• Residential similarity preference;
  • Environmental justice;
  • Racial parity;
  • Agent-based modeling;
  • Residential choice constraints


  1. Top of page
  2. Abstract
  3. What Can Constrain Residential Choice Sets?
  4. An Agent-Based EJ Model
  5. Analysis
  6. Discussion and Conclusion


In the environmental justice literature, uncertainty exists about the underlying causes of environmental risk disparities, especially as they relate to residential choices. To simplify, the two dominant views are racism/discrimination versus inevitable market dynamics. In this article, we move aside from these to examine the potential role of various residential choice constraints on environmental injustice and how they may be interrelated.


Using an agent-based simulation model, we examine the interaction of race-based constraints with other experimental conditions that can affect minorities’ residential choice sets.


Simulation experiments demonstrate that if the minority holds relatively lower similarity preferences, the environmental quality gap declines when other conditions are held constant. However, racial parity in communities also decreases the environmental quality gap, as do slower population growth and larger geographies.


These results enable us to look at the problem of race-based environmental injustice more holistically, and begin to think about holistic solutions that may finally address what has heretofore been an intractable social problem.

One major focus of environmental justice (EJ) research concerns the disproportionate collocation of environmental disamenities with the poor or racial/ethnic minorities. Substantial evidence of environmental injustice has been accumulated at various societal levels (e.g., Ringquist, 2005; Mohai, Pellow, and Roberts, 2009), but these studies are not without conflicting findings (e.g., Hamilton, 1995). In search of a reason for these inconsistent conclusions, EJ researchers have pointed out that the empirical results are sensitive to analytical decisions such as the spatial unit of analysis (Baden, Noonan, and Turaga, 2007; Noonan, 2008) and model specification (Bowen, Atlas, and Lee, 2009; Chun, Kim, and Campbell, 2012).

Though important, the debate does not necessarily answer the underlying causal question: What social processes or conditions drive differential racial exposures to environmental risk? Early EJ research was premised on the assumption that racially disparate outcomes occur because some firms (or other organizations) have a discriminatory motive behind their siting decisions, so hazardous facilities and firms need to be a primary target of regulation and policy (e.g., Bullard, 1990). This assumption requires further scrutiny, not only because it is difficult to examine the true motives of firms’ decisionmakers, but also because firms’ siting decisions occur in a dynamic urban system where residents move in and out at the same time (Pastor, Sadd, and Hipp, 2001), and residential choice must be carefully incorporated in order to illuminate the complex social process (Been and Gupta, 1997). Such a dynamic process poses analytical challenges. Social interactions and interdependencies among heterogeneous entities are not easy to conceptualize, capture, or analyze. In addition, neighborhoods within which firms and residents reside vary and change, in terms of racial composition, population growth, and regional landscape.

To begin to address this complexity, we rely on a simulation approach that allows us to model emergent results from the uncoordinated dynamic decision-making processes of firms and residents in different residential and cultural contexts, examining the consequences of these processes on environmental quality among different social groups. Firms’ siting decision scenarios are based on existing, competing theories in EJ. Residents’ location choices are conceptualized with an important constraint: the residential similarity preference (Schelling, 1978), which has been generally overlooked in previous EJ research, though considered more in urban economics and demography (Epple, Filimon, and Romer, 1984; Banzhaf and Walsh, 2008). Using a simulation model, we attempt to disaggregate how the heterogeneous preferences of different social groups to live in mixed-race communities alter EJ outcomes, especially as other factors add constraints.

What Can Constrain Residential Choice Sets?

  1. Top of page
  2. Abstract
  3. What Can Constrain Residential Choice Sets?
  4. An Agent-Based EJ Model
  5. Analysis
  6. Discussion and Conclusion

While the historically prevailing view in the EJ field on the link between race and environmental risk (or quality) disparity is that environmental injustice mainly results from organizations’ siting processes (Bullard, 1996), an alternative explanation based on market dynamics has also gained currency in the EJ literature. Under this view, the firms’ intention is not necessarily discriminatory nor does discrimination serve as a sufficient explanation of environmental risk disparity (Been and Gupta, 1997). Firms’ main intentions are to maximize profits and reduce the cost of doing business, seeking locations where the land, labor, or materials are cheap, and where the potential costs of an accident are low (Hamilton, 1995). Such areas also tend to be where lower-income minority groups live. Thus, firms do not intend to discriminate against racial and ethnic minority groups, but make location decisions based on an economic rationale. In addition, the racial and ethnic composition of residents may change after disamenity-producing firms and facilities are located within a neighborhood because these firms produce negative effects. Some residents may want to leave such areas, and these individuals who are able to leave are more likely to be comparatively affluent (Hamilton, 1995). Poor residents, who have fewer resources, may not have such an option. The movement of affluent people out of the area may depress property values, making housing more affordable, which further attracts the poor to move in. As space becomes more constrained, low-cost options that are away from disamenities become harder to find, and the quality with which the poor live degrades. Because minority residents are disproportionately poorer than white residents, observed disparities will appear to be caused by racial factors, but are, rather, results of market mechanisms (Been and Gupta, 1997).

