Introduction

“Social Construction” is a term that has been used in different fields and has many definitions. For example, David Demeritt (2002) states that geographers use “social construction of nature” to “refute particular claims about the world and to make a variety of philosophical critiques of conventional understandings of nature and society” (Demeritt 2002, 786). He distinguished between two types of Social Construction: construction-as-refutation and construction-as-philosophical-critique. The first one refers to refutation of “false beliefs about the world and is consistent with …positivism and realism” (ibid.). The second one questions “the culture/nature, subject/object and representation/reality dualisms” that distinguish true conceptions of nature from false ones. The Sociology of Scientific Knowledge and Discursive Constructionisms is one of the constructions-as-philosophical-critiques. The first one explains “socially the once sacrosanct and epistemologically self-evident belief in scientifically valid knowledge and phenomena,” while the Discursive Constructionism refers to the “‘role of language in the construction of social reality’” (Demeritt 2002, 772 and 773). Both critiques are heavily used in sociology and other social sciences.

In the field of Public Policy, Schneider and Ingram’s Policy Design for Democracy uses these two critiques in a variety of ways, and particularly to question the impact of perception and political manipulation of perception on political behavior. Since their book’s publication in 1997 (and an earlier article published in 1993), Schneider and Ingram’s view has generated a considerable literature on the (political) social construction of target populations and whether such social constructions affect policy. In fact, as of 2014, there were over 100 works known to have used this framework (Sabatier and Weibel 2014: Table 4.1; Pierce et al. 2014).

However, many of the works in this literature focus on policy outcomes for a single target group, such as persons with AIDS or persons with mental health issues (e.g., Schroedel and Jordan 1998; McSween 2002). An overlooked implication of the Schneider and Ingram social construction (SC) model is that outcomes that affect all types of socially constructed groups—policies with widespread outcomes—should affect them in rank-ordered ways, such that benefits to the Advantaged should exceed benefits to Contenders, which should exceed benefits to Dependents, which should exceed benefits to Deviants. Similarly, if policies lead to undesirable social outcomes, we should see the opposite results, with the magnitude of undesirable outcomes ordered by social construction of target groups such that the Advantaged receive the least and the Deviant the most, with Contenders and Dependents in between. In this paper, we present results of an analysis that embraces this more complex, rank-order hypothesis about the way widespread, apparently universalistic, policy outcomes affect different socially constructed populations in a degenerative political system.

Air pollution has been studied from a social justice perspective, including by Liu (2001), and Pastor et al. (2001, 2004, 2005, in various author orders). In the environmental policy literature, there is also significant discourse on uneven distributional impacts of costs and benefits of pollution abatement, including concern about “hot spots” (e.g., Stavins 1998; Schwarze and Zapfel 2000). Air pollution affects, to greater and lesser degrees, all people in an area; thus, it affects people in all four quadrants of the social construction typology that Schneider and Ingram (1997) create. Here, we examine whether such a widespread policy outcome as actual levels of air pollution affects groups in the rank-ordered way implied by Schneider and Ingram’s (1997) social construction theory. Though not something identified by Schneider and Ingram (1997), analysis of air pollution incidence fits within the Schneider and Ingram concern with the outcomes of policy as distinct and perhaps quite often importantly different from the claims about public policy.

Air pollution outcomes are an interesting test since there are significant information asymmetries such that citizens are affected by air pollution regulations but may not be aware of actual air pollution incidence and further may assume that air pollution regulations work the same for all. Yet (as discussed below), Advantaged groups are much more likely to know about air pollution incidence and effects at the same time that, under the Schneider and Ingram (1997) model, the degenerative policy system will not wish to impose harms on these groups and will wish to provide them benefits. Thus, if the incidence of air pollution is influenced by social constructions as hypothesized, this supports the idea of policy makers as manipulating the policy system to punish and reward those in different social groups.

Theoretical grounding

In Policy Design for Democracy, Schneider and Ingram (1997) argue that the social construction of “target populations”—those populations that will be the targets of some policy or another—is an important political and policy phenomenon in a degenerative democratic system, for there is a multistep causal structure such that social construction and power shape policy designs, which in turn determine the benefits and harms target groups actually receive from policies. Further, the benefits and harms that target groups actually receive may be quite different from those stated as the goals of the policy—necessitating examining actual policy outcomes rather than legislation or policy statements alone.

Social construction theory uses a two-by-two taxonomy to divide target populations into four groups based on strength (political power) and deservingness: Advantaged, Contenders, Dependents, and Deviants. Advantaged are both politically strong and constructed as deserving; Contenders are politically strong but constructed as undeserving; Dependents are constructed as weak but deserving; and Deviants are constructed as weak and undeserving. Though the 2 × 2 taxonomy leads to four categories, it is also possible for groups to be “between categories” as society struggles over which social construction of many competing ones will dominate (Schneider and Ingram 1997; Pierce et al. 2014; and see several chapters in Schneider and Ingram 2005). Figure 1 below is based on Figure 5.2 in Schneider and Ingram (1997:109) and shows some, but not all, of the groups Schneider and Ingram present in that figure.

