

likelihood of being located in a Domestic Pollution Haven county by 3.7 percent. The

coefficient for the League of Conservation Voters score, -0.057, indicates that a one unit

increase in LCV score decreases the likelihood of being located in a Domestic Pollution

Haven county by 5.5 percent.

Finally, the log in-migration rate is highly significant and negative, meaning that

as more people move into a county, the likelihood of it being a Domestic Pollution Haven

decreases. The coefficient for log in-migration rate, -0.882, indicates that a one unit

increase in log in-migration decreases the likelihood of being located in a Domestic

Pollution Haven county by 58.6 percent.

Table 6.2 examines the relationship between the Domestic Pollution Havens

measure and several predictors. There is relatively strong support for the local political

culture hypothesis (H6.2), with the voter turnout and League of Conservation Voter score

variables being significant and in the expected direction; strong support for the racial

component of the environmental justice hypothesis (H6.3), as the race variables are

significant and in the expected direction; expected support for the dirty industry

concentration hypothesis (H6.4); and expected support for the quality of life hypothesis

(H6.5).

Students also viewed