While we acknowledge that economic-based firm decision making can certainly affect the distribution of environmental quality across different social groups, our current research is based on the premise that considering only this effect is insufficient for understanding—and thus creating policies for—the broader social phenomenon. Specifically, constraints on residential choice such as the relative magnitudes of similarity preferences and racial proportions, along with varying spatial contexts, may be intertwined as important social mechanisms that contribute to EJ outcomes. Ignoring the complex social conceptions of race and racism and looking at the constraints that may affect minorities even without these factors helps us disentangle the many elements that may cause the environmental injustice that we observe. Understanding unintended causal factors, as well as intended ones, should help us design more effective policy.

Residential Preferences

The basic processes examined here are in line with the work of Schelling (1978), who suggested that micro motives (or preferences or choices) of individuals could lead to unexpected (and unintended) macro patterns in neighborhoods. In his self-forming neighborhood model for dichotomous mixing (e.g., white vs. black, girls vs. boys, and so forth), Schelling developed a thesis on residential preference, assuming that people's preference-based decisions are dynamic, so they move around until a location satisfying their preference is identified (e.g., being near some fraction of others who are “like them”). In Schelling's examples, residential location decisions are assumed to be driven by preferences, but he recognizes that it is not always easy to isolate individual preference- or color-based discrimination or segregation from other reasons such as economically induced and collectively enforced factors.

To explore segregation, Banzhaf and Walsh (2008) incorporated this concept into a general equilibrium model based on Tiebout's (1956) sorting mechanisms. They found that the relationship between public goods and spatial residential patterns was altered, sometimes in unintuitive ways, by agents with preferences for living in proximity to other similar agents. Segregation was discerned in two ways: with the wealthy excluding the poor from locating near public goods, and also with the majority ethnic group opting to locate away from a minority ethnic group. Adjusting the location of public goods or public “bads” did not eliminate segregation, but, rather, reinforced it, especially when the minority group was, on average, poorer than the majority group.

Taking a similar idea out of comparative statics and into a dynamic simulation, Eckerd, Campbell, and Kim (2012) explored how uncoordinated social interactions between firms and residents as they make location choices can lead to environmental injustice. Firms’ siting decisions and their impacts on environmental quality disparity were examined using three scenarios: siting based on the lowest cost, a political rationale to stay away from the majority (cf. Hamilton, 1995), and a discriminatory motive to be close to the minority. Residential choice is constrained by the similarity preference, which limits potential locations that are available for residents. Although the two studies are not directly comparable, Eckerd, Campbell, and Kim (2012) reported that the key to large environmental disparities was high similarity preferences held by both groups, while Banzhaf and Walsh (2008) found that residential similarity preferences mean that environmental disparities could worsen for the minority group, even as overall environmental conditions improved. Both works show the importance of unintended consequences from interactions.

Even after these two studies, several nuanced but important questions surrounding the effects of residential preference remain. For example, Schelling (1978) made no distinction between the similarity preferences of groups and assumed a rather low proportional preference. Clark's empirical study (1992, using data from Omaha, Kansas City, Milwaukee, Cincinnati, and Los Angeles) found that while Schelling's (1978) conception of residential choice is generally correct, the magnitude of similarity preferences regarding neighborhoods is much larger than what Schelling assumed. Banzhaf and Walsh (2008, 2010) assumed different preference levels for the two groups based on previous empirical research. It appears that minority groups tend to have lower similarity preferences than whites, possibly because of the paucity of residential options that would be available for minorities under higher similarity preferences (Banzhaf and Walsh, 2008). If this is the case, an assumption that the minority's similarity preference is equal to that of the majority (as assumed in Schelling (1978) and Eckerd, Campbell, and Kim (2012)) may be too rigid. Precisely because they are in the minority, the minority group with similarity preferences makes residential choices with far fewer alternatives than their majority counterparts. Thus, the question arises: When relaxing the minority similarity preference to a lower level as suggested by Clark (1992), would the environmental quality gap be smaller than when the minority groups hold the same high level of similarity preference as the majority?