Fig. 1
figure 1

Sample social constructions. Note: this table is modified from Figure 5.2 in Schneider and Ingram (1997:109)

Schneider and Ingram (1997) argue that, within a degenerative democratic system, policy makers tend to shape policies based on these social constructions and how voters perceive each of these groups. For example, during elections, candidates tend to focus their campaigns on policies that satisfy Advantaged groups, such as business owners and the middle class, or design policies that put burdens on Deviant groups, such as criminals. The reason the candidates do this is because Advantaged groups are politically strong, and they are seen as deserving; thus, providing them benefits is accepted by the public and can provide gains for politicians since these groups are powerful (Schneider and Ingram 1997: 108). Deviant groups such as criminals are usually politically weak in the USA, and the public considers them undeserving. Thus, politicians seem to consider groups socially constructed as Deviants their easy target for policies that carry burdens (Schneider and Ingram 1997: 120). The groups in between are the Contenders and Dependents. These two groups are not readily favored by politicians because of the controversial debates that often surround them. Contenders tend to be politically strong, but they have bad reputations; Dependents are politically weak, but have good reputations, so they are considered to be deserving of and in need of help (Schneider and Ingram 1997: 116 and123). Putting burdens on Contenders and openly helping them are both politically risky; thus, benefits to these groups will be hidden (deceptive) and effective burdens rare (Schneider and Ingram 1997: 116). And, since Dependents are politically weak, and hence, little political capital is to be gained (within a degenerative democracy) from helping them, policies that appear to provide them benefits will tend to be hollow (also deceptive) (Schneider and Ingram 1997: 126). Schneider and Ingram (1997) are ultimately concerned with actual outcomes of policy rather than with what may appear to be, or are stated as, the effects of policy.

The social construction literature

As mentioned above, up to 2014 there were over 100 publications that used some form of Schneider and Ingram’s social construction theory (Sabatier and Weibel 2014: Table 4.1; Pierce et al. 2014). Pierce et al. (2014) provide an interesting and useful analysis of 123 of these works across several dimensions, reporting that (1) qualitative work predominates, with only 24 % purely quantitative studies and an additional 14 % using mixed methods; (2) 111 of the works apply the theory (as opposed to engaging in theory development), and 60 publications (or only 54 %) out of the 111 publications “explicitly identify at least one target population within one of the ideal categories of advantaged, contender, dependent, or deviant” (Pierce et al. 2014: 13); and (3) only 6 % are on the topic of the environment.

Further, often research using social construction theory focuses on outputs, rather than outcome. Differentiating between outputs and outcomes is important because outputs refer only to the “plans, projects, and other tangible items produced directly by the collaborative effort,” (Koontz and Thomas 2006). An example is legislation (e.g., Anglund 1998; Sidney 2005; Donovan 1993). However, outcomes were defined by Koontz and Thomas (2006) as the effects of the collaborative process and its outputs on changing social and environmental conditions. Though outputs are of course important (and indeed are steppingstones to outcomes), Schneider and Ingram are intimately concerned with outcomes and recognize that policies that appear to be universalistic may nonetheless have unequal outcomes:

One of the effects of this dynamic in sophisticated political institutions is that political leaders will take care not to appear to give special privileges but will emphasize purportedly universalistic application of rules that appear to treat everyone equally. The highly specific elaboration of presumably universalistic rules in legislation sometimes masks the myriad ways that the policy will produce outcomes mainly for advantaged groups (1997 p. 116; emphasis added)

Within this context, we note that our article is quantitative, applies the theory using seven explicitly identified target populations—one or more from each of the four ideal categories—and uses as its topic environmental policy in order to focus explicitly on a policy outcome that is based on presumably universalistic rules: the actual measured level of air pollution.

As mentioned in the Pierce et al. overview, other scholarship in this area often focuses on single groups that are socially constructed in a particular way, such as the mentally ill (McSween 2002), people with disabilities (Schur et al. 2003), veterans (Jenson 2005), social housing tenants (Laffin 2013), and prisoners (Nicholson‐Crotty and Nicholson‐Crotty 2004). For example, Jean Reith Schroedel and Daniel R. Jordan (1998) used social construction theory to evaluate the US Senate’s response to AIDS between 1987 and 1992. They found that the social construction model provides mixed results but important insights into how political processes affected AIDS policy design during this period (Schroedel and Jordan 1998). McSween (2002) argued that mental health receives less government support than does (physical) health because of the stigma attached to mental illness. Schur et al. (2003) studied perceptions of the disabled vis-à-vis government responsiveness to their needs and found that non-employed people with disabilities significantly believe that government is unresponsive to their concerns. Also, Nicholson‐Crotty and Nicholson‐Crotty (2004) provide empirical evidence of the important role of the perception of target populations in the case of inmate health. They argued that social construction, rather than need or fiscal capacity, appears to determine the decisions of state legislators. Diane Hirshberg (2002), at the Annual Meeting of the American Educational Research Association, presented a paper studying the construction of race and ethnicity by Members of the Health, Education, and Social Services committees in both houses of the Alaska State Legislature. She found that the social construction of Alaskan Natives undermines sound policy decision making.

