Racial and Ethnic Residential Segregation in Nonmetropolitan Communities

Abstract

For at least half a century, much of the literature on residential segregation has primarily focused on large metropolitan areas, where most of the population resides in one or more high-density urban cores and medium-density, outlying suburban environments. Many influential landmark segregation studies have focused on small samples featuring primarily the largest 50–60 metropolitan areas in the country. In contrast, our knowledge of residential segregation outside of metropolitan contexts is very limited, even as nonmetropolitan communities are becoming more racially and ethnically diverse. This is due in large part to the challenges with measuring residential segregation in nonmetropolitan contexts, which we address in this book. By using segregation measures that are corrected for index bias, measuring segregation of households rather than persons, and relying on the often overlooked separation index, we draw substantive conclusions about patterns and trends of White-Black, White-Asian, and White-Latino residential segregation in nonmetropolitan communities from 1990 to 2010. We also further demonstrate the superiority of our methodological choices by comparing our findings to those that would be produced through conventional approaches and by reviewing case studies of selected nonmetropolitan areas. Substantively, we find that segregation in nonmetropolitan communities is often not as high as what is observed in metropolitan areas, especially for Asian and Latino households. However, segregation is certainly capable of being high in nonmetropolitan communities, even when the minority proportion is very small. White-Latino segregation is typically quite low and has remained stable in micropolitan areas while slightly declining in noncore counties. This is more so true for White-Asian segregation in nonmetropolitan communities where segregation is very low and has remained low since at least 1990. Methodologically, we conclude that our innovations to segregation measurement expand opportunities to broaden and deepen the literature to understand the nature of residential segregation in nonmetropolitan communities.

An earlier version of this chapter was prepared and presented as a conference paper for the 2022 Annual Meetings of the Population Association of America

4.1 Overview

For at least half a century, much of the literature on residential segregation has primarily focused on large metropolitan areas, where most of the population resides in one or more high-density urban cores and medium-density, outlying suburban environments.¹ Indeed, many influential landmark segregation studies such as Duncan and Duncan (1955) and Massey and Denton (1988) focused on small samples featuring primarily the largest 50–60 metropolitan areas in the country. More recent benchmark analyses of broad trends in segregation patterns across the United States have used expanded analysis datasets that include a broader set of metropolitan areas (Frey, 2018; Iceland, 2014; Logan & Stults, 2011), but studies often report summary statistics for segregation indices with cases weighted using criteria that give disproportionate influence to the very largest metropolitan areas. Analysis datasets often exclude smaller metropolitan areas and rarely include nonmetropolitan communities of any kind. The heavy focus on segregation in metropolitan settings is in part a matter of tradition, with studies of urban residential patterns dating back to the earliest days of American sociology through the work of scholars like W.E.B. Du Bois (1899) along with researchers at the Chicago School who developed their urban ecological theories by observing group settlement patterns across neighborhoods in Chicago (e.g., Park & Burgess, 1925). The tradition, therefore, is that segregation research and theorizing is centered on the urban contexts of the nation’s largest metropolitan areas. But even as the literature expanded in the late twentieth century, with revived sociological and demographic interest in residential segregation exemplified by Massey and Denton’s American Apartheid (1993) and calls by leading scholars to recognize the sociological importance of segregation in nonmetropolitan communities (Lichter & Brown, 2011), there has continued to be a hesitation to systematically analyze segregation in nonmetropolitan contexts.

The focus on racial and ethnic segregation in metropolitan areas has been and continues to be well justified. Large metropolitan centers are highly important, especially as the U.S. population has over time become increasingly concentrated in metropolitan communities. The metropolitan context is also changing through rising levels of spatial complexity due to suburban and exurban sprawl and steady trends of growing racial-ethnic diversification outpacing those seen in nonmetropolitan communities (Sharp & Lee, 2017). However, segregation in smaller metropolitan areas and nonmetropolitan communities is also highly relevant and important in its own right, despite historically receiving less attention and, in the case of nonmetropolitan communities, despite demographically losing population due to natural decrease and net population outmigration (at least, up until recently (Cromartie & Vilorio, 2019)). It is understandable that the striking and compelling patterns of segregation observed in large exemplar metropolitan areas would receive outsized attention. But it is important to not lose sight of the fact that segregation is observed across a wide range of communities and fundamental questions regarding how levels, patterns, and trends in segregation vary across communities cannot be answered by analyses with a limited focus on large metropolitan areas which, while undeniably important, are not at all representative of the breadth of variation in communities across the United States.

It is valuable therefore to examine segregation in nonmetropolitan settings to consider implications for theories of segregation and avoid the risk that prevailing theories may be overly tailored to metropolitan environments. As only one example, consider the highly plausible and widely accepted “White flight” hypothesis that high levels of racial segregation exist in metropolitan areas because higher-status White households gravitate to suburban settings that are predominantly White and higher-status rather than effort to remain in neighborhoods that are racially transitioning (Frey, 1979; Massey & Denton, 1993). Because these suburban neighborhoods they choose to occupy instead are spatially and administratively separated from the more diverse neighborhoods of the central city and surrounding suburbs that have racially transitioned (Kye, 2018), suburban White households simultaneously pay lower tax rates and enjoy attractive location-based amenities including, most notably, higher quality schools for their children. It is undeniable that White households residing in many suburban settings benefit from these consequences of segregation. But, if the hypothesis has identified a broad and powerful driver of segregation, it could be seen as implying a prediction that segregation would lower in nonmetropolitan settings because residential separation does not lead to racial segregation in public schools as, at least since the 1970s, public schools are desegregated and it is common for nonmetropolitan communities to have a single campus for high school and also at lower levels. Thus, residential segregation may not confer educational advantages to middle-class White households in nonmetropolitan settings. And, relatedly, White households whose children attend non-public schools, a strongly emergent pattern for White families after 1970 (Cready & Fossett, 1998), do not need to be residentially separated to achieve exclusivity and advantage in schooling. If, in fact, racial segregation is observed to be significantly lower in nonmetropolitan settings, it constitutes evidence consistent with the “White flight” hypothesis. But, if segregation is equally high or even higher in nonmetropolitan settings, it raises questions regarding what the fundamental drivers of segregation are. One possibility would be that different, but equally powerful, dynamics of segregation exist in both metropolitan and nonmetropolitan settings. Another possibility is that similar dynamics drive segregation in both settings and researchers need to refine theories to acknowledge the commonality.

Racial and ethnic segregation in nonmetropolitan communities particularly warrants greater attention in recent years due to the striking demographic shifts occurring in nonmetropolitan communities. Lichter and Brown (2011) argue that rural areas are often overlooked and misunderstood as socially isolated from the dynamics of urban contexts, when in fact they should be seen as increasingly interdependent with metropolitan areas, particularly due to migration patterns that have diversified the nonmetropolitan United States (Winkler & Johnson, 2016). While these areas have been characterized over the last few decades by stagnant White population growth or even decline, an opposing force has offset this trend: the migration of minoritized racial and ethnic groups to nonmetropolitan communities (Johnson, 2006; Lichter, 2012; Lichter et al., 2018; Sharp & Lee, 2017; Winkler & Johnson, 2016). Latino migrants are primarily driving this trend, but Asian presence is also substantial in some nonmetropolitan communities and can be anticipated to grow (Sharp & Lee, 2017). Additionally, many rural communities in the South have long been home to a significant number of Black residents whose history is tied to the South’s agricultural economy which relied first on enslaved Black people and later on Black sharecroppers to perform most of the cultivation and production labor. Adding to this is a reversal of the Great Migration to Northern urban areas that characterized the mid-twentieth century, with Southern areas seeing a new surge of Black migrants (Hunt et al., 2008, 2013) with evidence of long-term settlement (DeWaard et al., 2016). Growing minoritized racial populations in nonmetropolitan communities combined with White population decline can lead to what appears to be growing nonmetropolitan diversity, although Lichter et al. (2018) argue that these forces do not necessarily create conditions of integration or harmonizing race relations if White residents exit these communities in a traditional “White flight” dynamic. The need to focus on residential segregation in nonmetropolitan settings is apparent, and we shift our attention to nonmetropolitan communities in this chapter through an application of the methodological innovations in segregation measurement that motivate this book.

4.2 Challenges for Nonmetropolitan Residential Segregation Research

Segregation in nonmetropolitan settings has received less attention in empirical studies of the past not due to lack of interest on the part of researchers but primarily because measuring segregation in nonmetropolitan settings involves significant challenges. One major problem has been limitations of data availability for spatial units appropriate for measuring segregation in nonmetropolitan communities. These problems have been overcome in recent decades as the U.S. Census Bureau achieved full block-level coverage of the United States, including all nonmetropolitan counties, in 1990, and from that time has distributed summary file tabulations of racial-ethnic distributions at the block-level along with related tabulations by age, sex, and other key demographic variables. Consequently, the literature has witnessed an increase in studies on segregation in nonmetropolitan communities starting in the early 2000s, with particularly important contributions by Daniel Lichter and colleagues who called for segregation researchers to devote more attention to segregation in small-towns and rural communities and provided exemplars of how such research can be undertaken (Lichter et al., 2007a). We endorse Lichter’s observation that “Rural minority populations are spatially segregated and invisible in ways not usually found in America’s metropolitan areas with large and densely settled inner-city minority populations” (2012: 4) and we also endorse his arguments that these patterns are compelling and justify the view that more scholarly attention be given to the often overlooked minoritized racial populations of the nonmetropolitan United States.

Studies by Lichter and colleagues and by others have indeed brought needed attention to the residential patterns of rural areas. This is especially welcome because increased attention is occurring at a critical point when the demographic composition of many nonmetropolitan communities has become more diverse, often changing in dramatic ways in comparison with earlier times when their racial-ethnic composition was more homogenous (Sharp & Lee, 2017). But, despite these welcome developments, our knowledge of nonmetropolitan residential segregation, including how it compares to metropolitan segregation and why it matters, remains incomplete. To emphasize why we may be missing something important about understanding the origins and dynamics of residential segregation, Lichter and Brown (2011), in their review article of the rural United States in relation to our national focus on urban contexts, concluded that there is a “blurring” of rural-urban spatial boundaries which “ironically…has been accompanied by the hardening of aspatial boundaries (e.g., race and class)” (Lichter & Brown, 2011: 584). If indeed the boundaries of race and class are solidifying in nonmetropolitan contexts, then we must investigate the spatial boundaries within nonmetropolitan communities for evidence of patterns of segregation that often accompany intensifying racial divisions. However, many of the studies that have attempted to expand our knowledge of nonmetropolitan segregation come with limitations and withholdings, bringing us to the primary reason why the literature remains so sparse.