Residential Conditions

Ultimately, the EJ argument is space based. While varying similarity preferences may be an important cultural factor to explore, it leads to the more general question of whether environmental disparities can arise simply because the minority has a smaller set of locations from which to choose. Relaxing similarity preferences is one way to increase the locational choice set for the minority, but there are other ways to expand the minority choice set. For example, when the minority proportion is closer to the majority proportion, this implies more options for minorities; would racial parity narrow the environmental quality gap holding constant similarity preferences? Though the true impacts of minority status are more than just numerical, in this article we analyze the impacts of residential choice constraints on minorities separately from other social constraints in order to examine their independent effects. By understanding unintended parameters of the EJ problem, we hope to gain insight into where policy leverage is available and what factors may stymie policy effectiveness. As Banzhaf and Walsh (2010) show, in complex systems seemingly obvious interventions (such as adding amenities to minority neighborhoods) may have unintended results.

If there are indeed interdependencies among residents’ similarity preferences, racial composition, and the location choice sets available, would the EJ outcome emerging from these interdependencies remain the same under different residential conditions, such as varying neighborhood characteristics? Evidence indicates that certain neighborhood characteristics matter for EJ outcomes, especially as a pressure factor on firms’ decisions and behaviors, such as siting and toxic release (Arora and Cason, 1998; Campbell, Peck, and Tschudi, 2010). High population density in a neighborhood—indicating more people in a given area—is assumed to be a disincentive for environmental disamenities due to potentially higher compensation costs (Hamilton, 1995). But, from a residential choice point of view, a higher population density indicates fewer remaining residential choices. If the resident is a member of a numerical minority with a strong similarity preference, then the resident is more likely to be forced to select an (environmentally) inferior location than is a majority.

Using this line of reasoning, we consider two unexamined residential conditions—differences in growth rate and region size—for which the conditions may alleviate or aggravate environmental quality disparities by influencing residential choice sets within a finite neighborhood. In a rapidly growing region, finite land can be quickly filled by residents and firms, continuing to decrease the available residential choice sets. In a slower growing region, we expect fewer constraints on choices. In a relatively small region, the choice set for both groups is more limited than in a larger region and, most likely, the numerical minority is the one more affected by this limitation in regards to residential choice, which may aggravate environmental disparities. We explore these questions, expanding the EJ agent-based model used for Eckerd, Campbell, and Kim (2012).

An Agent-Based EJ Model

  1. Top of page
  2. Abstract
  3. What Can Constrain Residential Choice Sets?
  4. An Agent-Based EJ Model
  5. Analysis
  6. Discussion and Conclusion

The base simulation “world” consists of 51 × 51 grid cells (for a total of 2,601 plots) representing pieces of land. The world is initialized with one firm (producing either an environmental amenity or disamenity) at the center, and 50 residents at randomly selected locations within the 20 × 20 plots that surround the firm. The population growth rate for the base model is set to 5 percent. Each firm employs a set number of residents and new firms are created as there is labor demand. In contrast to most Tiebout and Schelling models, agents are not infinitely mobile; only newly introduced firms and residents seek locations. Those that remain in the world remain at their initially chosen location until they “die,” which opens the space.1

Plots of land have two key attributes: price and environmental quality. At the beginning of the simulation, the attributes of the plots are homogeneous and set to a hypothetical expected value of 50, ranging from 1 to 100 in a uniform distribution. When new firm and resident agents are introduced to the world, they influence the price and quality of plots as described below.

The price function is:

  • display math(1)

where the price of plot j at time t is a function of the price of plot j at t − 1 and a gap between quality and price at t − 1 weighted with two other factors at t: a utility of plot j (Equation (2)) and an occupancy rate (inline image) of the surrounding eight plots for plot j . If a plot j is underpriced or overpriced given the quality at t − 1, the price at t is adjusted based on the difference. The utility score of plot j is as follows:

  • display math(2)

where the utility score of plot j at time t is determined by the price and quality of plot j at t − 1 and the distance between the plot j and any nearest firm k at time t. When the price of a plot and the distance between a plot and the nearest firm increase, the utility of plot j diminishes. On the other hand, as the quality of plot j increases, the utility score of plot j increases. The balancing parameters of each variable are set to 0.5 in the simulation. In other words, the resident agent values price, quality, and distance equally. All else equal, residents have a higher demand for plots with a higher level of occupancy nearby at time t (Torrens and Narra, 2007).

The environmental quality of plots is influenced by their proximity to firms. Environmental-disamenity-producing firms negatively affect plot quality, whereas firm agents modeled as producing little pollution have positive effects. This latter incorporates the sometimes-ignored point that communities desire job-producing businesses. The positive or negative effects of firms decay with distance within two concentric rings2 from the plot; for example, plots adjacent to disamenity-producing firms experience quality degradation more severely than plots further from them. The positive effects of environmental amenities also diffuse in the same manner, increasing the quality of nearby plots.

