Abstract
In this first chapter of empirical findings, we provide an overview of trends and patterns in the residential segregation of households by race and ethnicity from 1990 to 2010 across metropolitan areas, micropolitan areas, and noncore counties. These three community types give our analysis wide geographic coverage across the United States. We examine segregation between White, Black, Asian, and Latino households across all areas over the two decades bracketed by the three decennial census years of 1990, 2000, and 2010. What distinguishes the analyses in this chapter from previous studies is that we apply new methods for measuring residential segregation which include correcting for index bias, using a measure that is better suited for distinguishing between different patterns of uneven distribution, and measuring residential segregation of households rather than persons. These adjustments allow us to expand the scope of our analysis to include more communities beyond the largest metropolitan areas while measuring residential segregation at the block-level, which is necessary for measuring segregation in less densely populated areas. While our findings at times reflect what past studies have found, it is more often the case that we reveal different levels and patterns of segregation, especially when focusing on White-Latino and White-Asian segregation where patterns of uneven distribution are more dispersed than previously understood. This first empirical chapter sets new benchmarks for residential segregation measurement and analysis in U.S. communities.
You have full access to this open access chapter, Download chapter PDF
3.1 Overview
In this first chapter of empirical findings, we provide an overview of trends and patterns in the residential segregation of households by race and ethnicity from 1990 to 2010 across metropolitan areas, micropolitan areas, and noncore counties. Metropolitan areas are identified as a cluster of one or more counties associated with at least one urban core with a minimum population of 50,000. These areas include the most populated cities in the United States such as New York City, Los Angeles, Chicago, and Houston along with their associated suburbs and exurban regions as established by commuting patterns and other criteria. Micropolitan areas are defined in a manner similar to metropolitan areas but on a smaller scale, having urban cores with a population of at least 10,000 but less than 50,000. Finally, noncore counties consist of counties that are not associated with an urban core and thus are not included as part of a metropolitan or micropolitan area. Noncore counties are generally considered to be rural in character with no or only limited commuting patterns to urban areas. These three community types give our analysis wide geographic coverage across the United States. We examine segregation between White, Black, Asian, and Latino households across all areas over the two decades bracketed by the three decennial census years of 1990, 2000, and 2010.
The purpose of this chapter is twofold. While many studies have provided broad descriptive analyses of segregation patterns over time for large metropolitan areas, a comprehensive analysis of all communities in the United States, additionally including all smaller metropolitan areas, all micropolitan areas, and all noncore counties, does not exist. Studies by Lichter and colleagues (2007, 2010) have reported analyses for place-based areas including urban, suburban, and rural contexts and stand as the best previous efforts to document segregation in rural and nonmetropolitan communities as well as in large metropolitan communities. We acknowledge and build on these important contributions. Importantly, however, we take advantage of new developments in segregation measurement to more effectively address and overcome the problems and challenges that have complicated the measurement and analysis of segregation in nonmetropolitan communities and smaller metropolitan areas.
Previous empirical studies of trends and patterns in segregation, for reasons that we review in more detail here and anticipated in our discussions in Chap. 2, have generally focused on describing and analyzing segregation in larger metropolitan areas and even there only for groups that are fairly large in size. This somewhat narrow focus has not necessarily been by choice. Instead it reflects researcher concerns about the problems associated with measuring segregation in a wider and more representative set of communities and group comparisons. In the past, these concerns were well-founded and have led most researchers to view it as necessary to be selective and limit the scope of research designs to exclude communities and group comparisons where scores for conventional segregation indices are susceptible to distortion by index bias. We overcome the limitations of previous research by drawing on new methods for segregation measurement that allow us to obtain valid and reliable unbiased index scores across a much wider range of communities and group comparisons than could be considered in previous research. Thus, the analyses we report in this chapter provide a more comprehensive descriptive analysis of changes in racial and ethnic residential segregation over time across the United States than has previously been possible because we obtain measurements of segregation that are free of index bias and therefore are more accurate and trustworthy for sustaining comparisons of segregation across nearly all communities, ranging from the largest metropolitan areas to the smallest noncore counties and for group comparisons involving both large and small group populations. In doing so, we also establish benchmarks against which anticipated analyses of segregation using the 2020 census can be compared.
Because a portion of this chapter involves reanalyzing segregation in large metropolitan areas that have already been widely studied and analyzed, we anticipate the patterns that we document in those communities may not deviate much from previous findings even though we are using new improved measurement methods that will sometimes lead to changes in index scores. This is largely because, as mentioned previously, most past studies have been careful to restrict the scope of their analyses to include only communities and group comparisons where they do not see a high risk of conventional measures being significantly distorted by the problem of index bias. For example, a common practice in segregation research is to impose restrictions on the set of communities in the analysis sample by including only the largest 50 or 100 metropolitan areas or group comparisons where both groups meet a combination of minimum absolute size and minimum population share (e.g. Frey, 2018; Iceland, 2014; Massey & Denton, 1988). These restrictions, particularly the focus on large metropolitan areas, exclude a significant portion of communities across the United States from segregation analysis.