Some scholars consider more than one type of socially constructed group at a time, but we have found only one that includes all four types in a single analysis (see Husmann 2015, below). For example, Donovan (1993) studies persons with AIDS and includes two groups, arguing that those with AIDS are either categorized as Dependents or Deviants (p. 6); also, this article focuses on outputs—that is, the specific structure of the Ryan White Act of 1990—rather than outcomes of the Act. Using survey experiments and focusing on the relationship between “problem targeting,” “place targeting,” “people targeting,” and public opinion (ibid.), Lawrence et al. (2010) look at three questions, including “Third, is public support for urban public policies contingent on the social construction of the target group?” In their analysis, they focus on one dimension of the Schneider and Ingram (1997) typology,—social construction, positive or negative—abstracting from the power dimension, and include children and elderly as positively constructed and government workers and single low-income mothers as negatively constructed (p. 426). Their findings focus on survey experiment support/opposition responses, rather than actual policy outcomes (and they find that positive and negative social constructions are associated with important differences).

On the other hand, some scholars choose to focus on situations in which social constructions may be changing between deserving and/or undeserving, and this often leads to consideration of a group over time or place, or of more than one group (Barrilleaux and Bernick 2003; Schriner 2005; Van Oorschot 2006; Lawrence et al. 2013). In a qualitative study, Hudson and Gonyea (2012) argue that the social construction of the aged has moved from Dependents to Advantaged and, via the Boomer generation and contextual factors, to Contenders; thus, they look at three of the four Schneider and Ingram (1997) categories. Their article provides significant and detailed forecasts of implications of the change to Contender status for the aged, but does not assess actual outcomes. A qualitative study conducted by Salomon and Hogan (2008) assesses the evolution of one prison’s AIDS policy. She finds that the changing perceptions of the agents and targets allowed a more humanistic approach to prisoners with AIDS. More recently, Schneider herself applied social construction theory to cross-state differences and similarities in US punishment policy and finds support for social construction theory (2012).

Barrilleaux and Bernick (2003) did a pooled cross-sectional time-series analysis for 1990–1996 and found that greater electoral competition results in more benefits for “deserving” poor but fewer benefits for a less attractive constituency, the “undeserving” poor (Barrilleaux and Bernick 2003:17).Footnote 1 Scholars Michael W. Link and Robert W. Oldendick (1996) have used Schneider and Ingram’s (1993) social construction (SC) theory to study how the “social construction differentials in the minds of white Americans affect their attitudes toward the issues of equal opportunity and multiculturalism” (Link and Oldendick 1996: 149; and see DiAlto 2005; Newton 2005; Nicholson-Crotty and Meier 2005; Bensonsmith 2005). Beaton and Tougas (2001) also studied the attitudes of 264 women and men, mostly scientists and engineers from the academy, toward groups that are targeted by affirmative action, such as women, visible minorities, and disabled persons. Their results indicate that the reaction to social justice concerns is affected by the group targeted (Beaton and Tougas 2001: 61).

Other researchers have also examined several different groups in a single study but without assigning them to the Schneider and Ingram (1997) categories. Van Oorschot (2006) “examines [cross-country] European public perceptions of the relative deservingness of four needy groups (elderly people, sick and disabled people, unemployed people, and immigrants)” (Van Oorschot 2006: 24). He found that Europeans share a common and fundamental deservingness culture: Across countries and social categories there is a consistent pattern that elderly people are seen as most deserving, closely followed by sick and disabled people; unemployed people are seen as less deserving still, and immigrants as least deserving of all (Van Oorschot 2006: 23). Although Van Oorschot’s (2004) concern was ranking these groups with regard to deservingness, he did not categorize his four groups based on SC theory. He did not assign each of his groups to any of the Schneider and Ingram (1997) groups, and it is not clear where his four groups fit within the four categories of the SC theory. For example, it is not clear whether he considered elderly people as Advantaged or Contenders. Also, it is not clear which group is the Dependent group: disabled or unemployed people—or both. Sharp (2009) focuses on the policy tools part of Schneider and Ingram (1997) and the argument regarding the chilling effect of certain policy tools on democratic engagement (cf. Mettler 2011). Though Sharp (2009) does not focus on the four types of socially constructed target populations, her findings indicate that using tools that have a chilling effect influences all types of socially constructed citizens—not only those (generally non-Advantaged) targets who are the direct recipients of chilling policies (p. 190).

Lawrence et al. (2013) also examine the social structure of four groups: elderly, children, government workers, and poor single mothers. They study “the role that party identification and political ideology play in the tendency of members of the public to respond to target groups that vary in public support” (Lawrence et al. 2013: 199). Consistent with Schneider and Ingram’s theoretical claims, they found that “policies that target what [they] thought would be more popular target groups (elderly people and children) tended to enjoy higher levels of support” (Lawrence et al. 2013: 206). However, like Van Oorschot (2006), they did not have a clear categorization of their four groups that fits with Schneider and Ingram’s four groups.

A recent article about obesity published by Maria A. Husmann in September of this year (2015) used a between-subject experiment to study whether the 256 participants would support a penalty or benefit for the following four socially constructed population groups: middle-class hardworking obese/overweight individuals (Advantaged); members of the food, advertising, or weight-management industries (Contenders); overweight and/or obese children (Dependent); and obese individuals of low socioeconomic status (Deviant). Consistent with Schneider and Ingram’s theory (2005), she found that the public’s perception of benefiting and punishing policies can be changed by changing the narrative of the target population (Husmann 2015: 20). The focus on experimental data precludes analysis of actual policy outcomes.