Agents and Decision Rules

Firm agents have two attributes, a number of jobs they provide and a pollution level. For simplicity, it is assumed that firms provide an equal number of jobs and the nature of those jobs is the same. Currently, a firm in the simulation “hires” 100 residents at maximum. When the number of residents in the world exceeds the sum of jobs that existing firms are providing, a new firm is established. Each new firm is generated with a pollution level that is randomly assigned within a range from 0 (no pollution) to 9 (highest pollution). The pollution level is drawn from a uniform distribution. When the pollution level of the new firm is strictly greater than 5, that firm is defined as a disamenity. Otherwise, the new firm is defined as an amenity in the simulation. When they are established, disamenities pollute their neighborhoods. Disamenities with higher pollution levels more severely degrade the quality of neighbor plots. We use a strict cut-off point in order to ensure that the majority of firms created are not high polluters; under the uniform distribution approximately 60 percent of the created firms should be low polluting and 40 percent high polluting.

While the first firm is generated at the center of the simulation world, subsequent firms’ siting decisions are informed by three rules. First, firms are boundedly rational, and when a firm is newly established it considers randomly selected plots (not to exceed 100 plots) in which to apply subsequent decision rules. Second, there is an agglomeration benefit to firms (Wolverton, 2009). Amenities prefer to be within 20 concentric rings of other amenities, and disamenities prefer to be within 20 concentric rings of disamenities. If there are plots that meet this preference, the new firm will limit its choice to those plots. If there are no plots that satisfy the agglomeration criterion, the firm selects a plot according to the third set of decision rules. Third, amenities always choose plots based on the lowest price, but disamenities randomly make choices based on one of three different scenarios: choosing a plot with the lowest price (i.e., neoclassical behavior), choosing a plot where a low proportion of majorities live because disamenities may want to stay away from the potential political action of residents (i.e., political behavior; cf. Hamilton, 1995), or choosing a plot where a high proportion of minorities live (i.e., discriminatory behavior).

Each resident agent is defined as either a majority (MA) or minority (MI). Residents are assumed to be the same in all aspects other than “race,” incorporating an assumption of homogenous income/wealth. It is well established that income variation is an important explanatory variable for EJ (e.g., Kriesel, Centner, and Keeler, 1996; Been and Gupta, 1997), but also that it is insufficient to fully explain environmental disparities by race (Ringquist, 2005). Since our interest in this research is better understanding the independent effect of residential constraints on those in the minority, we effectively hold income constant for all resident agents. While a simple majority/minority distinction does not encompass the full complexity of socioeconomic status, making simple distinctions enables us to model the independent effects of the modeled factors.

The goal of residents is to find a plot that satisfies their location decision criteria. A resident's decision is constrained by the similarity preference regarding neighborhood racial composition, and a choice is then made based on the utility score of the plots considered (using Equation (2)). For the present experiments, we follow Clark's findings (1992) and set the majority's similarity preference at 80 percent but vary the minority's preference from the same level (80 percent) to lower levels (50 percent or 20 percent). In other words, the majority only considers locating on plots where at least 80 percent of the same type of agents exists in the surrounding 12 plots within two concentric rings from a plot. If there is no such plot, the majority picks a plot with the highest utility score within the full decision set. However, minorities can have different similarity preferences and will consider locating in plots with 80 percent minorities, 50 percent minorities, or 20 percent minorities in the surrounding 12 plots, depending on the experimental scenario. Table 1 summarizes the parameterization of the model and different experimental conditions: varying the minority similarity preference, the ratio of majority to minority (parity levels), the growth rate, and the region size.

Table 1. Parameterization of the Model and Experiment Scenarios
ParametersBase Model Value [Range]Experimental Conditions
Random seedVaries 
Initial price and quality of plots50 [1, 100] 
 (uniform distribution)  
Initial number of firms1 
Pollution level (uniform distribution)[0, 9] 
Initial number of residents50 
Maximum number of employees100 
 per firm  
Utility score[0, 1] 
Vacancy rate[0, 1] 
Balancing parameters (α,β,γ)0.5 
[Exp. 1] Minority similarity80%50%, 20%
 preference when the majority  
 similarity preference is fixed  
 at 80%  
[Exp. 2] Racial composition:70% MA/30% MI55% MA/45% MI,
 majority/minority 90% MA/10% MI
[Exp. 3] Growth rate5%1%
[Exp. 4] Region size51 × 51101 × 101
 (2,601 plots)(10,201 plots)


  1. Top of page
  2. Abstract
  3. What Can Constrain Residential Choice Sets?
  4. An Agent-Based EJ Model
  5. Analysis
  6. Discussion and Conclusion

The simulation data include the outcomes of 108 different scenarios (2 region sizes × 2 growth rates × 3 racial parity scenarios × 3 disamenity decision scenarios × 3 minority similarity preference scenarios). The simulation was run 100 times per scenario with different random initial residential set ups. Each scenario stops after 60 time-steps, which we found to be long enough for noticeable patterns to emerge in the simulation. Also, with the growth rate of 5 percent, the base simulation world is reasonably occupied with residents and firms. For the model outcome, we tracked the average aggregate level of quality for majority plots and minority plots. In Figures 1 and 2 we focus on a subset of results that indicate the general pattern of outcomes.