Two other related practices include avoiding measuring segregation using neighborhood geography (spatial units) smaller than census tracts (e.g. Iceland et al., 2002) and differentially weighting cases to deemphasize the impact of cases more susceptible to index bias. The choice to measure segregation by operationalizing neighborhoods using census tracts for spatial units leads to underestimation of segregation outside of large metropolitan areas because large spatial units cannot accurately capture segregation that is manifest at smaller spatial scales in nonmetropolitan communities and smaller metropolitan areas. There are a few prior studies that have measured segregation using smaller spatial units at the census block level (e.g., Lichter et al., 2010; Allen & Turner, 2012), who also go beyond metropolitan contexts to study segregation in small towns and rural counties. However, these studies have faced an unavoidable dilemma with no good choices. Researchers are left with the strategies that we have previously described, which include working with segregation index scores that in some, perhaps many, cases are likely to be inflated by index bias at nontrivial levels that vary in magnitude across different communities and different segregation comparisons or excluding communities and group comparisons to minimize these problems. Tending toward the second choice has the severe practical consequence of essentially foregoing the possibility of studying segregation in nonmetropolitan communities and smaller metropolitan areas. Thus, in the past, researchers wishing to study segregation in nonmetropolitan settings have had to hope for the best and cope with higher levels of index bias than they would otherwise wish to, or otherwise avoid studying segregation in nonmetropolitan communities altogether.
Because researchers most often choose to avoid the problem by studying segregation of large subgroups in large metropolitan areas, we anticipate that many general patterns of segregation in large metropolitan areas that have been reported in previous studies will generally, though not necessarily always, be replicated in this chapter. Even so, our choice to measure segregation using data for households rather than persons, and including a larger range of group comparisons, may yield results that will expand and refine what we know about segregation in metropolitan contexts and potentially lead to new findings and insights. In contrast, we strongly anticipate that the findings we report for nonmetropolitan communities – namely, micropolitan areas and noncore counties – will be significant new additions to knowledge about segregation outside of large metropolitan contexts. This will be the case regardless of what we find because currently concerns about the challenges of measuring segregation in nonmetropolitan settings are unsettled. If findings of previous research are largely replicated, it will be valuable to know that concerns about measurement are resolved in a way that leaves previous research findings intact. If findings of previous research are not consistently replicated, it will be valuable to know that concerns about measurement were justified in some degree and research going forward must use newer, more appropriate methods of measurement to accurately document trends and patterns in segregation.
3.2 Previously Observed Trends in White-Black, White-Latino, and White-Asian Segregation
Among the major nonwhite panethnic populations in the United States, Black households have been the most highly segregated from White households across the nation on all major dimensions of segregation and thus experience conditions of hypersegregation (high levels of segregation on several of the five dimensions of segregation identified by Douglas Massey and Nancy Denton (1988)) in many of the large metropolitan areas of the Midwest and Northeast (Massey, 2020; Massey & Denton, 1989; Massey & Tannen, 2015; Wilkes & Iceland, 2004). Black segregation, particularly in urban areas, is a deeply entrenched pattern that has been molded by over a century of overt and covert discriminatory practices to exclude Black households from White neighborhoods and undervalue the neighborhoods where they reside (Massey & Denton, 1993). Although researchers in recent decades have documented steady, albeit small declines in White-Black segregation over time (Frey, 2018; Iceland, 2014; Iceland et al., 2002), patterns of White-Black segregation are still distinct and carry on serious consequences within and across generations that have resisted change to a greater degree than for any other group comparison (Sharkey, 2013). High levels of White-Black segregation enable other inequities to persist that restrict opportunity and negatively impact the well-being of Black people (Massey, 2020). According to the 2010 Census, 14 percent of the United States population identified as Black alone or in combination with one or more other races. This is a 12 percent increase from 2000, a growth rate faster than that of the U.S. population as a whole (U.S. Census Bureau 2010). The Black population is also overwhelmingly native-born, meaning that the dynamics of residential and other social outcomes play out differently than they do for non-Black Latino and Asian households because the role of ethnic enclaves supported by influxes of immigrants is less pronounced.