While availability of relevant data has improved significantly, research on nonmetropolitan residential segregation has faced a second major challenge in measuring segregation. This is that standard approaches to measuring segregation can and often do lead to misleadingly high scores under conditions that are common in rural communities and nonmetropolitan communities with small populations. Namely, measuring segregation in nonmetropolitan communities includes having to measure segregation using data for small spatial units when groups vary widely in relative size across communities. Either condition presents a major practical problem and together the problems are compounded. Studies of segregation in nonmetropolitan settings have until recently had to take one of two paths for dealing with these practical problems. One path is to carry over practices used in studies of segregation in metropolitan areas with minimal changes, with the consequence that analysis samples are small and nonrepresentative. The other is to modify practices used in earlier studies of metropolitan areas to achieve larger, more representative analysis samples, but with the consequence that index scores are more susceptible to being distorted by index bias.

Some studies of segregation in nonmetropolitan communities have, with only minor adjustments, adopted the methodological practices used in studies of segregation in metropolitan areas. In these cases, communities are screened for inclusion in the analysis based on highly restrictive minimum population thresholds and sometimes at a level needed to sustain segregation measurement using census tracts as neighborhoods – often out of not unfounded wariness of problems associated with measuring segregation using block-level data with standard segregation indices (Fossett, 2017). For example, Byerly’s (2019) study of American Indian and Alaska Native (AIAN) segregation used an area-level sample restricted to metropolitan and micropolitan areas where there were at least 1000 single-race AIAN individuals and 1000 multiracial AIAN individuals with segregation measured at the census tract level. At this end of the spectrum of methodological choices, the sample restrictions adopted have undesirable consequences of distorting our understanding of nonmetropolitan segregation due to limiting attention to a small and decidedly nonrepresentative set of communities that are larger in size and to group comparisons where both groups are larger in both absolute and relative size. In particular, case restrictions that apply high minimum population thresholds for the groups in the comparison exclude a large swath of nonmetropolitan communities where minoritized racial populations are newly emerging and potentially impacting residential distributions, thus precluding the opportunity to directly observe how segregation patterns initially form in small communities and change as new groups grow in absolute and relative size.

This is both concerning and ironic because we find that the restrictions do not necessarily lead to more effective measurements of segregation. Measuring segregation at the census tract level in nonmetropolitan communities may screen cases in a way that reduces the impact of index bias, but it carries an unwelcome consequence of systematically underestimating the level of segregation in nonmetropolitan settings because census tracts are too large to capture patterns of segregation as they occur in smaller communities; specifically, they obscure clear patterns of segregation that occur across smaller spatial units such as census blocks by combining blocks that differ on racial composition into much larger census tracts that then misleadingly appear to be substantially integrated. This is not necessarily a problem in large metropolitan areas where individual tracts contain a small share of the population in the community and, due to clustering dynamics, segregation typically is manifest in patterns that can be captured by tracts. But these conditions do not necessarily hold in smaller metropolitan areas and they certainly do not hold in nonmetropolitan settings.

To avoid the problem of having small non-representative samples where segregation is systematically underestimated in smaller communities, researchers must consider the alternative path of modifying practices used in studies of segregation in large metropolitan areas. Relaxing sample restrictions to use lower screening thresholds on absolute and relative group size will yield larger, more representative samples. Measuring segregation using block-level data will capture segregation in smaller communities as well as in larger communities. But adopting these changes leads to increased risk that scores obtained using standard formulas for calculating values of segregation indices will be inflated by index bias that varies in magnitude across communities and is especially high in communities where new groups are small in absolute and relative size. Lichter et al. (2007a) elected to measure segregation using block-level data because they would otherwise not be able to adequately detect the sort of small-scale segregation that occurs in small towns and rural communities. But it required accepting the risk that segregation index scores were potentially distorted by index bias.

We recognize previous researchers have faced difficult choices and sympathize with their dilemmas. One of our major goals for this book is to identify and use strategies for measuring segregation more accurately and appropriately, especially in noncore and micropolitan communities and in new destination communities (see Chap. 5) where groups may be small in absolute and/or relative size. On this point we bring welcome news. Specifically, new developments in methods for measuring segregation have introduced solutions that overcome these longstanding problems in measuring segregation in nonmetropolitan communities. Adopting these new methods enables us, and other researchers, to conduct the most inclusive and precise analysis of nonmetropolitan residential segregation to date and set accurate benchmarks and methodological guidelines for future analyses.

4.3 Segregation in Nonmetropolitan Communities: What We Know, and What We Question

The research over the past few decades on segregation in nonmetropolitan communities has been valuable, but also limited and somewhat inconsistent. Hwang and Murdock (1983) produced some of the earliest research in this area, examining segregation in nonmetropolitan communities and metropolitan areas of Texas and finding that segregation was highest in nonmetropolitan communities that were not adjacent to metropolitan areas. In a subsequent study, Murdock et al. (1994) noted that there had been very few attempts to study and understand segregation in nonmetropolitan communities, making it difficult to answer even the most basic questions about the nature of segregation in nonmetropolitan contexts or draw out comparisons with metropolitan areas where patterns of segregation were better understood. They tried to address this gap in the literature by examining block-level segregation in Texas cities but were not able to cover all nonmetropolitan communities due to the limited coverage of block-level census tabulations at the time. Nearly 30 years later, their initial observation on the state of the literature still holds mostly true with the notable exception of significant contributions by a few research teams. Despite much better census data coverage and public-use data availability, research on segregation in nonmetropolitan communities remains limited. The literature, while growing over the last decade, is still far from comprehensive or definitive.

Studies that have sought to describe patterns and trends of segregation in nonmetropolitan communities have at times offered inconsistent findings on whether racial segregation is higher or lower in nonmetropolitan communities compared to metropolitan areas. For example, while the majority of studies have argued that White-Latino segregation is higher in nonmetropolitan communities (Hwang & Murdock, 1983; Lichter et al., 2007a, 2010; Murdock et al., 1994), some studies have found White-Latino segregation in nonmetropolitan communities to be lower, including one notable study by Wahl et al. (2007) where White-Latino segregation in micropolitan areas was considerably lower than in metropolitan areas, on average. The literature is also conflicted on how segregation is changing in these areas over time and why Murdock et al.’s 1994 study found substantial White-Black segregation declines from 1980 to 1990 in both metropolitan areas and nonmetropolitan communities in Texas, with larger declines occurring in areas with population growth. Lichter and colleagues’ 2007 research reported similar findings in a national-level study. Both studies also reported findings that White-Latino segregation was declining as well (Lichter et al., 2007a; Murdock et al., 1994). Other studies, sometimes using non-standard approaches to segregation measurement (e.g. Logan & Parman, 2017), have found more varying trends over time.

Lichter et al. (2007a) have to date made the most comprehensive effort to measure segregation in smaller and rural communities with their analysis of place-based segregation, and thus we treat their research somewhat as the benchmark for this chapter. Their article importantly recognizes that segregation observed in nonmetropolitan communities is substantively meaningful and consequential. They also echo the observations of Murdock and colleagues 13 years prior – that there continues to be little social scientific interest in the residential patterns of nonmetropolitan communities. These researchers describe the social and demographic conditions that have existed in the nonmetropolitan United States which set the stage for segregation to rise in response to the demographic trend of steady nonwhite population growth in nonmetropolitan communities over the past three decades. These conditions include the persistence of residential patterns established during the Jim Crow era for Black households in the nonmetropolitan South and of the concentration of Native American households on tribal reservation lands (Lichter et al., 2007a), the history of informal and formal tactics of discrimination and violence that created and maintained all-White “sundown towns” above as well as below the Mason-Dixon line (Loewen, 2006), the lower socioeconomic standing of nonwhite groups moving to nonmetropolitan areas, the foreign born status and limited English-language ability of Latino immigrants, and pre-existing and persistent White racial intolerance (Lichter et al., 2007a).

Measuring segregation using the dissimilarity index (D), Lichter et al. (2007a) find that White-Black segregation overall is extremely high with levels in nonmetropolitan communities slightly higher than in metropolitan areas.² Allen and Turner (2012) similarly found that White-Black segregation was very high in nonmetropolitan communities. Lichter and colleagues additionally reported that White-Black segregation is declining, in a manner similar to that reported in studies documenting trends in metropolitan areas. High and declining levels of White-Black segregation may not be surprising, but what may come as a surprise is that the researchers also document moderate to high levels of White-Latino segregation in both metropolitan and nonmetropolitan communities. Indeed, when we use the standard dissimilarity index, we also find a sizable percentage of nonmetropolitan communities with medium to high scores for both White-Black and White-Latino segregation (Table 4.1). But, while it is useful and perhaps reassuring to replicate past findings, we caution against placing undue confidence in these particular results because the standard version of the dissimilarity index is far more likely to be affected by index bias in exactly these scenarios.

Table 4.1 Distribution of nonmetropolitan communities across low, moderate, and high levels of segregation, standard and unbiased dissimilarity index

When we compare scores for the standard and unbiased versions of the dissimilarity index, we find important differences for White-Black segregation. Not surprisingly, average index scores are lower, and more communities register low-to-moderate scores on the unbiased dissimilarity index. The reductions are especially large in nonmetropolitan communities with Black populations that are small in absolute and relative size.

We find the impact of bias is even larger and more concerning in the case of White-Latino segregation, where the contrast between the values of the standard and unbiased versions of the dissimilarity index is nothing short of dramatic. We find that 19 percent of nonmetropolitan communities in 1990 had moderate scores on the standard dissimilarity index and 80 percent had high scores when measuring White-Latino segregation. This distribution between moderate and high values of D is more pronounced than that reported by Lichter et al. (2007a), who found that 56 percent of areas had moderate scores and only 30 percent had high scores for White-Latino segregation. One reason for this is that our case selection criteria can be more inclusive and less restrictive as a direct benefit of using unbiased index scores. Thus, our analysis dataset includes more communities and the extra cases are ones that would have been excluded in early studies based on concerns that standard index scores were likely to be distorted by upward index bias.