Figure 1. Quality Comparison Between the Majority and Minority at t = 60

Note: For each box plot, the first number (e.g., 1 percent) tells the growth rate; the second number (e.g., 55) tells the starting percentage of the majority, and the third number (e.g., 20) tells the minority's similarity preference during that scenario. The top row represents slow-growth scenarios (1 percent), while the bottom row is for faster growth scenarios (5 percent). For all these data region size is 2,601, the smaller region, and disamenities are assumed to behave in three different ways—seek for a lowest priced plot, a plot near minorities, a plot such that a firm can stay away from majorities.

Download figure to PowerPoint


Figure 2. Dynamics of Quality Gaps for the Three Parity-Level Scenarios

Note: Numbers (e.g., 20 percent) indicate the MI similarity preference. Words (e.g., “low price”) indicate the disamenity firm's location criterion. Different lines (solid, dashed, or dotted) indicate the level of MA/MI disparity.

Download figure to PowerPoint

Figure 1 presents snapshots of the average quality levels for MA and MI at the last simulation step under various combinations of growth rate, majority/minority proportion, and the MI similarity preference (all of these hold region size constant at 2,601 plots, and disamenity firm behavior randomly distributed in three different ways as mentioned before). Each box plot has a heading with three numerical values: the first number represents the growth rate, the second number indicates the starting percent of MA in terms of racial composition, and the last number shows MIs’ residential similarity preference. In all cases, MA similarity preference is fixed at 80 percent.

Using Figure 1, we can first compare the environmental quality level for MA and MI under two different growth rate scenarios. Comparing the box plots in the top and bottom rows, the median quality level for both groups is noticeably lower with a 5 percent growth rate (bottom). The quality level remains around 50, whereas it stays around 55 when a 1 percent growth rate is used. On average, both resident groups are better off in a slowly growing region in terms of residential environmental quality. However, the variation in environmental quality, both within and between groups, is larger in the slowly growing region than in the rapidly growing region.

Next we compare the environmental quality level under the three different MA/MI composition scenarios. The first six box plots, three each from the top and bottom rows on the left, show the quality levels when the racial composition for MA and MI are 55 percent and 45 percent, respectively. Here, the numerical minority is close to parity with the majority. The six box plots in the top and bottom rows of the middle columns show the environmental quality for MA and MI when the “racial” composition is 70 percent and 30 percent, respectively (the base case). The last six box plots on the right show the results with 90 percent MA and 10 percent MI. Thus, in the last set there is extremely high numerical disparity in terms of the racial composition. Overall, the environmental quality for MI in the disparity scenarios is relatively lower than that of MI in the parity scenarios, holding constant the growth rate and similarity preferences.

In each MA/MI scenario, three different numerical identifiers show the relaxation of the similarity preference for MI. For example, the first three box plots in the top row show the environmental quality levels with a similarity preference of 20 percent, 50 percent, or 80 percent for MIs. The MI group is better off when relaxing the residential similarity preference in the location choice as long as they are not in the extreme (90 percent/10 percent) racial disparity scenarios. In the near-parity scenarios (55 percent /45 percent), MIs enjoy higher environmental quality by reducing the similarity preference to 50 percent or 20 percent. In the high-disparity scenario (90 percent /10 percent), the relaxation of the similarity preference from 20 percent to 50 percent does not improve the environmental quality for MIs. When MIs hold a similarity preference of 80 percent, they seem to be worse off in every racial composition and growth rate scenario. These results support a hypothesis (mentioned above) that actual minority groups may have lower similarity preferences in order to increase their residential options along other dimensions.

We showed above how the overall quality levels may vary when the similarity preference of the minority is relaxed (i.e., 50 percent or 20 percent) and the majority holds a strong similarity preference (i.e., 80 percent). In the analysis of the dynamics of environmental quality levels under differential similarity preferences, there are two scenarios when the median quality for MIs is at a higher level than that of MAs at the end of the simulation. In a slow-growing region, if MIs relax their similarity preference to 20 percent or 50 percent with MIs no less than 30 percent of the population (and if disamenities do not have a discriminatory motive in their siting decisions), MIs experience at least the same level of environmental quality as MAs—and their quality may be higher.