According to Pierce and coauthors (2014), social construction theory has only rarely been used to examine environmental policy. Instead, the majority of work looking at the effect of US air pollution on social groups is embedded in the environmental justice literature and focuses on race/ethnicity and/or income. Examples are national studies that “have generally concluded that minorities (particularly Hispanics, Asians/Pacific Islanders, and African Americans) have been at (potentially) higher risk of exposure to air pollution induced by criteria air pollutants at the national level (Gelobter 1992; Wernette and Nieves 1992; Liu 1998)” (Liu 2001: 203). In addition, Morello-Frosch et al. (2001) analyzed the combined lifetime cancer effects of 148 air pollutants in Southern California and found that increases even in such an overall, combined indicator of increased air pollution effects was influenced by race and ethnicity holding constant income and housing status. These findings in the environmental justice literature do provide evidence that air pollution affects different social groups differently and also document an ongoing concern with the social justice effects of air pollution, but do not directly address social construction theory.

In this paper, inspired by the rank-order hypothesis implied by Schneider and Ingram’s political form of social construction theory, we examine the relationship between the level of air pollution—because it is a widespread policy outcome, conceptualized in a universalistic manner, and affecting multiple types of groups—in California’s Central Valley cities, and the social constructions of several different subpopulations within these cities. Here, we further contribute to the Schneider and Ingram (1997) social construction literature by moving beyond outputs and examining outcomes—actual levels of air pollution as a concrete expression of policy. The focus on outcomes fits Schneider and Ingram’s concerns, for they ultimately care about “who wins and who loses” and the effect of policy designs on “the distribution of wealth and other resources within a society” (1997: 101). In focusing on outcomes, our analysis uses inferential statistics, rather than the qualitative methods that are often used in the social construction literature (see, e.g., Hogan 1997; Menahem 1998; Chanley 2005).

The research approach

California’s Central Valley is arguably one of the most air-polluted areas in the USA; within the top twenty-five most air-polluted cities in the USA, eight cities out of twelve cities in California are in the Central Valley (American Lung Association 2011: 13). Central Valley cities such as Fresno, Modesto, and Bakersfield are ranked among those that have the highest unhealthy air environment in the nation (ibid). Such an unhealthy atmosphere has its impact on residents’ health. People in the Central Valley have high rates of asthma, heart attack, and several other diseases that are related to air pollution (California Department of Public Health 2007). California’s Central Valley has significant problems with air pollution even though California is recognized as a world leader in air-quality policy (UCLA School of Public Health 2008).

Because of its size and demographic diversity, California presents a rich variability of social groups, including migrant farm-workers and politicians in charge of one of the largest state budgets in the USA. Analyzing causes of all of California’s air quality incidence, however, is complicated because of the geographic diversity of the state, which includes 840 miles of coastline, desert tracts of which the Mojave is only one, and six different mountain ranges (Netstate 2011, California). As stated by Liu (2001), “The quantity of emissions is only one factor for determining eventual environmental impacts; weather conditions and topography are particularly important for fate and transport (dispersion) of air pollutants” (195). The Central Valley, about 450 miles long (op. cit. Netstate), is a distinctive geographic structure within the state.Footnote 2 Because of the mountains that surround the area, the air pollution that develops in the Central Valley itself, and the air pollution that comes from surrounding areas over the mountains, largely remains within the Valley. Using only the Central Valley for this analysis of social construction theory rank-order hypotheses for widespread environmental outcomes simplifies the geographic and weather controls that could be required for a statewide (or larger) analysis. Air quality may be poor for all (as Morello-Frosch, Pastor and Sadd 2001, find for Southern California), but if decisions are made in a degenerative system, results still should be better for some groups than others based on their social constructions.

Although air quality is largely a public good, it is still a political concern of all developed countries as well as some developing countries. People in developed countries already consider having clean air a necessity rather than a luxury, as it might be considered in some of the developing countries. In California, the acts and regulations that aim to enhance air quality have increased dramatically since 1930 to reach a peak in 2007 (Air Resource Board 2015). These regulations indicate political concern regarding air quality. However, although regulations apply to the whole state and are discussed in a universalistic way as applying to everyone equally, the level of implementation and enforcement can vary. Indeed, though the air quality in the Central Valley may be poor for all, there is still variation between cities. As Schneider and Ingram (1997) note (see above), in a degenerative political system apparent universalism may mask the way that outcomes affect different socially constructed groups differently.

The Central Valley is divided into nineteen counties, and these counties have 101 cities and towns (Umbach 1997). In order to apply Schneider and Ingram’s model to our research, we measure attributes of these cities and towns based on the political strength and the social constructions of their residents. Based on information regarding social attitudes toward groups, we use the percentage of each city’s population with bachelor’s degrees or higher, as well as the number of firms, to identify the Advantaged in cities. The number of banks is used to detect the Contenders in cities, and children as well as single mothers are used to identify the Dependents in cities. Deviants are identified by per capita crimes and the number of prisons. These are also groups that are generally relevant to social justice implications of air pollution. We further explain these particular choices below.