The Latino population is now the largest nonwhite racial-ethnic group in the United States, having grown by 43 percent from 2000 to 2010 to stand at 16 percent of the national population by 2010 (U.S. Census Bureau 2010). The majority of this rapidly growing, multicultural population is of Mexican origin, with the second largest portion of the Latino population being of Puerto Rican origin. Despite sustained and fast population growth, past research indicates Latino households have been and continue to be only moderately segregated from White households on the two most widely studied dimensions of uneven distribution and isolation (Charles, 2003; Frey, 2018; Iceland et al., 2014; Massey & Denton, 1987). However, holding uneven distribution constant, population growth resulting from both natural increase as well as immigration have necessarily led to higher levels of isolation and a decrease in exposure to White households (Charles, 2003; Massey & Denton, 1987). And while White-Latino uneven distribution has in general been moderate, overall the uneven distribution of the Latino population in metropolitan areas has not declined as observed for Black households but instead has at least remained stable and may in some cases have increased, particularly in metropolitan areas where there has been greater Latino population growth (Frey, 2018; Iceland, 2014; Iceland et al., 2014; Iceland et al., 2002; Logan & Stults, 2011). At the individual level, research suggests that group differences in social and economic characteristics may be a significant contributing factor in White-Latino segregation. These studies note that, particularly in comparison to Black households, Latino households often experience greater levels of residential contact with White households as they acculturate and assimilate on socioeconomic status (Alba & Logan, 1993; Charles, 2000; Chetty et al., 2020; Crowell & Fossett, 2018, 2020, 2022; Massey & Fong, 1990). A caveat here is that this may not hold true for Black Latino households.
While studies of Latino segregation most often give attention to “traditional” or “established” areas of Latino settlement along the Southwest border and in major metropolitan areas, Saenz (2004) and Vásquez et al. (2008) have reported evidence that Latino households in general are moving away from the traditional areas of Latino population concentration such as the Southwest and entering new areas of settlement and residence that previously did not have sizable Latino populations across the Midwest and the South. This movement has inspired a new direction in the Latino residential segregation literature as researchers have begun to examine the residential patterns of Latino households in these “new destinations” (Lichter et al., 2010). This situation is of great interest both because of the rapid growth of the Latino population nationally and the special methodological challenges involved in tracking White-Latino segregation over time in new destination communities. Accordingly, we give separate and focused attention to these trends in Chap. 5.
Although well short of approaching the absolute size of the Latino population, the Asian population in the United States also has been growing rapidly in recent decades and therefore has been receiving greater attention in studies of residential segregation. In 2010, persons who identified as Asian either alone or in combination with one or more other races comprised 5.6 percent of the total U.S. population, a 45.6 percent increase since 2000 (U.S. Census Bureau 2010). Over the past half century Asian immigration has transformed the overall U.S. Asian population from being predominately Chinese and Japanese to also including other groups who ethnically identify as Filipino, Korean, Asian Indian, Vietnamese, Cambodian, and Laotian. Still, as of the 2010 census, the majority of the Asian population was comprised of the Chinese, Asian Indian, and Filipino subgroups (Hoeffel et al., 2012). Like the Latino population, the Asian population is a fast-growing group, but this growth is primarily due to immigration and less to natural increase. In addition to a small set of immigrant “gateway” metropolitan areas and a few other areas of historical Asian presence, metropolitan areas in non-traditional areas such as the South have seen significant Asian population growth in recent decades, suggesting that there may also be an Asian new destination phenomenon emerging (Flippen & Kim, 2015; Hoeffel et al., 2012).
Past research has consistently reported comparatively low-to-moderate levels of White-Asian uneven distribution and minimal change over time, as well as relatively low, albeit rising, levels of Asian isolation with relatively high but slightly declining exposure to White households (Charles, 2003; Frey, 2018; Iceland, 2014; Iceland et al., 2014). Similar to Latino households, much of the documented change in overall contact patterns is primarily due to Asian population growth, since uneven distribution has been mostly stable over time (Iceland et al., 2014; Logan & Stults, 2011). The literature shows that among minoritized racial populations, the Asian population as a whole is generally the least residentially segregated from White households and also that Asian households experience greater residential contact with White households as they acculturate and make socioeconomic gains in comparison to other minoritized racial groups, which in turn may lead to less residential segregation (Crowell & Fossett, 2022; Massey & Denton, 1987; Massey, 2020; Sakamoto et al., 2009; Zhou & Logan, 1991). Of those who identify as Asian alone, approximately 70 percent are foreign-born, and the foreign-born Asian population makes up 28 percent of the total foreign-born population in the United States (American Community Survey 2007–2011). Because of the nature of Asian immigration to the United States, the Asian population is also highly selective on higher educational and socioeconomic standing, although certain nationalities represented in the United States such as the Cambodian and Hmong populations, who arrived in the United States in the context of violent political conflict, exhibit lower levels of socioeconomic standing on average.