One might understandably hope to find that the ad hoc strategies for dealing with the problem of index bias used in previous studies of segregation would be adequate in some sense when analyzing White-Latino segregation in nonmetropolitan communities. Unfortunately, this is not so. When we review scores obtained using the unbiased version of the dissimilarity index, we find a remarkably different distribution of communities along a low-to-high continuum of levels of segregation. Based on the unbiased dissimilarity index, 76 percent of nonmetropolitan areas have low scores, 22 percent have moderate scores, and only 1 percent have high scores. This distribution is completely opposite to what we observed using the standard version of the dissimilarity index and is very different from the patterns reported by Lichter et al. (2007a). The contrasts are clear and stark – scores for the standard version of D are inflated by index bias. The magnitude of bias varies in complex ways across cases, but it is never negligible. Instead, it ranges from moderate to severe and on average is high and thus shifts the distribution of scores to a fundamentally different range and pattern. These differences indicate that researchers have been right to worry about the impact of index bias on findings. New methods now make it possible to eliminate the impact of bias directly at the point of measurement so index scores can be examined and analyzed as is, removing concerns that individual scores and scores for particular kinds of communities cannot be trusted.

Our discussions of methods in Chap. 2 make the case in more detail. Here we briefly assert that the methods we use to deal with index bias are superior to any used in previous research with the most fundamental advantage being that all individual scores are accurate, valid, and free of bias as calculated and thus can be interpreted individually and compared across cases without concern for how findings might be distorted by bias. All previous strategies for dealing with index bias have necessarily worked with inherently flawed scores, with researchers attempting to minimize the impact of bias by excluding the most severely flawed cases and discounting less severely flawed cases based on screening and weighting variables that are presumed to be correlated with bias. By drawing on new methods (Fossett, 2017), we dispense with the need to use proxy correlates of bias to identify cases where standard scores are inflated by bias. We have direct estimates of bias based on the difference between the values of standard and unbiased scores for the same cases.

More importantly, having accurate unbiased scores wholly negates the need to unnecessarily exclude valid cases from the analysis and/or discount valid cases based on concerns about bias. Thus, as evidenced in Table 4.1, this also allows us to expand the study design to include a larger number of communities. When measuring segregation using block-level data, as is crucial in studies of nonmetropolitan communities, the problem of bias cannot be dealt with effectively by imposing selective restrictions on analysis samples. It can only be addressed by directly adjusting the index formula itself to eliminate bias at the point of measurement. Doing so produces substantially different results. Therefore, our knowledge about nonmetropolitan segregation, echoed in a more recent study by Lichter et al. (2016) where they again found that Latino segregation in nonmetropolitan communities is “exceptionally high” (Lichter et al., 2016: 512) with the dissimilarity index reaching scores as high as 60 – scores that we would also categorize as “very high” – must be reexamined and reconsidered in light of the different findings that emerge when measures are adjusted to eliminate the impact of index bias.

4.4 The Choice of Segregation Index for Nonmetropolitan Segregation Research

A major strength of our study is that we adopt a careful and nuanced approach to measuring segregation that is especially important for obtaining a more complete understanding of the nature of levels and trends in segregation in nonmetropolitan settings. In particular, we identify multiple methodological factors, including some that are not recognized in previous research, and we address them by using measurement strategies that are superior to those used in previous research on segregation in nonmetropolitan communities. The single most troublesome problem is the upward bias inherent in scores obtained using standard segregation index formulas. It is no exaggeration to characterize the problem as critical in studies of emerging segregation for new groups in nonmetropolitan communities. We argue, and present evidence to support our view, that findings based on measuring segregation in nonmetropolitan communities with scores obtained using the standard formula for the dissimilarity index should not be accepted at face value.

One might acknowledge the problem of index bias and yet have hope that certain findings regarding trends in segregation, variation in segregation across communities, and differences in levels of segregation across different group comparisons will nevertheless be unaffected. This welcome result would be possible in principle if bias inflated index scores in a uniform way across all circumstances. If so, one might acknowledge that scores are inflated by bias but could still be confident in findings that, for example, White-Black segregation is higher than White-Latino segregation in nonmetropolitan communities or that levels of segregation are declining over time. Unfortunately, we document that, in fact, this situation does not hold. The reason is both simple and devastating. The impact of bias on index scores in nonmetropolitan settings is far from uniform. It is sometimes small and sometimes very large, and the variation affects assessments of trends over time, variation across communities, and levels for different group comparisons. All of these problems are particularly pronounced for communities that are seeing sustained influxes of new groups, especially the many Latino new destination communities.

Index bias is not the only problem that causes us to reconsider and re-evaluate patterns of segregation documented in previous studies using only the dissimilarity index. Whether adjusted for index bias or not, the scores of the dissimilarity index are not able to distinguish between two very different patterns – namely, polarized unevenness associated with prototypical segregation and dispersed unevenness associated with a more benign pattern that is rarely discussed in the literature despite being surprisingly common (Fossett, 2017). Both patterns are common in nonmetropolitan contexts so the distinction between the forms of segregation associated with these patterns is highly relevant in the present study. As we explained in more detail in Chap. 2, the inability of the dissimilarity index to distinguish between polarized unevenness and dispersed unevenness takes on much greater practical significance in contexts where one group is disproportionately larger than the other (i.e., with a larger to smaller group ratio reaching 6:1 or higher). For example, consider a nonmetropolitan community where 98 percent of the pairwise population is White. If the typical minoritized group household lives on a block where the composition of the block is 96 percent White, they will technically live in a neighborhood that departs from parity on percent White. While the departure from parity on percent White is quantitatively small (i.e., 2 points), the prevalence of this pattern can easily produce very high scores on D because D is extremely sensitive to this aspect of uneven distribution, which we term dispersed unevenness. In simplest terms, at a given value of D, uneven distribution is maximally dispersed when as large a share of the minoritized group population as needed to produce the value of D in question resides in below parity areas that are as close to parity as possible.

Technically, the value of D in this situation will be correct as calculated and the usual interpretations will apply; for example, a value of 70 would indeed indicate that the majority-minoritized group difference in percentage residing in areas at or above parity is 70 with the consequence that at least 70 percent of the households in one group would have to change neighborhoods to bring about exact even distribution. The problem is that practices in the literature have fostered assumptions about the implications of the value of D that not only are not always correct but often are incorrect and highly misleading. Specifically, a high score on D is likely to be misinterpreted as signaling that a prototypical pattern of segregation associated with polarized unevenness is present when the reality of the situation is that this may be far from the case. The basis for this mistaken assumption is that didactic illustrations of segregation involving high scores of D (e.g., Iceland et al., 2002; Jaret, 1995; Taeuber & Taeuber, 1965) invariably show a pattern of polarized unevenness where neighborhoods that depart from even distribution are polarized on group composition into, for example, all-White (or nearly so) and all-Black (or nearly so) neighborhoods. What is never shown (at least to the best of our knowledge) outside of Fossett (2017) and this book, is that high values of D can arise in the more benign situation where the minoritized group generally or even exclusively lives alongside the majority group in areas that are close to parity on percent majority group in a pattern of dispersed unevenness that is rarely acknowledged and for which it is much harder to make the case that segregation in this form carries actual or potentially meaningful consequences for life chances.

We find that not only is it logically possible for D to register high scores when the two groups in question are unevenly distributed without a pattern of prototypical segregation, it is empirically common. This generally unrecognized possibility for the dissimilarity index to take high scores based on a pattern of dispersed unevenness is particularly relevant for measuring White-Latino and White-Asian segregation in nonmetropolitan communities, and, to a lesser degree, also for White-Black segregation. That is, White-Black segregation is more often characterized by a prototypical pattern of segregation wherein White and Black households are truly living in different neighborhoods in nonmetropolitan communities as well as in large metropolitan areas. In contrast, White-Latino segregation in nonmetropolitan communities is more varied and frequently takes the pattern of dispersed unevenness wherein the dissimilarity index takes high scores, but Latino households co-reside extensively with White households and rarely reside in predominantly Latino neighborhoods, if ever. Accordingly, analysis of White-Latino segregation must be measured in a more careful and nuanced way that can distinguish between the distinctly different possibilities for patterns of Latino settlement and residential distribution. The limitations of D are even more salient when evaluating patterns of White-Asian segregation as high values for D are almost never linked to patterns of prototypical segregation as typically seen for White-Black segregation. To be clear, this is not a technical issue in measuring White-Asian segregation. White-Asian segregation logically could take the form of prototypical segregation and sometimes does. But these cases are the exception and the overwhelming pattern is that White-Asian segregation takes the largely unrecognized pattern of dispersed unevenness.

To amplify the point, White-Black segregation takes the prototypical form of group separation into homogeneous enclaves on a more frequent basis. But this result is not dictated by any technical considerations such as the group being small or large in absolute or relative size. White-Black segregation logically can take the form of dispersed unevenness, and it occasionally does. But this pattern is the exception. The variation in these distinctly different patterns across group comparisons, across communities, and over time is sociologically important. It cannot be identified in studies using only the dissimilarity index. Over the past four decades, nonmetropolitan communities have been reshaped in relatively dramatic ways by migration with many areas experiencing non-trivial Latino settlement for the first time and racially diversifying in other ways. But in general, most of these areas remain predominately White with the median pairwise percent White ranging between 93 percent and 99 percent from 1990 to 2010. These disproportionate racial and ethnic compositions create scenarios where high values of the dissimilarity index often are the result of uneven distribution in the form of dispersed unevenness. For example, most Latino residents live on blocks that are slightly less White than the area overall but are nonetheless predominantly, and often overwhelmingly, White.

White-Latino segregation in nonmetropolitan communities can, and we find sometimes does, take on a more prototypical form in nonmetropolitan communities, particularly those communities where Latino populations are more settled and racial dynamics that might lead to enclave formation or racial conflict and group stratification have had time to take effect. But we cannot rely on the dissimilarity index to distinguish between the communities where this happens and the communities where it does not. This limitation of D – namely, the potential for a high score to reflect either polarized unevenness or dispersed unevenness, is fundamental. The only way to distinguish between the pattern of polarized unevenness and the pattern of dispersed unevenness – the former of which there is a strong consensus that the pattern is substantively meaningful and potentially highly consequential for life chances and the latter of which is rarely discussed and has never been identified as substantively important – is to examine alternative measures that are sensitive to this aspect of uneven distribution.