For Figure 2 we transform the outcome variable to a direct measure of the environmental quality gap between MAs and MIs. The quality gap is calculated by subtracting the average quality level of the minority from the average level of the majority. A positive number indicates that the majority is relatively better off and by how much, while a negative number indicates that the minority is relatively better off and by how much. Zero indicates that there is no difference in the quality level experienced by the two groups.

Figure 2 presents quality-gap dynamics under different MA/MI composition scenarios and breaks out the effects of the disamenity firm's decision criteria. The lines represent the quality gap between MA and MI over the 60 steps of each scenario. The solid lines show the scenario of 70 percent MA and 30 percent MI. Dashed lines show the quality gap under the racial near-parity scenario: 55 percent MA versus 45 percent MI. Dotted lines show the high-disparity scenario: 90 percent MA versus 10 percent MI. The three columns indicate whether disamenities decide where to locate based on finding the lowest price (“low price”), staying away from the majority (“fewest MA”), or targeting the minority (“most MI”).

First, consider Figure 2(g)–(i), in the third row. Both MA and MI hold strong similarity preferences (80 percent), but racial compositions differ across the lines. All three quality-gap lines end at greater than zero in the three charts, showing that MAs are better off regardless of disamenities’ siting decision scenarios. Further, relative MA/MI composition does not matter much since quality-gap lines remain at very similar levels in each. The numerical near-parity of groups does not appear to alleviate the environmental quality gap between different groups.

Figure 2(a)–(f) shows the same experiment with the MIs’ similarity preferences relaxed to 20 percent (top row) or 50 percent (second row). Here, there is no clear pattern as to what level of MA/MI disparity is best for minorities, supporting our view that there are complex interactions at work. What we can note is that if disamenity firms target minorities (third column), no level of parity or disparity and no level of relaxed MI similarity preference can prevent MIs from doing much worse than MAs. In the other two types of disamenity behavior, MIs can occasionally be better off (we see some lines ending below zero).

Lastly, in Table 2 we compare the relative importance of each scenario in explaining the quality gap observed in the entire set of simulations. Instead of separately examining the effect of residential conditions—growth rate and region size—on the environmental quality gap, here we investigate them together with firms’ and residents’ decision scenarios. We created a data set including the mean value of the quality gap for each of the 108 scenarios. At the last step of the simulation, the mean quality gap was 0.48 with a standard deviation of 1.09. The minimum value was −2.48 and the maximum was 5.61.

Table 2. OLS Results, DV is Quality Gap at t = 60
 CoefficientStd. Err.p-value
Disamenities: fewest majority nearby0.100.180.581
Disamenities: most minority nearby1.280.180.000
[Exp. 1]   
MIs: 50% similarity preference0.330.180.071
MIs: 80% similarity preference0.380.180.035
[Exp. 2]   
70% MA/30% MI0.400.180.029
90% MA/10% MI0.860.180.000
[Exp. 3] Growth rate: 5%0.640.150.000
[Exp. 4] Region: large−0.410.150.006
Disamenities established (%)
N = 108   
F(9,98) = 13.3   
Adj R2 = 0.51   

If disamenities locate their facilities near areas mostly composed of minorities, then the quality gap between the two social groups increases as compared with the situation where disamenities pursue the lowest cost as the siting decision criterion, even when other factors are held constant (Table 2). MAs are much better off in this scenario. When MIs hold a similarity preference at the same level as MAs, the gap is significantly larger than when MIs have a lower similarity preference. As the proportional size of the racial disparity increases, the quality gap increases. Finally, the quality gap is higher in a rapidly growing versus slow-growth region, and the gap is lower in a larger region than in a smaller region. The first set of results support the traditional idea that if polluting organizations target minorities they will be worse off. But the others, taken together, indicate that environmental injustice can be seen without this discriminatory motive, and that in general residential choice constraints harm minorities. The one factor that minorities can readily control is their level of similarity preference, which may explain the observation of lower similarity preferences among minorities.

Discussion and Conclusion

  1. Top of page
  2. Abstract
  3. What Can Constrain Residential Choice Sets?
  4. An Agent-Based EJ Model
  5. Analysis
  6. Discussion and Conclusion

This article examines how differential experiences of environmental quality (or quality gaps) between social groups arise from varying spatial constraints. The focal point of this research is constraints that may limit residential choice sets. Overall, the simulation experiments show that location choice constraints could, in part, explain observed real-world environmental quality variations. In the experiments, when minorities lower their similarity preference (reducing their choice constraints), they enjoy environmental quality at a level similar to the majority. For MIs, holding a high similarity preference has a negative relationship with living in an area with high environmental quality. Racial composition and residential conditions can also constrain or expand residential choice sets for the minority, contributing to different levels of exposure to environmental harm. In the simulation, combinations of these variables explain differences in environmental risk exposure between different social groups.