Demographic trends in the Asian and Latino populations pose special problems for tracking trends in segregation over time and variation in segregation across communities. In particular, these subpopulations vary considerably in absolute and relative size across different communities and within given communities over time. These demographic patterns create the possibility that segregation comparisons using standard index scores may be impacted by index bias in complex and unwelcome ways. National and local changes in the size of the Black population have been more modest in recent decades, especially in comparison with changes that took place during the Great Migration era from 1910 to 1940. But the Black population has diffused in some degree in recent decades to areas which previously had minimal Black population presence. Our use of new methods for obtaining unbiased index scores will allow us to examine these trends and patterns with greater confidence that the variation observed is real and not artifactual, thus providing more clarity in understanding these trends and patterns. These brief demographic descriptions of the minoritized racial populations included in our analysis serve two purposes. The first is to provide context for understanding variations in segregation patterns across areas and over time. Understanding the populations involved and their characteristics allow us to go a step beyond descriptive analysis to speculate on the underlying reasons for any changes we observe. The second purpose is to acknowledge previous segregation studies of these populations which have set the basis for received wisdom regarding trends and patterns in the segregation of minoritized racial groups in the United States. Because our approach to measuring segregation differs from the approaches used in prior empirical studies in the segregation literature, we will be interested to see whether our findings track or differ from the previously established baselines regarding the level and nature of segregation and how it varies across group comparisons, across communities, and over time.
3.3 The Historical Context of Segregation
Many factors are relevant to observed variations in levels of racial and ethnic residential segregation. These include the particular groups included in the comparison and basic demographic characteristics such as the population size of the community and the relative size of the groups in question. White-Black segregation is deeply woven into the fabric of urban spatial distributions from a history of formal, legal, and institutional segregation policies operating alongside informal, extra-legal behaviors, with both being driven by overt racial prejudice against the Black population. White-Black segregation in rural areas also has a distinct character shaped by the historical role of Black Americans in the agricultural economy of the rural South and the legacy of centuries of slavery, sharecropping, and Jim Crow segregation. White-Latino segregation has been impacted by the significant increases in immigration that began in the 1960s, bringing in large numbers of new arrivals with distinctive differences in language, culture, socioeconomic status, and legal standing. This population often encounters formal and informal constraints when choosing their residential locations. In addition, the Latino population is highly heterogeneous with wide diversity in racial identity, ethnic identity, and national identity across regions and among recent immigrant populations. For example, the highest levels of White-Latino segregation are found in the metropolitan Northeast where segregation looks similar to levels of White-Black segregation. One reason for this is that the Northeast has more Black Latino individuals, who identify as Latino but who also in many cases self-identify as, or are racialized by others as, Black. Latino individuals who racially identify as White or are perceived as White are likely to experience the lowest levels of segregation.
Similar observations apply regarding White-Asian segregation, as the Asian population in the United States has a unique and complex history. Prior to recent decades, the Asian population was small at the national level and was concentrated in a select number of communities in the United States where Asian communities were often subjected to overt legal and extra-legal discrimination from the middle−1800’s up to the Civil Rights Era. But, following changes in immigration policy in the 1960s, the Asian population grew rapidly through primarily legal immigration that was highly selective in terms of socioeconomic status and skilled employment and also in some instances by refugee resettlement programs that often involved support and sponsorship. The scale of immigration was such that the Asian population shifted substantially toward having a high percentage of foreign-born and also having greater ethnic and national diversity. As a predominately foreign-born population, Asian residential settlement patterns are shaped by the economic and political conditions that contextualize immigration patterns for each Asian ethnic group that has immigrated to the United States, enclave formations, and the degree to which they experience social distance from White households. In general, but with important exceptions, the Asian immigrant population differs from the Latino immigrant population of recent decades in terms of having a higher socioeconomic profile, a smaller undocumented population, and a greater concentration in large and growing metropolitan centers with higher wages. In particular, in comparison with Latino and especially Black households of similar socioeconomic standing, high-income, high-education Asian households are likely to have greater residential contact with White households. Asian ethnic groups with lower socioeconomic resources, such as groups with a history of refugee resettlement, are likely to be more segregated from White households and less likely to reside in affluent neighborhoods.
3.4 Data
For our empirical analyses we draw on census block-level tabulations of households by race and ethnicity reported in the 1990–2010 decennial census summary files. We use these data to calculate pairwise segregation scores between White, Black, Latino, and Asian households for metropolitan and micropolitan core-based statistical areas (CBSAs) and noncore counties as defined in the 2010 census. White householders are defined as those who are non-Hispanic and who racially identified as White alone, while other racial groups include those who identify as Latino or Hispanic, as this is how household tabulations are constructed for public-use summary files. Latino householders are defined as anybody who indicated that they were “of Hispanic, Latino, or Spanish origin” (Census 2010 Questionnaire). These racial-ethnic categories were relatively stable over the 1990, 2000, and 2010 censuses with the exception that beginning in 2000 respondents could select more than one race to reflect multiracial identity, a change that resulted in less than 10% of the U.S. population identifying as two or more races in either decade.