As we undertake our study, we acknowledge and appreciate the work done by those few groups of researchers over the past four decades who have advocated for giving greater attention to nonmetropolitan communities and segregation patterns and who have made important contributions to filling gaps in our knowledge in this area. But we also note that research in this area has had to deal with significant methodological challenges beyond what those researchers would encounter when investigating segregation in the largest metropolitan areas, including some challenges that have only recently become clear. Our goal in this chapter is to build on and extend their pioneering efforts and contribute to this body of research by using new methods to address and overcome these measurement challenges and thereby gain a clearer understanding of the state of segregation in nonmetropolitan communities and how these patterns are shifting over time. But before we do that, we must address a fundamental question: What does residential segregation mean in nonmetropolitan communities?

4.5 Debates Over Meaningfulness of Residential Segregation in Nonmetropolitan Communities

It is possible for a nonmetropolitan community to sustain a racially diverse population with high levels of integration and little to no systemic racial conflict, but this outcome is not a given. Studies focusing on large metropolitan areas have established that residential segregation serves as an effective mechanism, often being explicit in intent and design, for excluding minoritized racial and ethnic groups from access to resources that can be hoarded to the benefit and enhancement of White neighborhoods (Massey & Denton, 1993; Trounstine, 2018). Traditional place stratification perspectives emphasize this key motivation behind segregation, which whether by active intent and/or by inertia, serves as a tool for maintaining White privilege and advantaged status position (Logan, 1978). A key example of this would be the nature of school districts, with school funding tied to local tax bases and private donations. Racial inequities in K-12 education can be dramatic in segregated metropolitan areas and primarily harm communities of color, with no negative educational attainment effects on White children (Kozol, 2005; Quillian, 2014).

Segregation also has consequences for urban development, with neighborhoods where minoritized racial groups predominate being more likely to be disrupted by highway expansions or industrial zoning. Finally, Massey and Denton (1993) made the compelling argument that the residential concentration of minoritized racial groups in homogeneous ghetto or enclave neighborhoods can intersect with concentrated poverty and enable profound economic disadvantage, with these neighborhoods bearing the brunt of economic downturns and being more likely to experience high levels of poverty. While some of the specific aspects of segregation in metropolitan environments may not directly translate to nonmetropolitan settings (e.g., racial segregation in public high schools), we may still ask: Do similar consequences of segregation occur in nonmetropolitan communities? If so, what form do they take? As the presence of minoritized racial and ethnic groups in nonmetropolitan communities grows, we must consider the hypothesis that social conditions of competition for resources may crystalize along racial lines. This competition can lead to racial intolerance, racial conflict, persistent racist ideology, and structured racial inequality across many domains (Fossett & Kiecolt, 1989).

Cities and towns in micropolitan areas and noncore counties are by definition less populated and also often less densely settled, and therefore do not replicate some kinds of enduring, large-scale patterns of segregation observed in large metropolitan areas like Chicago, Detroit, Los Angeles, or New York City where large portions of groups reside in deep racial isolation, resulting from expansive regions of adjacent neighborhoods that are highly polarized on group composition. This condition sharply inhibits intergroup interactions and shared experiences and creates structural conditions that make group inequality on location-based outcomes logically possible. But patterns of segregation in nonmetropolitan settings can and often are enduring and consequential in their own right. Intriguingly, much of nonmetropolitan America has seen growth in minoritized racial populations in the post-Civil Rights era, raising questions regarding how settlement patterns form given the existence of fair housing laws and the end of most de jure segregation practices. Furthermore, neighborhoods in nonmetropolitan communities are smaller in scale compared to the ethnic ghettos, barrios, and enclaves seen in some metropolitan areas. Thus, in small towns and rural communities, there are more opportunities for intergroup interactions within a municipality where it is feasible, at least in principle, for most residents to access the same shops, services, and communal spaces while children attend the same schools.

However, anyone with even passing familiarity with nonmetropolitan communities will understand that differential spatial distributions can still be consequential in these settings and can create the logical potential for group inequality on location-based outcomes when racially polarized neighborhoods form at a smaller spatial scale. For example, residential segregation and associated spatial distributions can signal the state of race relations in the community, sometimes in very dramatic and explicit ways that carry practical as well as symbolic import. As Lichter (2012) cautions, the closer proximity and higher levels of interaction between White and minoritized racial groups in nonmetropolitan communities can potentially foster “mutual understanding” but it can also provide more opportunities for group conflict and the emergence of relations of racial hierarchy and dominance (2012: 26) as suggested by group competition theory (Blalock, 1967; Olzak & Nagel, 1986), which have been supported by findings from previous research on racial inequality in nonmetropolitan communities (Fossett & Therese Seibert, 1997). As an example of how this conflict can manifest, Lichter et al. (2018) discussed how heightened political divisiveness and anti-immigrant sentiment could be affecting reactions to growing Latino populations in nonmetropolitan communities, which has largely been driven by foreign-born migrants (Lichter et al., 2018).

Powerful historical evidence of the possibility of conflict and segregation in nonmetropolitan communities is exemplified by the “sundown towns” that emerged across the United States beginning in the early twentieth century, which were indicative of the resurfacing of overt racism and racial discrimination in the post-Reconstruction era throughout the nation in tandem with Jim Crow segregation taking root in the South (Loewen, 2006). Communities of all sizes and predominately outside of the South drove out Black households through intimidation, violence, and local law, creating intentionally all-White communities with Black households being excluded and relegated to rural settings, often outside of administrative boundaries for city services and political representation. James Loewen’s deep archival research and analysis of census data identified thousands of definite and probable sundown towns in the United States, some of which maintain this status to the present (Loewen, 2006). These extreme patterns of segregation in nonmetropolitan settings and their implications differ in key ways from urban neighborhood segregation, as they often resulted in entirely White municipalities with Black households fleeing to larger urban areas or to rural all-Black towns and enclaves shut out from economic and political advancement (Loewen, 2006).

As for other consequences of segregation that are more apparent in metropolitan areas such as economic inequality, school inequality, housing disparities, and health disparities, we may see different expressions of inequality in nonmetropolitan communities. For example, in smaller towns and communities, all the children likely attend the same schools given that many rural communities only have a single elementary school, junior high, and high school. Consequently, Logan and Burdwick-Will (2017) find that racial and ethnic school segregation is significantly lower in rural areas compared to urban areas, although rural schools tend to underperform as a result of higher levels of poverty. Even so, we still might find that integration in public schools in nonmetropolitan settings often turns out to be a phantom achievement. In the Jim Crow South, White and Black children attended schools that were separate and massively unequal, and the Civil Rights Era brought an end to this formal system. But Cready and Fossett (1998) documented a historical transition over the period 1969–1990 where the end of de jure school segregation and racial inequality in quality and quantity of education in the nonmetropolitan South was followed by large-scale movement of White families into White-dominated non-public schools and increasing neglect and even abandonment of public schools in counties where the Black population reached thresholds at or exceeding 10–15 percent of the population. Public schools then received lower funding as White families paying enrollment fees for non-public schools had reduced incentives to maintain the quality of public schools. In this broader perspective, separate and unequal did not really disappear. Consistent with this pattern, segregation may occur at the meso-level in noncore counties with multiple schooling options.³

What can also clearly vary in disparate ways is socioeconomic status, access to resources and services, and exposure to poverty. For example, Albrecht et al. (2005) studied nonmetropolitan minoritized group concentration and reported two key findings: minoritized racial groups experienced greater economic disadvantage when living in counties with higher minoritized group concentration, and White residents experienced greater advantage in counties with larger minoritized racial populations. While they did not measure segregation within counties, their findings suggest that racial inequalities can exist in nonmetropolitan communities which are tied to spatially bounded demographics. In micropolitan areas and noncore communities, unique problems not generally seen in large metropolitan areas can emerge for those who live near incorporated areas. Municipal and other administrative boundaries can be highly consequential in these contexts, especially for households without the socioeconomic resources needed to offset certain challenges that result when residing on the wrong side of the boundary. Excluded residents may be disadvantaged on many important dimensions including access to municipal services relating to healthcare, emergency services, road maintenance, treated water, sewage and sanitation services, existence and maintenance of drainage and flood control systems, internet service, and transportation (Johnson et al., 2004). Lichter and Parisi (2008) found that rural poverty disproportionately impacts Latino and Black households, leading to social and economic isolation. As they argue, the interplay of race and class dynamics that are known to correlate with segregation are also evident in rural contexts but with greater constraints as it is more difficult for those who are most disadvantaged to seek out new environments and opportunities (Lichter & Parisi, 2008).

We name here a few other examples of how segregation in nonmetropolitan communities may be consequential for disparities in access to resources. First, Julia Caldwell et al. (2017) found that Black and Latino residents in segregated rural communities reported having worse access to a usual source of healthcare, although they were also more likely to report that their healthcare needs were being met, which the authors attribute to a possible “ethnic density” effect, particularly in areas where Latino population growth via migration has been high. Second, Erin York Cornwell and Matthew Hall (2017) reported that the risk of exposure to neighborhood problems has increased in rural areas for Black and Latino residents, and that racial disparities in perceived neighborhood problems are on the rise in these same communities.

There is also mixed evidence in the literature that White-dominated communities may be selective in annexing new neighborhoods depending on the racial composition of the neighborhood, echoing the “sundown town” dynamic. Lichter et al. (2007b) studied municipal under-bounding in rural southern communities, where municipalities will choose not to annex areas if doing so would change the demographics of the community and extend public service access to marginalized populations. They found mixed results, but one telling finding is that predominately White communities were less likely to annex neighborhoods with predominately Black populations. In contrast, Wilson and Edwards (2014) found no conclusive evidence of ethnicity-based municipal under-bounding in Midwestern communities when looking at percent Latino in fringe areas. To the extent that it may occur, municipal under-bounding holds implications for the health and well-being of excluded populations, in addition to the costs that these residents face by having to rely on privatized services which would otherwise be publicly funded such as sanitation, water, road maintenance, and emergency services. These political decisions bear consequences for segregation and equal access to resources and opportunities. Therefore, segregation can still be meaningful if, for instance, a minoritized racial group is predominately residing outside of a town’s boundaries in rural enclaves, mobile home parks, and the like without services and amenities that are available in towns and nonmetropolitan cities.