Description of the data

To test the relationship between air pollution levels—the outcome of air pollution policy—and social construction theory, we created a dataset that includes one dependent variable, seven independent variables, and three control variables. Our goal is to advance our understanding of social construction theory by testing whether the cross-city variation in air pollution is explained by the social constructions of subpopulations within cities.

Dependent variable: PM10 pollution

To measure air pollution of major concern, a researcher can use any of the seven “criteria pollutants” as defined by the Clean Air Act: Carbon Monoxide (CO), Hydrogen Sulfide (H2S), Nitrous Oxide (NO2), Ozone (O3), Particulate Matter of 10 microns or less (PM10), Particulate Matter of 2.5 microns or less (PM2.5), and Sulfur Dioxide (SO2). A meta-analysis by Dr. Hazrije Mustafić, M.D., M.P.H., and his colleagues concluded that “all the main air pollutants, with the exception of Ozone, were significantly associated with a near-term increase in” myocardial infarction (MI) risk (Mustafić et al. 2012: 1).Footnote 3

We have chosen to use PM10 because it is one of those that “appear[s] to pose the greatest health concerns in California’s ambient air” (Air Resource Board 2015). Further, PM is one of the air pollutants studied by Liu (2001), and therefore, there already exists a stream of research since then on social justice implications of its distribution. Ozone is also of serious concern in California, but Liu (2001) points out that it is difficult effectively to model ozone.

The Air Resources Board approved 20 micrograms per cubic meter (µg/m3) as the California annual standard for PM10 because significant harmful health effects could occur among both adults and children if exposed to levels above 20 µg/m3 (Air Resource Board 2015). Yet in our data the average value for PM10 is over 28 µg/m3 (but with variance; see Table 1). In addition to heart attacks, an elevation of PM10 is likely to “increase respiratory disease and cause lung damage, cancer, and increased mortality” (ibid). In addition, air-quality monitoring-site data for most Central Valley cities are available for PM10, which is not true of all criteria pollutants. For example, data from the California Air Resources Board on Nitrous Oxides (NOx) were limited (Bushouse 2009). We chose the annual average of PM10 over PM2.5 (which is also quite harmful) because more data were available for PM10 than for PM2.5.

Table 1 Descriptive statistics: cities in the Central Valley, 2009

Based on these considerations, we use levels of PM10 in Central Valley cities in 2009 as our dependent variable. Even so, because of limited data, we observe the PM10 dependent variable for just 70 Central Valley cities.

Advantaged

As mentioned above and shown in Fig. 2, we include measures of all four of the Schneider and Ingram (1997) socially constructed group types, and we measure these using seven independent variables. The first two independent variables, Total Number of small Business Firms in the city, and the percent of the city’s population aged 25 or over with Bachelor’s Degrees or Higher, are used to measure the presence of Advantaged groups. In this study, we use US Census data for Small Business Firms and data for residents’ college educations, which are the averages for the years from 2005 to 2009 for each city.

Fig. 2
figure 2

Groups used to operationalize social construction theory

Anglund (1998) finds evidence over a long period (1953–1993) that “small businesses” are positively constructed in the USA as deserving and the “source of a desirable social condition” (45). According to the US Small Business Administration (US SBA), as legally defined, “small” businesses are those with fewer than 500 employees; about 99.7 % of all US firms are small businesses (US SBA, n.d.).Footnote 4

We also expect that the college educated are an Advantaged group. Educated people are positively constructed since they are seen as hardworking and useful to the country; other evidence indicates they are powerful since they are known to participate more in politics, and they have the wherewithal to participate more effectively than less-educated groups. As stated by Villasenor (2015),

From empirical research there is overwhelming evidence for the private and public benefits of postsecondary education, and also for the fact that such education is positively socially constructed. A poll conducted by Gallup and the Lumina foundation reported that 97 % of Americans feel that having a certificate or degree beyond high school is somewhat or very important.Footnote 5 … All this strongly suggests that a college degree is positively constructed….

Together, these elements make the college-educated Advantaged (positively constructed and politically more powerful).

These two groups—Small Business Firms and Bachelor’s Degrees or Higher people—are not just socially constructed as advantaged, but also hold positive environmental attitudes, making them relevant vis-à-vis air pollution outcomes. For example, Halme and Korpela (2014) argue that small entrepreneurs have a sustainability motivation and also the flexibility to adopt new environmental innovations. Van Liere and Dunlap (1980), McMillan et al. (1997) and Rasool and Ogunbode (2015) find that education has a statistically significant positive effect on environmental concern (cf. McMillan, et al. 1997, 98; Masurel 2007; and Revell et al. 2010).

Contenders

To measure Contenders, we use the number of active Banks in each of the cities.Footnote 6 The variable Banks is used to detect Contenders because historically bankers are viewed as powerful but greedy (Schneider and Ingram (1997): 108; and see Anglund 1998, and Sidney 2005: 127–128).Footnote 7 Given the housing crisis and the Great Recession—often blamed upon the banks (Muddy Water n.d.)—which began in 2007 or 2008, we believe it is especially plausible that in our studied year of 2009 Bankers are Contenders: still powerful due to relative wealth, but socially constructed as undeserving. In addition, banks usually represent money, and those who have money do value their health and have concerns about the environment (e.g., Shen and Saijo 2008; Franzen and Meyer 2010; Botetzagias and Malesios 2012). Finally, all the Banks we counted were opened before 2009.