As discussed in more detail previously in Chap. 2, we use data for households because this is the appropriate micro-level unit to use when calculating unbiased index scores. Data for persons are not generally appropriate for computing unbiased index scores for residential segregation because persons residing in multi-person households locate together, not independently. And, because households are overwhelmingly homogeneous on racial-ethnic group, the correlated locational outcomes of same-race members of households creates most of the bias in standard index scores. Basic tabulations of persons by race cannot sustain the proper calculations needed to obtain accurate unbiased index scores, but the proper calculations can be implemented using tabulations of households by race.Footnote 1 There is another side benefit of using data for households, which is that segregation scores based on tabulations of persons often include subpopulations not residing in households such as persons in institutions and/or in group quarters that sometimes can distort segregation comparisons.Footnote 2
The advantage of using county-based areas is that county boundaries are highly stable across time. For the handful of counties that changed boundaries across the three decennial census years considered in our analysis, we excluded the ones where boundary changes could not be reconciled to stable definitions over time. We also implemented the additional selection criterion of excluding CBSAs and noncore counties where the number of households for either group in the analysis is less than 50 households – which typically translates into 150–250 persons – or where the percentage share of the smaller group in the comparison is less than 0.5 percent. In comparison to selection criteria used in prior research, these restrictions are fairly liberal. This reflects the advantages of using new methods for segregation measurement. The unbiased indices we use can sustain valid, reliable measurement even when groups are small in absolute and/or relative size. In contrast, standard versions of indices of uneven distribution, and in particular the dissimilarity index (D), do not maintain acceptable behavior under similar conditions because their scores are distorted, often to a dramatic degree, by the impact of index bias.
The selection criteria, while liberal in comparison to those commonly used in previous research, still serve to screen out many logically possible combinations of group comparisons across communities. But this primarily reflects the demographic reality that for many communities the population in the community does not have the minimal numbers needed to sustain meaningful analysis of residential segregation for the excluded group comparison. After implementing these selection criteria, we are still left with a sizeable number of CBSAs and noncore counties for analysis for most segregation pairings (Table 3.1). The main exception is that very few noncore counties met the criteria for analyzing White-Asian segregation as the Asian population is overwhelmingly urban. As we review more closely below, the number of areas included varies depending on the year, area type, and the group comparison in question. Because the selection criteria we use are more inclusive, our analysis sample includes more communities and more segregation comparisons than would be the case if we used standard index scores and the more restrictive criteria needed when using standard scores. As a result, our analysis dataset is more representative of the full range of communities and group comparisons that could be considered.
3.5 Measurement
In this chapter we rely primarily on the separation index (S), which measures the dimension of evenness, or the extent to which the racial composition of neighborhoods deviates from the overall composition of the area. For comparison we also include an analysis using the more widely used dissimilarity index (D). Both indices have a fairly straightforward interpretation, especially when conceptualized in the difference-of-means formulation discussed in Chap. 2 (and in more detail in Fossett, 2017). In the case of segregation from White households, the widely used dissimilarity index can be interpreted as the difference in the proportion of each group (e.g. White households and Black households) that lives in a neighborhood where the proportion White for the neighborhood equals or exceeds parity (i.e., is equal to or greater than the proportion of the population that is White for the community overall). The separation index has an even simpler interpretation; it is the difference in the average neighborhood-level proportion White between the two groups in the analysis.
In Chap. 2 we described scenarios where the separation index and the dissimilarity index can deviate from one another. In these situations, the value of S gives the more reliable signal regarding whether the two groups in the analysis in fact live apart from each other in different spatial domains within the community – the hallmark of “prototypical segregation” which is characterized by polarized displacement from even distribution, or polarized unevenness, that can sustain group differences in location-based outcomes. In contrast, D cannot provide a reliable signal on group separation because D inherently reacts strongly to neighborhood departures from parity that are quantitatively small and thus can take on high values even when the two groups in the comparison live together in neighborhoods that are similar on neighborhood group composition and have similar levels of contact with the reference group in the comparison – a condition Fossett (2017) terms dispersed displacement from even distribution, or dispersed unevenness. The situation of dispersed unevenness always involves a particular combination of index scores; namely, a high score on D and a low score on S. We call attention to these situations for three reasons. One is that the possibility of these situations, not to mention their relatively common occurrence, is not widely appreciated by segregation researchers. The second is that, bluntly, the high value of D in these situations can be highly misleading because many incorrectly assume that a high value of D will involve group separation and the potential for group inequality in area-based outcomes (e.g., pollution, crime, opportunities, amenities, services, etc.).