In cases where the minoritized racial groups present in a nonmetropolitan community are a relatively new but growing population, a trend that emerged in the 1980s and 1990s, racial segregation can also serve as an indicator of the sort of reception these groups are given by the predominately White population established in these areas. Questions that may be asked in these situations include: What do initial settlement patterns look like?, How do these patterns shift over time? and, What role do changing demographics play in shaping the nature of social interactions as new migrants become permanently settled, start or are rejoined by their families, and interact more with the institutions of their new communities? The possibilities remain open for enclaves to form, for the newcomers to become fully integrated, or for racial conflict to emerge or intensify and lead to place stratification dynamics. This specific category of communities, referred to as new destinations, has been of particular interest in the nonmetropolitan segregation literature and is one that we focus on in the next chapter.

In sum, the literature on how and why nonmetropolitan segregation matters is mixed and far from conclusive, but there is enough to suggest that spatial residential distributions in nonmetropolitan communities can be and often are consequential for group inequality on location-based outcomes. To what extent and under what conditions remains to be understood and likely has much to do with local context and demographic changes. While we will not go as far as analyzing the consequences of segregation in nonmetropolitan communities, we undertake the important first step of producing valid measures of segregation in these areas that are free of the inherent biases which have vexed previous attempts to study nonmetropolitan segregation. The measurement choices that we make allow us to avoid the problems of upward bias and the risk of overstating the extent to which groups are residentially separated from one another without having to impose any major restrictions on the areas selected for analysis.

Perhaps with these refined baselines established, the literature can advance towards a better understanding of what nonmetropolitan segregation looks like and what it means for the people who experience it. We will summarize trends and patterns of racial segregation in nonmetropolitan communities. But we will also explore these patterns more deeply by conducting more aggregate-level analyses, mapping case studies, and comparing areas where prototypical segregation is occurring to areas where dispersed unevenness is evident and group separation is absent. In doing so we will further emphasize a central methodological point, which is that the choice of segregation measurement can be highly consequential for how we understand segregation, especially in nonmetropolitan communities.

4.6 Data

For the analyses in this chapter we continue to use data from decennial census summary files for 1990, 2000, and 2010, drawing specifically on census block tabulations of householder race and ethnicity data to calculate values of index scores for White-Black, White-Latino, and White-Asian household segregation in micropolitan areas and noncore counties. Micropolitan areas are similar to metropolitan areas in being Core Based Statistical Areas (CBSAs) constructed from one or more counties associated with a well-defined urban core. The main distinction is size and scale. Micropolitan areas have urban cores with populations between 10,000 and 50,000 and are thus smaller in size and scale in comparison with metropolitan areas, which have urban cores with populations from 50,000 up into the millions. Micropolitan areas are by definition not entirely rural but in many cases do have larger percentages of population residing in rural communities because they have smaller urban cores. Noncore counties, by contrast, are counties that do not contain an urban core of 10,000 and are not closely linked to a nearby urban core (e.g., through discernable commuting patterns).

We again impose minimal restrictions on our case selection, excluding areas where either group in the analysis has less than 50 households present in the area and areas where either group in the analysis comprises less than 0.5% of the pairwise population. This is to ensure that we are only measuring segregation in areas where block-level segregation could meaningfully occur. When one group in the analysis falls below these thresholds, it is highly unlikely that segregation could be sustained in any consequential way. Applying our selection criteria creates an analysis dataset that includes 46 percent of all U.S. nonmetropolitan communities for our White-Black analysis, 71 percent for our White-Latino analysis, and 18 percent for our White-Asian analysis by 2010.

4.7 Measurement and Approach

The majority of this chapter consists of descriptive analyses of White-Black, White-Latino, and White-Asian segregation in nonmetropolitan communities using direct quantitative measures of segregation as well as GIS mapping. As we did in Chap. 3, we adopt three innovative approaches to segregation measurement, which we hold are especially critical for gaining more accurate and informative assessments of the nature of segregation patterns in nonmetropolitan areas. First, we rely on the separation index (S) to measure important aspects of evenness that cannot be identified using the dissimilarity index (D). The separation index, like all measures of uneven distribution, registers positive values when the racial composition of one or more neighborhoods deviates from the overall composition of the community. The key for our needs is how different measures register the deviations. The separation index (S) takes high values only when deviations from even distribution are quantitatively large for at least one group (and maybe both groups). The popular alternative is the dissimilarity index (D), another measure of evenness that has historically dominated the segregation literature. Methodological studies note it is insensitive to the quantitative magnitude of departures from even distribution. But readers and even many researchers do not always appreciate how this can lead high scores on D to be misleading.

For the sake of self-containing this chapter, we offer here again a brief explanation of how the dissimilarity index and separation index are commonly calculated and interpreted, and how both are altered using the new methods developed by Fossett (2017) and employed in this book. Both D and S have fairly straightforward, easy-to-explain interpretations, especially when conceptualized in the difference-of-means formulation introduced by Fossett (2017). In this framework, all widely used measures of uneven distribution are reconceptualized as a simple arithmetic difference in group means on a neighborhood outcome (y) scored on the basis of area racial composition. The attractive quality of this framework is that it reveals very clearly how indices differ in registering large and small departures from even distribution. In the case of segregation from White residents, the commonly used dissimilarity index can be interpreted as the simple difference between the proportion of each group (e.g. White households and Black households) that lives in a neighborhood where the neighborhood proportion White is equal to or greater than the proportion of the population that is White for the community overall. The separation index has an equally easy interpretation; it is the simple difference in the average neighborhood-level proportion White between the two groups in the analysis. A thorough discussion of the implications of these differences in measurement for segregation research is presented in Chap. 2.

Significantly, the inherent level of upward index bias in S is always lower than the inherent bias in D. The difference in impact of bias on the respective scores of S and D can be large when measuring segregation using block data and it can be extremely large when measuring segregation for new, emerging groups. Fossett (2017) provides procedures for calculating versions of D and S that are free of index bias. This approach to computing unbiased index scores draws on the difference-of-means framework mentioned earlier. In this framework the source of bias can be described in fairly simple terms (see Chap. 2 for a more thorough technical discussion). Index scores for White-Latino segregation, for example, are computed as the White-Latino difference of group means on residential outcomes (y) for households scored on residential contact with White households as indicated by proportion White in their neighborhood of residence. The standard calculation of contact includes both contact with others and contact with self. Under random assignment, contact with others will have the same expected value for both groups and therefore does not contribute to index bias. In contrast, contact with self is fixed and cannot be randomly assigned. It is automatically higher for White households and lower for the minoritized group households. This is the sole source of bias in indices of uneven distribution (Fossett, 2017). This insight leads to a simple adjustment that eliminates index bias. It is to calculate contact for a household after removing the household from the terms of the calculation. Or in other words, contact should be computed for neighbors rather than for the entire neighborhood population. The logic is simple: Do not treat a household as its own “neighbor” and the source of index bias will be eliminated. We apply this correction in our formula of the separation index by first casting the separation index as a difference of means, as shown below:

$$ S={\overline{Y}}_1-{\overline{Y}}_2 $$

(4.1)

Where Y₁ is the average contact score for the first group in the analysis and Y₂ is the average contact score for the second group in the analysis. For the separation index, the contact scores are calculated as shown below:

$$ {p}_i^{\prime }=\left({n}_{1i}-1\right)/\left({n}_{1i}+{n}_{2i}-1\right)\ \mathrm{for}\ \mathrm{households}\ \mathrm{in}\ \mathrm{the}\ \mathrm{reference}\ \mathrm{group},\mathrm{and} $$

(4.2)

$$ {p}_i^{\prime }=\left({n}_{1i}-0\right)/\left({n}_{1i}+{n}_{2i}-1\right)\ \mathrm{for}\ \mathrm{households}\ \mathrm{in}\ \mathrm{the}\ \mathrm{comparison}\ \mathrm{group}. $$

(4.3)

Where n_1i is the count of households belonging to the reference group in the analysis in the reference household’s spatial unit, or neighborhood, i, and n_2i is the count of households belonging to the comparison group in the analysis in the reference household’s spatial unit. The bias correction can be found in these equations, where the reference household is subtracted. In the case of households that belong to the reference group, they are subtracted from both the numerator and denominator. For households that belong to the comparison group, they are only subtracted from the denominator because their counts are not included in the numerator (for example, in a calculation of pairwise proportion White, only households with White householders would be removed from both the numerator and the denominator).

Finally, the third measurement innovation is that we chose to measure segregation of households as opposed to the convention of measuring segregation of persons. Most empirical studies of segregation calculate segregation index scores using tabulations for persons. There are several understandable reasons for this. Person data tabulations are more widely available and person data tabulations are the first to be released after any decennial census. And, it is substantively reasonable to wish to consider the full populations of the groups involved when assessing segregation. Unfortunately, segregation index scores based on person data are susceptible to a source of index bias than is not generally recognized. As a result, the problem of index bias is more severe than is widely appreciated and sound, effective options for dealing with index bias when using person data are not available.⁴

The last statement may seem odd since the sections above outlined multiple ways to obtain unbiased scores for segregation indices. Note, however, that the methods for obtaining unbiased index scores reviewed above are appropriate for application to data for households but they are not appropriate for application to measuring residential segregation using data for persons. This distinction between households and persons is important but not widely appreciated. To explain the issues involved, we now consider the nature of the results obtained when the procedure for obtaining unbiased index scores given in the last section is applied with person data instead of household data. The key to the procedure is to adopt a refined formula for calculating contact with the reference group that excludes contributions of self-contact – the source of bias in standard (biased) computing formulas. When the procedure for working with data for households is applied to data for persons, the exercise will reduce the level of bias in the obtained index scores in comparison to scores obtained using standard (fully biased) formulas. But, importantly, the reduction in bias will only be partial rather than complete. Data we review in Chap. 2 suggests that on average, eliminating self-contact for persons eliminates only about a third of index bias that originates in fixed same-race contact within households. So, the scores obtained are closer to standard (biased) scores than to fully unbiased scores obtained using data for households.

This disappointing result traces to a simple but highly consequential fact. It is that most individuals do not locate independently; instead, most individuals locate in coordination with a cluster of individuals that together form a household and in an overwhelming majority of cases the households are racially homogeneous. This fact makes all of the procedures for obtaining unbiased index scores outlined above inappropriate for use with person data. To help draw out the basis for this conclusion, consider the following. Under random assignment of persons as members of racially homogeneous households, they are assigned in n-person clusters of same-race where n is the number of persons in the household. This will produce index scores that are much higher than when persons are distributed independently of the other members of their household, and the difference can be large. To bring these higher index scores down, one would have to break up many households and redistribute the individuals in them to other neighborhoods, and that is obviously a non-sensical proposition.