We found that even in the absence of any nefarious intent by firms, minorities tend to be worse off when (1) minority residents hold high similarity preferences, (2) there is high numeric disparity between the two social groups, (3) there is a high regional growth rate, and (4) the region is small or otherwise geographically constrained. Each of these situations reduces the ability of minorities to find plots that suit their preferences, and each generally results in worse environmental outcomes for minorities. The interactions between these several factors (which are certainly not all of the relevant factors) may help explain why the EJ literature contains such different findings for different regions. Empirical EJ studies have not considered growth rate, similarity preferences, geographic constraints, and so forth.

A complication worth noting is that we may be confounding two different issues when we discuss similarity preferences. Similarity preferences are just one social structure that may lead to segregated communities, and may not be the most prominent one. While we have expanded the minority choice set by decreasing the minority residents’ similarity preference, we might also have modeled this dynamic as a lessening of exclusionary practices by majorities. It is worth considering whether we would find similar results if majorities relax their exclusionary practices. However, in practice it may be much more complex to craft a policy to encourage majorities to reduce their exclusionary preferences, than to encourage minorities to decrease their similarity preference.

The finding with respect to racial compositions also raises interesting questions for EJ research. Going beyond mere collocation of minorities and environmental disamenities, Arora and Cason (1998) state that “a larger percentage of non-white residents may be associated with a higher level of releases in the southeastern states” (1998:416). Certainly, “a larger percentage of non-white residents” can mean different situations such as those modeled here as different MA/MI relationships. Should EJ research pay attention to the relative racial composition of residents in neighborhoods?

Also, an interesting empirical result in Campbell, Peck, and Tschudi (2010) is that, controlling for many other factors, the presence of Asian minority groups is associated with increases in new Toxic Release Inventory Facilities in Arizona's Maricopa County (which contains Phoenix, the fifth largest city in the United States). Given the history of this southwestern area, this seems surprising, but it may be less so if the issue is locational constraints on this small minority group rather than racism by firms directed against Asians. In general, some of the surprising findings with respect to certain minority groups’ exposure to environmental risk may need to be reexamined along with residential similarity preferences and other minority choice constraints.

For simplicity and control, the simulation model was designed to examine the role of race without confounding with income. Empirically, race is highly correlated with income, and others have reported that poverty may be an explanatory variable for environmental injustice. We do not discount the explanatory power of income variability on EJ outcomes; however, we do note that income is insufficient to explain the EJ problem with respect to race (Ringquist, 2005). Our model helpfully explores minority-based social processes in a homogenous wealth framework, but we acknowledge that the role of poverty versus race on EJ outcomes requires thorough examination in future analysis. A further limit of this study is that there was no residential mobility. Residents “die,” which opens up old plots for new residents, but residents do not relocate during the simulation. Given that Americans exhibit residential mobility (even if at declining rates), it is also worth considering how the ability to move to find a better plot might affect EJ outcomes.

In conclusion, environmental injustice occurs in the context of a dynamic social system. When firms are overtly discriminatory in their siting decisions, an environmental quality gap can emerge between majority and minority residents. However, the same outcome can occur when minority residents have a strong preference to live in communities with a high proportion of other minority residents, when a region is growing rapidly, or when there are geographical barriers to outward expansion—in short, when residential choices are constrained. In one way, this complicates the issue: it is no longer sufficient to assign specific blame for environmental injustice. There may be instances where lines of fault are clear, but the results of our simulations indicate that the causes of EJ are likely to be much more complex than the simple behavior of one actor, and therefore useful policy interventions are likely to be more complex. These results enable us to look at the problem of race-based environmental injustice more holistically, and begin to think about holistic solutions that may finally address what has heretofore been an intractable social problem.

  1. 1

    This immobility constraint is worth consideration. In both standard Tiebout and Schelling models, relocation costs are assumed to be zero, so agents are free to move at any time. We changed this assumption, implicitly imposing high relocation costs such that once agents select a location, they remain. The United States is more mobile than many other societies, and even in the United States mobility rates are declining, so we found this more realistic than full mobility. As a check, the model was tested with limited mobility based on a “satisficing threshold,” with similar results to the immobility assumption. We retain the latter model for the sake of parsimony.

  2. 2

    In the ABM toolkit used (NetLogo), the surrounding area was defined by a primitive, in-radius (which we refer to as a “concentric ring” here). This primitive returns only those plots whose distance from the caller is less than or equal to number. Therefore, “two concentric rings” include plots whose distance from the center plot less than or equal to 2 (i.e., 13 neighboring plots, including the caller).