The third reason is that our analyses document important systematic patterns in the occurrence of segregation involving dispersed and polarized unevenness. For example, polarized unevenness – situations where D and S are similar – are typical for White-Black segregation. This means that Black households live apart from White households in different spatial domains in the community and thus have low contact with White households and can experience location-based disadvantages that do not affect White households. This fact, plus the fact that segregation studies from the 1940s to the 1980s typically considered only White-Black segregation, may partly account for why so many incorrectly assume that high values on D indicate that groups are separated across spatial units and thus can (and may be likely to) experience group inequality on location-based outcomes. In contrast, dispersed unevenness – situations where D is high and S is low – are typical for White-Asian segregation. This means that in general, Asian households live alongside White households in the same spatial domains in the community and thus experience high contact with White households and cannot experience location-based disadvantages that do not affect White households. We also find that the situation for White-Latino segregation is more complicated because both patterns of segregation – dispersed and polarized patterns of unevenness – are common. Dispersed unevenness is common in new destination communities where Latino households are a new and relatively small presence in the community and few Latino households live in neighborhoods that are predominantly Latino. Polarized unevenness is more common in established communities where Latino presence is larger and long-standing and where it is likely that a substantial fraction of Latino households will live in neighborhoods that are predominantly Latino.
These are important distinctions because the pattern of polarized unevenness creates the maximum differences in group contact and the possibility of opportunity-hoarding and differential group disadvantage on location-based outcomes. The separation index will more reliably signal when segregation of this nature is occurring. Thus, towards the end of the chapter we comment on the empirical importance of index choice for generating findings and review how observing S and D together can be informative for describing changing patterns of unevenness between dispersed and polarized configurations. Under the sorts of conditions that we identified where index bias is prevalent and D can take high values at the same time that S does not, the two indices often do not change in the same ways over time. We are able to demonstrate in this chapter how this index divergence is not a flaw but rather is reflecting an observable pattern transition in the type of uneven distribution that is present. But even so, in situations where values of D and S differ in our empirical results, we assign priority to the value of the separation index for drawing substantive conclusions about trends and patterns in racial and ethnic residential segregation across the United States.
To interpret these scores, we use the schema shown in Table 3.2 (adapted from Fossett, 2017). The table shows the guidelines we will follow when characterizing scores for the separation index and the dissimilarity index as ranging from low to very high. One thing the table indicates is that for any given category the numerical range for D runs well above the numerical range for S. Scores for D inherently run higher than scores for S because D always responds more strongly than S when neighborhood departures from parity are not fully polarized (i.e., do not involve homogeneity of either group in the comparison). The boundary ranges for D make allowances for this. Thus, we do not characterize moderate D-S differences as indicating discordance. However, we do characterize step differences across categories – for example a high score on D (in the range 50–69) and a medium score on S (in the range 15–34) – as indicating D-S discordance, which signals the segregation pattern involves unevenness that is dispersed rather than polarized.
This chapter is also our first opportunity to empirically demonstrate the importance of using the unbiased formulations of segregation indices as described in Chap. 2 and developed by Fossett (2017). To review, nearly every commonly used measure of segregation that is based on some calculation of neighborhood-level composition is susceptible to an artificial upward bias when calculated using conventional formulas. Most segregation researchers are aware of this problem and avoid it by excluding cases where the bias is most likely to occur. The formulas we use to obtain unbiased index scores make case exclusion unnecessary as the formulas eliminate the source of upward bias that distorts scores obtained using standard formulas, thus yielding scores that can be treated as valid and reliable as given and eliminating any need to consider post hoc adjustments or differential weighting of scores across cases.
As a brief reminder, we note again that the difference-of-means formulas for calculating index scores pinpoint the sole source of upward bias in standard index scores; it is the incorporation of self-contact in the calculation of an individual household’s level of contact with the reference group in the comparison. The crux of the matter is that self-contact is fixed (it cannot be randomly assigned) and it varies systematically by group. Thus, if White households are designated as the reference group when calculating index scores for White-Black segregation, self-contact is always positive for White households and it is always zero for Black households. This creates an inherent value that is greater than zero for the group difference in average contact, even under random assignment. In contrast, contact with White households among others (excluding the focal household) will have the same expected value for White households and Black households under random assignment and thus has no impact on index bias. Revised formulas reviewed in Chap. 2 and in Fossett (2017) eliminate self-contact from index calculations and in so doing yield unbiased index scores (i.e., scores that have expected values of zero under random assignment). When unbiased scores differ from standard scores, the unbiased scores should be preferred. If bias is not a problem, the scores will not disagree. Therefore, the optimal choice is to use the unbiased indices.