4.7.1 Summary of Methodological Approach

Our segregation measurement choices make it possible to draw out conclusions that reflect the reality of segregation in nonmetropolitan communities over time more accurately and fully. Measuring segregation of households using an index that is free of bias and that is up to the task of indicating when two groups are truly living in different neighborhoods, regardless of the size of either population or the spatial unit, makes it possible to study more nonmetropolitan communities across the U.S. than has been done before. Although our substantive interpretations are restricted to results from the unbiased separation index, we still take the opportunity in this chapter to compare outcomes measured with both the unbiased separation index and the unbiased dissimilarity index because nonmetropolitan communities are prime candidates for the sort of discordance that can occur between the two indices. We limit our analysis and discussion of this issue to summary scores and a selection of case studies that represent circumstances when the indices are in alignment and when they are not. This exploration of measurement issues is bolstered by GIS mapping, which allows us to visualize the extent to which groups are actually experiencing prototypical residential segregation in a nonmetropolitan community. This gives us a deeper, more nuanced analysis of segregation in nonmetropolitan contexts, and also permits us to showcase some of our methodological points, with the primary point being that segregation indices can react in considerably different ways to uneven distribution that occurs without high levels of residential separation. Shapefiles for GIS mapping are obtained from the National Historic Geographic Information System (NHGIS) through IPUMS at the University of Minnesota (Manson et al. 2022).

4.8 Changing Demographics of Nonmetropolitan Communities

The changing sizes of our analysis samples over time using our household population-based selection criteria hint at the changes taking place in nonmetropolitan communities with regards to racial and ethnic diversity (Table 4.2). The criteria that neither group in the pairwise analysis have a household population of less than 50 in the area means that the number of cases included for analysis varies from one decade to the next. Typically, this results when the minoritized racial group in the analysis is small in 1990 but grew in size in the following two decades – a pattern that is especially common for the Latino population. In 1990, these selection criteria give us 808 communities for analyzing White-Black segregation, 694 communities for analyzing White-Latino segregation, and 123 communities for analyzing White-Asian segregation. The majority of the communities included in our White-Latino and White-Asian analyses in 1990 are micropolitan areas, which by definition tend to be larger in overall population size than noncore counties. What is notable is that by 2010, the number of communities included for analysis increased by 72 for White-Black comparisons, 671 for White-Latino comparisons, and 227 for White-Asian comparisons. For our analyses of White-Latino and White-Asian segregation, these are sizable increases that were primarily driven by growing racial diversity, especially in noncore counties. While not all micropolitan areas are uniformly diversifying or experiencing nonwhite population growth, these overall trends demonstrate that the nonmetropolitan U.S. is increasingly heterogeneous on race-ethnicity of persons and households. In Table 4.3 we also present the pairwise percentage of each nonwhite group across time and communities. What is notable here is the pairwise percentages of Latino and Asian populations in micropolitan and noncore counties are quite low, which creates the conditions under which standard segregation indices may generate misleading results impacted by index bias.

Table 4.2 Nonmetropolitan communities included in analysis by year, community type, and pairing

Table 4.3 Racial composition (pairwise) by year and community type

Nonmetropolitan communities in general have seen increases in nonwhite populations from 1990 to 2010 (Table 4.4), with the largest increases occurring for the Latino population. Based on household data, which exclude group quarters and institutionalized populations, Latino percentage growth rates in all nonmetropolitan communities are keeping pace with metropolitan areas at a median of 200 percent over the two decades, and the median growth rate is higher when we look only at the subset of nonmetropolitan communities included in our analysis (i.e., leaving out areas where population sizes are especially small and either group’s share of the population is below 0.5 percent). Median Asian population growth rates in nonmetropolitan communities are lower than in metropolitan areas, but nearly on par when we only look at the subset of nonmetropolitan communities included in our analysis.

Table 4.4 Median percent changes in minoritized group by community type, 1990–2010

Finally, median Black population growth rates are always lower in nonmetropolitan communities than in metropolitan areas, but especially when we only look at the nonmetropolitan communities included in our analysis. Black population growth rates are also lower in comparison to Latino and Asian growth rates, in part because growth of the latter groups is bolstered by immigration in addition to natural increase. These numbers lend support to the call for more research on segregation and racial diversification in the nonmetropolitan communities of the United States. With this sense of growing diversity in nonmetropolitan communities, the central questions we ask next are: What does residential segregation look like in nonmetropolitan communities, and how has it changed?

4.9 Overall Trends in Nonmetropolitan Residential Segregation

In Table 4.5 we summarize segregation according to the separation index for White-Black, White-Latino, and White-Asian comparisons in 2010 by community type. Generally, we find that segregation is higher in noncore counties for White-Latino and White-Black segregation, communities that are by definition smaller in population size and more remote from urban centers than micropolitan areas. For White-Asian segregation, we find no important differences between noncore counties and micropolitan areas, likely because the nonmetropolitan Asian population comprises a much smaller share of overall community populations, making it unlikely to find anything other than low levels of segregation regardless of type of community. Unsurprisingly, given documented national trends, segregation is highest between White and Black households, but not as high as what we have observed in large metropolitan areas where patterns of White-Black spatial distributions often coalesce into pronounced levels of hypersegregation – that is, high levels of segregation on several additional dimensions of segregation beyond uneven distribution (Massey, 2020; Massey & Denton, 1989; Massey & Tannen, 2015; Wilkes & Iceland, 2004; also see Chap. 3).

Table 4.5 Separation index (unbiased) by year, community type, and pairing

White-Black segregation was high in noncore counties and micropolitan areas and declined to moderate levels in micropolitan areas, while White-Latino segregation and White-Asian segregation have been low in both types of areas. Notably, White-Black segregation is declining across nonmetropolitan communities, tracking national trends towards lower, albeit still relatively high, levels. What differs drastically from previous research on nonmetropolitan segregation is that we find no evidence that White-Latino segregation is typically high in nonmetropolitan communities. Indeed, White-Latino segregation scores barely reach medium levels, and that is only observed in 1990 in micropolitan areas. Since 1990, nonmetropolitan White-Latino segregation has declined to low levels in micropolitan areas and has held steady at low levels in noncore counties. While less is said about nonmetropolitan White-Asian segregation in the literature, our findings clarify it would be wrong to adopt a default assumption that White-Asian segregation is high or even medium in nonmetropolitan communities. Instead, we find White-Asian segregation has been steadily at low levels over the decades of our analysis in both micropolitan areas and noncore counties.

Our finding that segregation scores in nonmetropolitan communities are lower than have been previously reported applies across all major White-nonwhite group comparisons. Consequently, much of what we know about the relative differences between White-Black, White-Latino, and White-Asian segregation based on earlier research focused on metropolitan areas also applies in nonmetropolitan communities. Most importantly, Black households are usually the most segregated while Asian households are the least segregated. However, because this knowledge is mostly derived from studies of metropolitan contexts, we must consider the different demographic circumstances and dynamics of changing racial composition occurring in these nonmetropolitan communities where populations generally are more homogenous and disproportionately White than in metropolitan communities and where minoritized racial groups initially comprise smaller shares of the population, but in many cases are growing rapidly. Thus, we next turn to a bivariate analysis of the relationship between minoritized group population growth and levels of nonmetropolitan segregation in 2010.

In Table 4.6 we correlate the percent change in the (pairwise) minoritized racial population with point changes in the separation index from 1990 to 2010. For White-Black and White-Latino segregation we find moderate correlations and for White-Asian segregation, where patterns are more static over time, we find a weak correlation. In general, minoritized racial population increases are correlated with rises in segregation for the minoritized racial group in the comparison, holding implications for the many nonmetropolitan communities across the United States that have been racially diversifying over the last few decades. Previous research on this issue has already speculated on what it means for race relations, with recent work by Lichter et al. (2018) finding that White flight from nonmetropolitan communities could be undermining the potential for integration and intergroup exposure.

Table 4.6 Correlations between minoritized population change and changes in the separation index

Our overview of segregation trends in nonmetropolitan communities relied on the separation index. As explained previously, this is because the more widely used dissimilarity index is not a good choice for describing levels of segregation in nonmetropolitan contexts because of its inability to distinguish between the different patterns of dispersed and polarized unevenness, both of which are common in nonmetropolitan communities. The technical basis for this decision is established in Chap. 2. However, we provide a less-technical review of examples to highlight how very different patterns of uneven distribution can produce equally high scores on the dissimilarity index and to help explain why the literature has so far reported high levels of segregation in nonmetropolitan communities, when that is mostly not what we have found here. Thus, in the next section we elaborate on some of our methodological points about segregation measurement which become especially relevant for studying nonmetropolitan communities. We also ask a related substantive question: How have patterns, rather than levels, of uneven distribution been changing over time in these communities?

4.10 Diverging Measures of Segregation and Patterns of Uneven Distribution

We previously described why and how the dissimilarity index can report high levels of segregation when close review of the residential distributions for the two groups in the comparison reveal that they in fact are living together, occupying the same neighborhoods, experiencing similar levels of contact, and, by logical implication, experiencing similar averages on location-based outcomes. We have also highlighted how considering both indices together can reveal more about the patterns of unevenness that are occurring in communities. Here we again review the qualities of the dissimilarity index and the separation index in more detail to document and clarify the residential patterns that prevail when the two indices diverge, which occurs frequently in nonmetropolitan contexts. Thus, we are capitalizing on the inherent limitations of the dissimilarity index and the superior qualities of the separation index to describe patterns of uneven distribution, a term that we tend to use interchangeably with segregation, in nonmetropolitan communities. We draw on Fossett’s (2017) terminology that distinguishes between prototypical segregation associated with polarized unevenness and the more benign pattern of dispersed unevenness. Both residential patterns involve particular aspects of uneven distribution, but each with different implications for intergroup residential contact.