Dependents

For the Dependent group, we used US Census data on the percent of Persons Under 18 years of age (Children), and the percent of residents who are Single Mothers (the latter data are an average from 2005 to 2009). It seems clear that children are classic “Dependents” as defined by Schneider and Ingram: children are historically viewed as loved but helpless (and see Bushouse 2009). The city-level Census data for persons under 18 years were available only for either 2000 or 2010; we did not use 2010 data since that would include children unborn in 2009 (the year of PM10 observations), but by 2009 the 2000 data are very old, so we averaged the data for those two years.

For single mothers, we used the US Census Bureau’s data for Single Mother, which is defined as “Female householder, no husband present with own children under 18 years” (US Census Bureau, State and Country QuickFacts 2012b). The social construction of Single Mothers is potentially more complex than that of children—women are often seen as worthy of protection (“women and children first”), mothers are loved (consider Mother’s Day sales as just one indicator), but single parents can be seen as either noble or promiscuous. The latter fits with Lawrence, Stoker, and Wolman’s (2010) assessment of low-income single mothers as negatively constructed—so we may expect worse air-quality outcomes for Single Mothers than for Children. This also fits with the Schneider and Ingram categorization within the social construction space: Although Schneider and Ingram present four distinct categories, they also present a space within which groups can be located. Even holding constant a category (e.g., Dependents), some may receive more benefits (or fewer costs) than others, depending on whether they are closer to Advantaged or Contenders.

Much research indicates that children are more susceptible to most types of pollutants (e.g., Olson 2016; World Health Organization 2007; 62 Federal Register 19885 1997). In keeping with this, Federal law gives especial attention to protecting children from pollution (Olson 2016; and see Executive Order 13045). Therefore, children are inherently a relevant population for the air pollution policy domain—and articulated statements regarding policy goals suggest that this group should be especially protected, which contradicts the prediction if they are Dependents in the sense meant by Schneider and Ingram (1997).

However, although children are more likely to be affected by pollution than adults, their interest is usually expressed through their parents or government (Pope et al. 1995; Ward and Ayres 2004; O’Connor et al. 2008; D’amato et al. 2010; Akinbami et al. 2010). In a single-mother family (and in the US most single-parent families are headed by females), the single mother is the only person primarily available to express air pollution concern for the children. Also, several scholars argue that females often are more concerned about environmental problems than males because of their “sense of emotional empathy (Arnocky and Stroink 2011), a stronger response to the harmful effects of deteriorating environmental conditions (Stern et al. 1993), and socialization processes that promote female interdependence and an ethic of care (Zelezny et al. 2000)” (Rasool and Ogunbode 2015, 278). These considerations suggest that single mothers should be especially concerned regarding air pollution, and so if their presence increases air pollution as hypothesized via the rank-order implications of Schneider and Ingram (1997), this is a strong indication of the importance of the social construction.

Deviants

To measure the presence of Deviant populations, we use the per capita number of Crimes per city in 2009,Footnote 8 and also the number of Federal and state prisons in each city. It seems almost definitional that society socially constructs criminals as undeserving, and often as “deviants” even in the common meaning of that term. We do not claim that criminals are particularly relevant to air pollution considerations, but our notion is that convicted criminals are, by their nature, so socially constructed as Deviants that punishment—including through air pollution—is generally acceptable.Footnote 9

Controls

In addition to the variables of theoretical interest, we include three control variables: Rainfall, number of Households,Footnote 10 and City Size. Increases in the level of Rainfall lead to a decrease in the level of air pollution because rain washes air pollution out of the air (though onto the land and water). We also control for the number of Households because residences are the primary source of fuel combustion in California, and fuel combustion generates PM10 pollution (EPA 2012). City size was included because big cities are very different from small cities in many political, economic, and cultural aspects. Big cities attract more people from the whole range of socially constructed groups, and big cities are more likely than small cities to be more polluted.

Under Schneider and Ingram’s (1997) model, we expect that, generally, Advantaged groups will receive the least amount of air pollution, Contenders more pollution than Advantaged, Dependents more than Contenders, and Deviants the most (cet. par.). Translated to the coefficients to be estimated in our models—and assuming that the air pollution control policies for PM10 were created within a degenerative system—we anticipate that the estimated coefficients on variables measuring Advantaged residents (the total number of non-bank small business firms and the percentage of residents with advanced levels of education) will be smallest in magnitude and possibly negative. The coefficient on the variable measuring the presence of Contenders, each city’s number of Banks, will be larger, and if both Advantaged and Contender coefficients are negative, the Banking coefficient will be less negative (i.e., closer to zero). Coefficients on variables measuring Dependents, percentages of Children Under 18 years and Single Mothers, will be higher compared to those for the Advantaged and Contender variables. Coefficients for variables measuring Deviants, Crimes Per Capita and numbers of Prisons, will have the highest coefficients in the models to be estimated. However, we need to note that not all variables are measured in the same units, and so relative magnitudes will not always be readily apparent. We clarify this further in discussion of the results.