3.6 Trends and Patterns of Racial and Ethnic Residential Segregation, 1990–2010
We begin by reviewing levels of White-Black, White-Asian, and White-Latino segregation as measured by the separation index (S), which are summarized in Table 3.3 and presented by decade and type of community alongside the more familiar dissimilarity index (D) in Table 3.4. While the separation index is our optimal index for measuring segregation and what we use to draw substantive conclusions about patterns and trends of residential segregation, we recognize that most readers are more familiar and comfortable with the dissimilarity index and that it has been the index behind much of what we know from the literature so far about residential segregation. Thus, we include it in Table 3.4 so that we can further explain why we prefer the separation index in comparison to the dissimilarity index and what impact that has on our findings and conclusions. Our choice to begin by examining Black, Latino, and Asian segregation from White households also warrants explanation. These three White-nonwhite group comparisons are a central focus in segregation research in the United States because residential segregation among racial-ethnic groups is a stratification outcome and is closely linked to group position and group inequalities across a wide range of location-based outcomes including basic living conditions, exposure to crime and social problems, amenities, social and economic opportunities, political influence, quality and responsiveness of government services, and more (Stearns & Logan, 1986; Massey & Denton, 1993; Firebaugh & Farrell, 2016; Krysan & Crowder, 2017). Given the White population’s historical standing as the majority group in racial-ethnic relations in the United States, predominantly White neighborhoods have consistently been found to be advantaged on location-based outcomes, and residential separation from White households has consistently been found to be associated with related White-nonwhite disparities and broad systematic disadvantages for nonwhite groups, especially for Black households. This context for segregation theory and research makes the separation index an excellent choice for measuring uneven distribution because, among all widely used indices, S best indicates when groups occupy different residential spaces, thus creating the conditions that make group inequalities in location-based outcomes possible.
After reviewing White-nonwhite residential segregation patterns, we will next examine segregation between nonwhite groups including Black-Latino, Black-Asian, and Latino-Asian residential segregation patterns from 1990 to 2010. It is not common for segregation studies to focus on patterns between nonwhite groups for the theoretical reasons stated above and also due to the methodological challenges described in Chap. 2 as well as in the previous sections of this chapter. Finally, in this section we also include some discussion regarding the impact of our measurement approaches on our empirical findings and provide explanations to reconcile findings that may differ from what has been previously asserted in the literature on residential segregation patterns and trends over time. The two primary issues here are the extent to which index bias has affected previous studies and the inherent shortcomings of the dissimilarity index, which has been the workhorse of segregation research for many decades. Both issues, fortunately, are fairly simple to explain and resolve.
3.6.1 White-Black Segregation
Overall, we find segregation between White and Black households to be the highest among the three White-nonwhite comparisons considered in this analysis and by a large margin in both relative and absolute terms. The values of the separation index for White-Black segregation vary widely across the communities in our study with a mean of 36.2, a median of 38.6, a standard deviation of 21.3, and an inter-decile range of 57.3 points extending from 5.6 to 63.0 at the 10th and 90th percentiles, respectively. The typical level of White-Black segregation across all communities is at medium-to-high levels. Noncore counties in 1990 have the highest average level of White-Black segregation for the groupings of communities reported in Table 3.4 at 49.2 points on S. Based on the guidelines we presented in Table 3.2, this is a high level of segregation. Substantively, a value of S on the order of 49 means that, for the community in question, the relative presence of White households among neighboring households is 49 points lower for the average Black household compared to the average White household.
Values of S in this range provide a clear signal that White-Black segregation consistently involves polarized unevenness, which Fossett (2017) terms prototypical segregation because it is the pattern that immediately comes to mind for broad audiences and segregation researchers alike when they are told the level of segregation is high. In this prototypical pattern, White and Black households generally reside in different neighborhoods where their neighbors are predominantly from their own group. This then creates the structural precondition for White and Black households to experience substantial differences on location-based outcomes. However, while we find high and prototypical White-Black segregation in noncore counties, the more commonly experienced outcome across areas is medium levels of segregation, especially by 2010. The separation index for micropolitan and metropolitan areas in 2010 averages below 30, which is firmly within the moderate range. This finding would appear to be in conflict with what past research has found using the dissimilarity index, and indeed the dissimilarity index, even after correcting for index bias, remains at high levels for all areas in every decade. The discordance between the two indices grows larger over time. This finding has a simple explanation when it is understood that the separation index responds to patterns of displacement from even distribution in a way that the dissimilarity index cannot. As the separation index drops to medium levels while the dissimilarity index stays high, the underlying patterns driving this change are shifting from polarized to dispersed unevenness. Black households are still typically living in neighborhoods that are below parity on proportion White, but they are having more residential contact with White households in their neighborhoods over time.