To review, Fossett defines patterns of prototypical segregation as “displacement from even distribution [that] concentrates the populations of the two groups into homogenous areas that differ by quantitatively large amounts on area racial composition” (2017: 78). In contrast, dispersed unevenness is defined as the opposite, where uneven distribution is occurring but “group residential separation and area racial polarization are far below the maximum levels possible for a given level of displacement” (2017:78). The research on nonmetropolitan communities has by and large reported that segregation is high in nonmetropolitan communities based on the dissimilarity index, to the point that it is treated as conventional knowledge. We argue that in many cases, the high scores on the dissimilarity index are produced by a pattern of dispersed unevenness rather than a pattern of prototypical segregation. We can support this argument by contrasting scores for the dissimilarity index with scores for the separation index, which only gives high scores under conditions of prototypical segregation. To put it another way, the separation index will never give a high score when the pattern of dispersed unevenness is present and a high score on S always indicates the presence of the pattern of polarized unevenness.

Therefore, in Table 4.7 we present summarized scores of the dissimilarity index in micropolitan areas and noncore counties alongside the previously reported separation index. Focusing first on White-Black segregation we observe that over time, there has been little overall discordance between the dissimilarity index and the separation index. As we expected, even in nonmetropolitan communities patterns of White-Black uneven distribution are more likely to manifest as prototypical segregation where White and Black households live apart from each other and occupy different neighborhoods, thus resulting in limited residential contact with one another and creating the possibility of the groups experiencing systematically different exposure to location-based outcomes. There is also only moderate discordance between the dissimilarity index and the separation index when measuring White-Latino segregation in nonmetropolitan communities, with quantitative differences between the two indices shrinking over time as the Latino nonmetropolitan population grows rapidly.

Table 4.7 Separation index and dissimilarity index side-by-side

However, there are substantive differences between the index scores when measuring White-Latino segregation that are worth noting because they hold implications for previous findings on this topic. Previous research had reported that nonmetropolitan White-Latino segregation reaches medium to high levels. We find significantly lower levels of segregation then previously reported when scores are calculated using the unbiased version of the dissimilarity index; indeed, scores shift down markedly to the lower end of medium levels. But as it turns out, even this change in results for D does not provide the full story of White-Latino segregation in nonmetropolitan communities because values of D do not typically indicate the same pattern of prototypical segregation that is present in White-Black segregation. In 1990 and 2000, we find that while the dissimilarity index signals medium levels of White-Latino segregation in micropolitan areas, the separation index indicates that White-Latino segregation is low. The dissimilarity index does not indicate low levels of White-Latino segregation in micropolitan areas until 2010. Here we draw out the substantive conclusions about shifting patterns of White-Latino uneven distribution based on comparing scores for D and S.

The discordance between D and S for micropolitan communities in 1990 and 2000 indicates the presence of the pattern of dispersed unevenness. The high value of D indicates that White and Latino households were unevenly distributed in the specific fact that, in comparison with White households, a greater proportion of Latino households were living in neighborhoods that were below parity on neighborhood proportion White. The substantially lower value of S indicates that in general White and Latino households were not living apart from each other and thus were not occupying fundamentally different neighborhoods as occurs under prototypical segregation. Latino households on average lived in neighborhoods that, while below parity on proportion White, were quantitatively close to parity. Thus, Latino households had average levels of residential contact with White households that were close to parity and we can conclude that Latino households necessarily experienced averages on location-based outcomes that were similar to those experienced by White households. However, by 2010, the separation index increases, indicating that Latino households are increasingly living in different neighborhoods apart from White households. This brings scores for D and S into closer alignment as scores for the dissimilarity index are more stable by comparison because scores for D are already high based on its strong response to dispersed unevenness and because D is much less sensitive than S when unevenness transitions from the more benign condition of being dispersed to the more potentially consequential condition of being polarized.

In the case of White-Asian segregation we find the most distinct discordance between the two indices out of all the comparisons. The dissimilarity index consistently shows medium levels of White-Asian segregation over time in both noncore counties and micropolitan areas while the separation index consistently shows very low levels of segregation. Unlike in the case of White-Latino segregation, this discordance does not subside over time because the pattern of dispersed unevenness for White-Asian segregation does not transition toward prototypical segregation. Thus, we have a clear example here of a situation where the dissimilarity index is reacting to uneven distribution without polarization (i.e. the two groups living apart from each other in neighborhoods that are polarized on group composition), while the separation index tells in a more straightforward way that White and Asian households in nonmetropolitan communities have quantitatively similar levels of residential contact with White households. In sum, there is no indication of prototypical segregation as the typical outcome for White-Asian segregation in nonmetropolitan communities. What appears to more often be the case is that Asian households in nonmetropolitan communities are more likely to live in neighborhoods that are slightly below parity on neighborhood proportion White – creating a pattern of dispersed unevenness which D, but not S, responds to strongly – but overall are still living in neighborhoods that are near-parity on contact with White households.

4.11 Case Studies: Areas with Dispersed Unevenness Versus Prototypical Segregation

To illustrate the differences between dispersed unevenness and prototypical segregation (polarized unevenness), we present a selection of case studies where the separation index and dissimilarity index are discordant, and when they are not. Using GIS mapping, we are able to demonstrate what residential patterns look like when both the dissimilarity index and the separation index are concordantly high and contrast those patterns to situations where the dissimilarity index is high, but the separation index is low. These comparisons will reveal quite strikingly what is meant by prototypical segregation versus dispersed unevenness and will illuminate the shortcomings of the dissimilarity index to distinguish between the two patterns of uneven distribution. We will present a pair of case studies for each group comparison, examining patterns of White-Black, White-Latino, and White-Asian uneven distribution.

For comparing patterns of White-Black uneven distribution, we selected two nonmetropolitan communities in Missouri and Kentucky. To serve as an example of prototypical segregation with a pattern of polarized unevenness, we focus on the case of the Sedalia, MO Micropolitan Statistical Area in 1990, which is composed of Pettis County, MO. In 1990 the area had 14,056 households. Of those households, 3.4 percent had a Black householder. White-Black segregation is measured with a score of 66.6 on the dissimilarity index and a score of 60.9 on the separation index. Values of both indices thus would be categorized as “high” segregation under the classification scheme we are using in this study (given in Table 3.2), suggesting prototypical segregation. Indeed, while Fig. 4.1 depicts a micropolitan area where most blocks in the less populated parts of the county are predominately White, a pattern of prototypical segregation is apparent in Sedalia, the central town and county seat of Pettis County. In this town, a cluster of neighborhoods north of a railroad track are 80–100% Black, while all other neighborhoods outside of this area are 80–100% White. It then follows that households from the two groups have little residential contact with each other because they live apart from each other in neighborhoods that are polarized on group composition. With the exception of the “clustering” aspect of segregation revealed in the figure, we can reach the main conclusions about the nature of segregation based solely on the value of the separation index, as the score is the difference in mean neighborhood percent White between White and Black households – a difference of over 60 percent. It is clear that within some of the central urban core of this county there is a pattern of prototypical segregation with White and Black residents living on opposite sides of the town. These patterns reflect an “other side of the tracks” form of segregation where there is often a physical boundary such as a road or railroad track that divides White and Black neighborhoods. Thus, while high levels of prototypical segregation are not as common in nonmetropolitan communities as in large metropolitan areas of the Midwest and Northeast, they are certainly possible, as we observe in this micropolitan area.

Fig. 4.1

Sedalia, MO MSA, 1990

We next examine the case in 2010 of Garrard County, KY, a noncore county, in Fig. 4.2. We first note that the racial composition of Garrard County is similar to that of Sedalia, with percent Black at 2.2 percent. However, as a noncore county, the area has a smaller household population of 6,668 households. Despite that, this county has a dissimilarity index score of 57.8, which, while a bit lower than in Sedalia, is still easily categorized as high. The major difference between the two counties is that the separation index in Garrard County is only 8.9, a value falling in the category of low (or even very low) segregation and over 50 points lower than the value for the separation index score of 60.9 for Sedalia. This documents that communities that have similar scores on the dissimilarity index can have fundamentally different patterns of group separation and levels of minoritized group contact with White households. It also documents that the potential for a high degree of discordance between values of D and S is not an artifact of group size. The two cases considered have a low level of Black population presence (3 percent for Sedalia and 2 percent for Garrard County) and yet differ dramatically on S. This is because the Black population in Sedalia is concentrated in racially polarized neighborhoods while in Garrard County Black households generally live alongside White households in neighborhoods where proportion White is near parity (which in this case is 97 percent White).

Fig. 4.2

Garrard County, KY, 2010

The discordance between the two indices in Garrard County is an indicator of dispersed unevenness, meaning that while uneven distribution is technically occurring, both groups live in neighborhoods that are near parity, which in this case is predominantly White, and the group differences on neighborhood racial composition are not remarkable. Indeed, when we map pairwise White-Black plurality in Garrard County, we find only one neighborhood where Black households constitute a numerical majority, and it is less than 80%. This is corroborated by a tabulation of blocks in Garrard County by levels of plurality. Thus, while it is more often the case that White-Black uneven distribution in nonmetropolitan communities takes the spatial form of prototypical segregation, as indicated by the on-average medium scores on the separation index, we cannot trust the dissimilarity index to tell us that, especially when we are trying to identify specific areas where prototypical segregation is occurring. For instance, without a better understanding of the nature of the dissimilarity index, we might incorrectly assume that a prototypical pattern of White-Black segregation prevails in both the Sedalia micropolitan area and Garrard County. GIS mapping reveals that this is clearly not the case in Garrard County.

In Figs. 4.3, 4.4, 4.5, and 4.6 we present comparable case studies for White-Asian segregation and White-Latino segregation. The two communities that demonstrate these divergent patterns for White-Asian segregation are the Morgan City, LA Micropolitan Statistical Area and the Midland, MI Micropolitan Statistical Area in 2010. These nonmetropolitan communities have similar scores on D of 60.2 and 57.0 but markedly different scores on S of 31.4 and 5.9, respectively. As in the previous example, both communities are similar on relative group size; the Asian population makes up between 1 and 2 percent of the pairwise and total populations in both communities and also is similar in absolute size. The difference in the values of S arises because White-Asian uneven distribution in Morgan City is polarized while the uneven distribution in Midland is dispersed. Choropleth maps (and associated block-level tabulations) document that the Morgan City micropolitan area contains a distinct and predominately Asian neighborhood along the Gulf Coast in a small community called Amelia. An examination of satellite images reveals that this predominately Asian neighborhood in Amelia consists of a sizeable mobile home park near a harbor out of which many Vietnamese-owned fishing and shrimping boats operate. Meanwhile, the choropleth maps (and associated block-level tabulations) for Midland reveal no predominately Asian neighborhoods. Habits of interpretation that are established in the literature could easily lead a researcher to mistakenly assume the comparable high scores on the dissimilarity index for both areas indicate that the level and pattern of White-Asian segregation is similar with Asian households living apart from White households in both communities. But the values of the separation index signal and clarify what the choropleth maps (and underlying block-level tabulations) reveal in more detail, which is that the comparatively high score for S for Morgan City indicates that White-Asian segregation takes the form of prototypical segregation associated with polarized unevenness while Midland, with a very low score on S and a high score on D, takes the much different form of dispersed unevenness.