In short, when it comes to exposure to air pollution, rather than having standard hypotheses about whether coefficients are different from zero, or even sign hypotheses, we have rank-order hypotheses, predicting the magnitude of the coefficients to be as follows:

$${\text{Advantaged}}\,< \,{\text{Contender}} \,< \,{\text{Dependent}}\,< \, {\text{Deviant}}.$$

Results of the statistical analysis

Table 1 shows the descriptive statistics for the data, which includes the maxima, minima, means, and standard deviations. As indicated in the table, we have seventy observations.Footnote 11

We used Ordinary Least Squares (OLS) regression with robust errors to estimate the following statistical model, and we tested for heteroskedasticityFootnote 12 and excess multicollinearity.Footnote 13 We used robust errors as a control for bias that could be caused by spatial autocorrelation, which is expected to be present when studying air pollution at the level of individual cities within a single geographic area that is a shared air basin.

$$\begin{aligned} {\text{PM10}} & = \alpha_{0} + \alpha_{1} {\text{SmallFirms}} + \alpha_{2} {\text{PopCollege}} + \alpha_{3} {\text{Banking}} + \alpha_{4} {\text{ChildrenUnder}}18 \\ & \quad + \alpha_{5} {\text{SingleMothers}} + \alpha_{6} {\text{CrimesPC}} + \alpha_{7} {\text{Prisons}} + \alpha_{8} {\text{Rainfall}} \\ & \quad + \alpha_{9} {\text{Households}} + \alpha_{10} {\text{CitySize}} + \varepsilon \\ \end{aligned}$$
(1)

As given in Table 2, we find that overall the social construction model explains variations in PM10 outcomes quite well. The R 2 for the social construction model is 0.597, with an Adjusted R 2 of 0.528, and the F indicates significance of the equation.Footnote 14 As expected, the control variable Rainfall is estimated to decrease PM10 significantly, with each inch of rainfall reducing PM10 by 0.425 μg/m3. Of the other two control variables, Households is estimated to have essentially no effect, but City Size is estimated to increase PM10 even controlling for all other variables.Footnote 15 We now proceed to the coefficients for the social construction variables.

Table 2 Regression results for 2009 levels of PM10 in Central Valley cities (range 13.3–42.5 μg/m3)

First we note that only two of the estimated policy coefficients are statistically significant: those for Population with College Degrees and those for Small Firms (both included as measures of Advantaged groups). However, as noted over and over by many statisticians (e.g., Agresti and Finlay 2009; Gujarati and Porter 2009), even in the case where the estimated coefficients are not statistically significant, our best guess regarding the coefficients is not zero; instead, our best guess is the estimates themselves.Footnote 16 Thus, we will focus on interpreting the rank-order of the estimates, rather than their statistical significance, as this is what is directly relevant to our hypotheses.

Consider the variables measured as percentages of the population within each city: Single Mothers, Children Under 18, and Population With College Degrees (BA or higher). Because they are measured in the same units, their magnitudes are directly comparable. Single Mothers measure Dependents who may also have some elements of a Deviant assessment and therefore be closer to the right side in the social construction space; Children are expected to be close to pure Dependents; and those with College Degrees should be Advantaged. As hypothesized in political social construction theory, of these groups Single Mothers are estimated to receive the most pollution, Children less than that, and Advantaged the least.

Prisons, Banks, and Small Firms are all measured in numbers, and so their coefficient magnitudes can be directly compared. We included these three groups to control for Deviants, Contenders, and Advantaged, respectively. A one-unit increase in the numbers of Prisons in a city is estimated to lead to the greatest increase among these three in PM10 pollution, and a one-unit increase in the numbers of Banks and Small Firms is both estimated to decrease the amount of PM10 pollution. Thus, the rank-order of Prisons/Deviants is the highest of these three, as expected. However, Banks are actually estimated to suffer less PM10 air pollution than Small Firms, and this is not as predicted by our use of Schneider and Ingram’s (1997) model since, if Small Firms correctly measure the Advantaged, and Banks Contenders, then Small Firms should receive the least of these.

Though difficult to compare directly to either percent measures or number measures given the difference in measurement, levels of PM10 pollution are estimated to increase with Crime per capita. A 0.1-unit increase in Crime per capita is estimated to lead to an increase of 0.37, making it one of the largest in magnitudeFootnote 17—further supporting the basic notion that Deviants can expect to receive more pollution.

Interpretations

Our goal for this analysis was to examine what we believe to be the heretofore unconsidered rank-order predictions of Schneider and Ingram’s (1997) social construction theory. To do this, we use quantitative empirical methods to estimate whether widespread environmental policy outcomes in California’s Central Valley—as measured by actual city-level PM10 air pollution levels—support rank-order predictions across all four SC categories of Advantaged, Contender, Dependent, and Deviant.