  1. Top of page
  2. Abstract
  3. What Can Constrain Residential Choice Sets?
  4. An Agent-Based EJ Model
  5. Analysis
  6. Discussion and Conclusion
  • Arora, S., and T. N. Cason. 1998. “Do Community Characteristics Influence Environmental Outcomes? Evidence from the Toxic Release Inventory.” Journal of Applied Economics 1(2): 41353.
  • Baden, B. M., D. S. Noonan, and R. M. Turaga. 2007. “Scales of Justice: Is There a Geographic Bias in Environmental Equity Analysis?” Journal of Environmental Planning and Management 50(2): 16385.
  • Banzhaf, H. S., and R. P. Walsh.2008. “Do People Vote with Their Feet?: An Empirical Test of Tiebout's Mechanism.” American Economic Review 98(3):84363.
  • Banzhaf, H. S., and R. P. Walsh. 2010. “Segregation and Tiebout Sorting: Investigating the Link Between Investments in Public Goods and Neighborhood Tipping.” NBER Working Paper Series 16057.
  • Been, V., and F. Gupta. 1997. “Coming to the Nuisance or Going to the Barrios? A Longitudinal Analysis of Environmental Justice Claims.” Ecology Law Quarterly 24(1):156.
  • Bowen, W., M. Atlas, and S. Lee. 2009. “Industrial Agglomeration and the Regional Scientific Explanation of Perceived Environmental Injustice.” Annals of Regional Science 43:101331.
  • Bullard, R. 1990. Dumping in Dixie: Race, Class, and Environmental Quality. Boulder, CO: Westview Press.
  • Bullard, R.. 1996. “Environmental Justice: It's More Than Waste Facility Siting.” Social Science Quarterly 77:49399.
  • Campbell, H. E., L. R. Peck, and M. K. Tschudi. 2010. “Justice for All? A Cross-Time Analysis of Toxics Release Inventory Facility Location.” Review of Policy Research 27(1):125.
  • Chun, Y., Y. Kim, and H. Campbell. 2012. “Using Bayesian Methods to Control for Spatial Autocorrelation in Environmental Justice Research: An Illustration Using Toxics Release Inventory Data for a Sunbelt County.” Journal of Urban Affairs 34(4):41939.
  • Clark, W. 1992. “Residential Preferences and Residential Choices in a Multiethnic Context.” Demography 29(3):45166.
  • Eckerd, A., H. E. Campbell, and Y. Kim. 2012. “Helping Those Like Us or Harming Those Unlike Us: Illuminating Social Processes Leading to Environmental Injustice.” Environment and Planning B 39:94570.
  • Epple, D., R. Filimon, and T. Romer. 1984. “Equilibrium Among Local Jurisdictions: Toward an Integrated Approach of Voting and Residential Choice.” Journal of Public Economics 24:281308.
  • Hamilton, J. T. 1995. “Testing for Environmental Racism: Prejudice, Profits, Political Power?” Journal of Policy Analysis and Management 14(1):10732.
  • Kriesel, W., T. J. Centner, and A. G. Keeler. 1996. “Neighborhood Exposure to Toxic Releases: Are There Racial Inequities?” Growth and Change 27(4):47999.
  • Mohai, P., D. Pellow, and J. T. Roberts. 2009. “Environmental Justice.” Annual Review of Environment and Resources 34:40530.
  • Noonan, D. S. 2008. “Evidence of Environmental Justice: A Critical Perspective on the Practice of EJ Research and Lessons for Policy Design.” Social Science Quarterly 89(5):115374.
  • Pastor, M., J. Sadd, and J. Hipp. 2001. “Which Came First? Toxic Facilities, Minority Move-In, and Environmental Justice.” Journal of Urban Affairs 23(1):121.
  • Ringquist, E. J. 2005. “Assessing Evidence of Environmental Inequities: A Meta-Analysis.” Journal of Policy Analysis and Management 24(2):22347.
  • Schelling, T. C. 1978. Micromotives and Macrobehavior. New York: W.W. Norton & Company.
  • Tiebout, C. 1956. “A Pure Theory of Local Expenditures.” Journal of Political Economy 64:41624.
  • Torrens, P., and A. Nara. 2007. “Modeling Gentrification Dynamics: A Hybrid Approach.” Computers, Environment and Urban Systems 31:33761.
  • Wolverton, A. 2009. “Effects of Socio-Economic and Input-Related Factors on Polluting Plants’ Location Decisions.” B.E. Journal of Economic Analysis & Policy 9(1). DOI: 10.2202/1935-1682.2083