Fig. 4.3

Morgan City, LA MSA, 2010

Fig. 4.4

Midland, MI MSA, 2010

Fig. 4.5

Greenwood, SC MSA, 2000

Fig. 4.6

Marshall, MN MSA, 2000

Finally, we show comparisons of two areas with divergent patterns of uneven distribution for White-Latino segregation. The first is the Greenwood, SC Micropolitan Statistical Area in 2000, where the pairwise percent Latino is 2.6, the dissimilarity index is 59.0, and the separation index is 44.5. The second is the Marshall, MN Micropolitan Statistical Area in 2000 where the pairwise percent Latino is 2.5, the dissimilarity index is 48.0, and the separation index is 8.6. The choropleth maps for each community again document the dramatically different patterns of racial composition of neighborhoods that can occur in communities that have similarly high scores on the dissimilarity index but very different scores on the separation index. The figure for Marshall shows scant evidence of Latino concentration in Latino neighborhoods as only a few very lightly populated blocks have a Latino plurality, with the highest Latino plurality neighborhood containing only five individuals. Instead, unevenness for Latino households involves dispersal across neighborhoods that are below parity but only by small quantitative amounts and therefore are predominantly White. In contrast, the city of Greenwood, the namesake and county seat of Greenwood County, has a clear pattern of polarized unevenness. The Latino population in the city of Greenwood lives apart from White households and is concentrated in neighborhoods on the southern side of the city that are predominately Latino in pairwise group composition. Note that since the overall racial composition of Greenwood is 45 percent Black, one would have to consider other measures such as overall Latino isolation (as measured by the P* contact index) and/or the value of S for the Black-Latino comparison to determine whether Latino households are separated from all other groups, or just White households.

What is very apparent from these cases is that when the dissimilarity index and the separation index are discordant, one will find no visual (or quantitative) evidence that the minoritized racial group is segregated into different neighborhoods from White households. A systematic GIS analysis would show that in every nonmetropolitan community where the dissimilarity index is high and the separation index is low, the neighborhoods in the community will not be polarized on racial composition but instead the (pairwise) racial composition of neighborhoods will vary in a narrow range relatively close to parity. While both indices are recognized measures of uneven distribution, D is less capable of distinguishing between the important difference between polarized unevenness associated with prototypical segregation and dispersed unevenness, which is more benign in terms of logical implications for the potential for groups to experience inequality on location-based outcomes.

Unfortunately, it is currently the case that the dissimilarity index is widely used in the segregation literature but with little awareness that a high score on the dissimilarity index may not correctly signal the presence of a prototypical pattern of polarized unevenness that most researchers will reflexively assume is present. Instead, it is often the case that high scores on D in nonmetropolitan communities are associated with the decidedly different pattern of dispersed unevenness. Thus, we return to the earlier methodological point, which is that it is appropriate to assign priority to reviewing scores for the separation index because, in addition to being far less susceptible to bias than the dissimilarity index when measuring segregation using block-level data, it will correctly signal whether uneven distribution takes the form of prototypical segregation associated with polarized unevenness and will do so reliably even when groups are small in absolute and relative size. As the GIS maps demonstrate, residential segregation in nonmetropolitan communities can take the highly polarized patterns we are accustomed to seeing in metropolitan areas, and we can rely on the separation index to indicate when this is so. The maps additionally demonstrate that segregation can also take the much different form of dispersed unevenness wherein S will take a low value and high values on D will be misleading if they are mistakenly interpreted as indicating that groups live apart from each other.

4.12 Summary

We undertook several major tasks in this chapter. First, we discussed how the literature has struggled to overcome the limitations of standard segregation measurement when analyzing segregation in nonmetropolitan communities. Then we illustrated how adopting new methods and refined formulations of familiar indices in combination with data for households rather than data for persons can overcome the problem of index bias and thereby open up a new era of research where studies of segregation in nonmetropolitan communities can include a larger and more representative set of communities and group comparisons. Next, we illustrated how considering the values of the separation index can, more so than any other measure of uneven distribution, reliably identify segregation comparisons that involve the prototypical segregation pattern associated with polarized unevenness, a necessary precursor for group inequality on location-based outcomes. Finally, we illustrated how one can identify the more benign segregation pattern associated with dispersed unevenness based on the discordant combination of a high value of D and a low value of S.

Our review of segregation measured using the separation index and block-level data for households identified new and important understandings of how patterns of White-Black, White-Latino, and White-Asian segregation vary across nonmetropolitan communities. Specifically, we found that the pattern of segregation we refer to as prototypical segregation does indeed occur in nonmetropolitan communities but not nearly as commonly or to the degree that previous research would lead one to believe. We find prototypical segregation involving polarized unevenness is more often the case for White-Black segregation, which tends to be relatively high even outside of large metropolitan areas. We also find that White-Latino and White-Asian segregation also sometimes takes the form of prototypical segregation in nonmetropolitan communities, but we find this occurs much less frequently than is seen in patterns of White-Black segregation. This leads to the major finding that, contrary what previous research would suggest, White-Latino segregation in nonmetropolitan communities is low rather than high and is stable or increasing rather than declining. The key takeaway here is that conclusions about White-Latino segregation in nonmetropolitan settings based on previous studies that relied primarily on scores for the dissimilarity index must be reconsidered for two reasons. First, index bias has substantial and complicated impacts on the levels and variation in values of D across communities and over time. Second, with much greater frequency than is generally appreciated, high values of D do not provide a reliable signal of the presence of prototypical segregation and register a more benign form of segregation we term dispersed unevenness. This same finding applies with equal force to conclusions about White-Asian segregation in nonmetropolitan communities where our analyses document that segregation is very low and has remained low since at least 1990.

Our primary substantive conclusion from both Chap. 3 and this chapter is that segregation in nonmetropolitan communities is often not as high as what is observed in metropolitan areas, especially for Asian and Latino households. However, segregation can be and sometimes is high in nonmetropolitan communities, even when the minoritized group proportion is small in absolute and relative size, a point that we highlighted through GIS mapping of case study areas. This finding should negate any skepticism that segregation in nonmetropolitan communities can approach levels seen in metropolitan areas. It can, and does, in particular communities and group comparisons. But the finding that segregation in nonmetropolitan settings is lower is not an artifact of methods of measurement; it is a sociological fact. One of the questions for future research is what consequences can flow from nonmetropolitan segregation. At the level of index scores there is overlap in distributions of scores, but it is not appropriate to project conclusions about the consequences of segregation gleaned from studies focusing on metropolitan areas to nonmetropolitan communities. The consequences and relevance of segregation in nonmetropolitan and rural settings are important, but they are not necessarily the same as in metropolitan settings.

We argue our conclusions regarding methods for studying segregation in nonmetropolitan settings should have a more immediate impact on segregation research. Our review of the existing literature on residential segregation in nonmetropolitan communities left us with one overarching assessment: past research encountered serious methodological challenges that prompted, or forced, researchers to both restrict analysis samples, invariably leading to smaller, nonrepresentative samples, and also to adopt a variety of questionable ad hoc practices in analysis. These decisions are all motivated by well-founded concerns about the potential for index bias to distort findings when segregation is measured at small spatial scales, especially when groups are small in absolute and/or relative size. We demonstrated how new methods of segregation measurement in combination with using data for households rather than persons provide highly effective solutions to the central problem of obtaining unbiased index scores, thereby freeing researchers from any need to adopt onerous sample restrictions and questionable strategies of analysis. Findings based on these new methods show that previous research reporting high levels of segregation in nonmetropolitan communities using the dissimilarity index must be called into question on two counts. First, because unbiased scores are much lower and vary in different ways across communities and over time. Second, because the dissimilarity index cannot distinguish between dispersed and polarized patterns of unevenness, the latter of which we term prototypical segregation. Prototypical segregation is the form of segregation that motivates research on segregation and researchers and lay audiences alike frequently and mistakenly assume this pattern is present when the dissimilarity index takes a high value. We document this is not the case both as a logical possibility and as a frequent empirical result in analyses that involve broader, more representative samples and a wider range of community settings. Accordingly, we caution researchers to avoid making the mistake of assuming high scores on the dissimilarity index are sufficient to support the conclusion that two racial groups are living apart in different neighborhoods that are polarized on racial composition and thus can experience inequality on location-based outcomes.

Other indices, in particular the separation index, are superior options that can be relied upon to provide a definitive signal that a pattern of prototypical segregation is present when the index value is high. This is important for studying segregation in nonmetropolitan communities, which often present circumstances of measurement under which the dissimilarity index is most likely to be problematic. Thus, we make two recommendations for measuring segregation in nonmetropolitan communities. First, always use the unbiased formulations of segregation indices as developed by Fossett (2017). Simply put, one is never worse off when using the unbiased scores, as they only deviate from standard scores in circumstances where the standard scores are flawed. Second, review scores of the separation index either alone or in combination with the dissimilarity index to get a complete picture of the nature of uneven distribution. The separation index is better suited than any other widely used index to indicate when the spatial distribution of groups across neighborhoods in a community takes the form of polarized unevenness associated with prototypical residential segregation that is invariably depicted in didactive presentations of high levels of residential segregation (e.g., White-Black segregation in Chicago).

Again, one is never worse off for examining values of the separation index. If values of D and S are concordant, the values of S provide confirmation that, by empirical coincidence, not logical necessity, high values of D are associated with prototypical segregation. If values of D and S are discordant, the low value of S provides the definitive basis for concluding the pattern of prototypical segregation is not present and instead the underlying pattern is one of dispersed unevenness. Adopting both options for segregation measurement will free segregation researchers to study residential segregation in nonmetropolitan communities across larger, more representative samples without being hampered by the long-standing and frustrating methodological challenges relating to index bias. These new approaches to measuring segregation yield superior measurements that can help researchers answer the call in the literature over many decades to better document and understand residential segregation in nonmetropolitan communities.