The results of this first empirical analysis of a complex rank-order hypothesis are quite suggestive. Most coefficients estimated for the PM10 model provide evidence in support of social construction theory’s rank-order prediction. Schroedel and Jordan, in one of the first uses of the Schneider and Ingram social construction model, found that the social construction model provided mixed results but important insights into how political processes affected AIDS policy design during the period studied (Schroedel and Jordan 1998). Our results, using data a decade later and looking at rank-order outcomes rather than single-group outputs, are similar: Results are mixed, but provide important insight.

In general, the results support our complex rank-order hypothesis, and some in ways that are directly contrary to other views of social policy. For example, in pollution-policy rhetoric, children are presented as a special population to be protected from pollution, but increases in their presence are estimated to increase the amount of PM10, not to reduce it as is estimated for groups we conceptualized as Advantaged or Contenders.

Fitting the hypothesis that Single Mothers are socially constructed as Dependents with Deviant aspects, the presence of Single Mothers is also estimated to increase the pollution outcome. Given the high residential mobility in the USA and the fact that housing prices are affected by type of amenities and disamenities in an area, it might be thought that a simpler explanation of this outcome is simple market dynamics. This claim has, in fact, been an enormous issue of debate in the environmental justice literature. However, significant research indicates that pollution imposition on populations is more important than low-social-status populations moving to the pollution. One of the best studies in this vein is by Pastor, Sadd, and Hipp (2001), who note that “Some critics have responded that the contemporary correlation of race and hazards may reflect post-siting minority move-in, perhaps because of a risk effect on housing costs, rather than discrimination in siting.” Yet their finding is that “disproportionate siting matters more than disproportionate minority move-in in the sample area” (cf. Campbell, Peck and Tschudi 2010). Thus, their findings indicate—as do others in the environmental justice literature—that pollution distributions with respect to different communities, though obviously caused by both, are more due to pollution policy decisions rather than to residential mobility.

The least supportive results are for Bank and Small Firms vis-à-vis each other, but this may be explained by either high multicollinearity or incorrect identification of their respective social constructions. Small Firms and Banks are highly correlated in the data, with a Pearson’s r of 0.9012 and statistical significant at 99 %. High multicollinearity does not cause bias, but it increases standard errors and so can cause overlap in the estimator for highly correlated variables that have similar coefficient magnitudes. Further, one of the problems with the Schneider and Ingram (1997) model is difficulty in identifying groups’ social constructions, which are, as they themselves note, often contested. Pierce et al. (2014: 13) report that Anglund (1998, 2000) identifies small firms as Dependents. If this identification is correct, that could explain the bank-to-small-firm ranking even within an SC model since Dependents should receive more pollution than Contenders, and here Small Firms are estimated to receive more pollution than Banks. Similarly, “Banks that lend money for mortgages are identified [by other scholars] as contenders (Hunter and Nixon 1999) but also as advantaged (Drew 2013)” (Pierce et al. 2014: 16). If Banks are Advantaged rather than Contenders, then their estimated rank order still fits an SC rank-order hypothesis.

As indicated above, these results are suggestive, but we cannot have as much confidence in them as ideal since only two policy variables are statistically significant. Both of the statistically significant variables are included to measure Advantaged groups, and their estimated coefficients suggest that these groups receive among the lowest amount of PM10 air pollution in California’s Central Valley. If in fact we should conclude that only the Advantaged succeed in reducing air pollution for their cities, and, for example, there is no effect from the presence of children, then this still supports Schneider and Ingram’s SC theory since these outcomes still suggest a deceptive policy system.

Thus, whether we accept all the rank-ordered estimates, or only those that suggest the Advantaged get less pollution than everyone else, these results unfortunately indicate that pre-2009 air pollution policy was set in a deceptive, degenerative policy system.

Conclusion

In this research, we analyze Schneider and Ingram’s (1997) social construction theory to help answer the following question: Is there support for Schneider and Ingram’s implicit hypothesis that policies with widespread effects should provide rankable outcomes for several different socially constructed subgroups? Overall, their social construction theory is intuitively appealing and has led to a significant literature, but our review of the literature indicates that all four types of socially constructed groups have not generally been examined at once, especially using inferential statistics and focusing on actual rank-order outcomes (the actual effects of policies) of a widespread issue, rather than narrower outputs (such as legislation).

The presented analysis provides strong rank-order results, though in this small dataset most coefficients are not statistically significant. Still, given the complexity of the rank-order hypothesis, the results suggest rejecting the null hypothesis that there is no relationship between the level of air pollution in different cities of the Central Valley and the social construction of populations within those cities. Further, for those scholars who wish to interpret only statistically significant results, the pollution benefits estimated to be received by Advantaged groups, in contrast to groups such as children, who are well known to be especially susceptible to pollution, are still supportive of a deceptive policy system.

It is important to remember that Schneider and Ingram’s (1997) theory requires at least two components: First, policies must be developed in a degenerative policy system, and second, groups must be relevantly and saliently socially constructed.Footnote 18 Further, for empirical testing, social constructions must be correctly identified by analysts. Though not conclusive, the results here indicate that social construction theory may be used in a more detailed way than it has in the past: To assess the relative outcomes, many differently constructed groups can expect from policies that have widespread outcomes. We believe that further empirical research should assess the rank-order hypothesis of Schneider and Ingram’s (1997) social construction theory, for we believe that the results presented herein suggest that their SC theory is even more useful as a predictor than has been realized.