The finding that White-Black segregation is highest among White-nonwhite comparisons holds across all three decades and all three community types. Regarding the level of segregation, we find that the average values of S for White-Black segregation are higher than the average values of S of other White-nonwhite comparisons by at least 25 points, a very large amount in both absolute and relative terms. Regarding changes over time, the average value of S for White-Black segregation declined substantially over the decades and falls by an average of 10 or more points from 1990 to 2010. Declines of this absolute magnitude are substantively important in their own right. They are equally if not more important when considered in relative terms as the average declines in raw scores represent relative declines of 20–25 percent over the two decades.
As for variation across types of communities, White-Black segregation as measured by S is highest in noncore counties compared to CBSAs by about 9–10 points. Among CBSAs, segregation is higher by 1–4 points in metropolitan areas compared to micropolitan areas. In 1990, the mean levels for all three areas were in the high range of 35–59 given in the schema in Table 3.2, with noncore counties in the top half of this range and CBSAs near the lower end of the range. Even after sizeable declines over the decades, White-Black segregation in noncore counties remained in the high range in 2010. In contrast, the similar declines of 10 or more points for White-Black segregation in metropolitan and micropolitan areas dropped the averages for these communities to below 30 and into the medium range of 15–34 points.
Because of the widely studied and understood circumstances of White-Black segregation that we are familiar with, where polarized unevenness is more common even in smaller, nonurban communities, we did not expect to uncover a story about White-Black segregation that is much different from what past studies have shown. When the pattern of unevenness is polarized, the separation index and the dissimilarity index will be in closer alignment. What we can conclude is what others have found, which is that White-Black segregation is on the decline from initially high levels across all communities, although these two groups still remain the most segregated from one another. But by focusing on the separation index, we contribute an added detail to our understanding of White-Black residential segregation and its trends over time. It is that in addition to the groups gradually becoming more evenly distributed, their pattern of unevenness is also becoming more dispersed. This means that overall, Black households are having more equal levels of residential contact with White households.
3.6.2 White-Latino Segregation
White-Latino segregation is a more complex and surprising story given the findings consistently reported in the literature of moderate and persistent White-Latino segregation with levels of segregation a bit below White-Black segregation. Therefore, this section warrants a more extended discussion. In contrast to previous findings, we find values of the separation index for White-Latino segregation are at levels well below White-Black segregation and vary in a narrower range across the communities in our study with a mean of 12.3, a median of 8.7, a standard deviation of 11.2, and an inter-decile range of 27.5 points extending from 1.2 to 28.7 at the tenth and 90th percentiles, respectively. On average, we find White-Latino segregation to be generally low, only approaching medium levels for noncore counties in 1990 and for metropolitan areas in 2000 and 2010. Previous reports measuring segregation of persons using conventional formulas have often reported medium and sometimes even high levels of White-Latino segregation, which understandably may raise questions about the low scores we produce here.
To address this likely concern, we note the following points. The first is that most previous descriptive studies of segregation have focused on the largest metropolitan areas in the United States, whereas our analysis extends beyond the large metropolitan context to include smaller metropolitan areas, micropolitan areas, and noncore counties because our methods can sustain meaningful analysis of segregation patterns in communities not considered in previous studies. In that regard, our analysis sample is more representative of the full range of White-Latino segregation across communities where Latino populations are present. Many of the communities that are often excluded in previous studies are less segregated than the large metropolitan areas that are more likely to be included in most previous studies. However, the inclusion of more communities, especially communities with Latino populations that are smaller in absolute and relative size, is a contributing factor, but not the overriding factor because the average scores we report for metropolitan areas also are lower than previous studies might lead readers to expect. Several other measurement practices we reviewed in Chap. 2 are more relevant. Two are especially important. The first is that we find White-Latino segregation is especially susceptible to distortion by index bias. The second is that, to a much greater degree than we expected, White-Latino segregation in many communities involves a pattern of dispersed unevenness instead of polarized unevenness as seen more commonly in White-Black segregation.
Regarding index bias, we highlight the following points. Many communities have relatively low levels of Latino presence in the local population. All else equal, this factor leads to higher levels of upward bias in the standard version of the dissimilarity index used in most previous research. In addition, Latino households tend to be larger than White, Black, and Asian households. Thus, the impact of bias on standard index scores for the communities in our study is much greater for White-Latino segregation than for White-Black segregation. And, conversely, the consequence of using unbiased index scores calculated using data for households instead of persons reduces the average scores for White-Latino segregation to a greater degree than for White-Black segregation. For example, for metropolitan CBSAs in 2000 (as a subset of cases included in many previous studies), the reduction in D based on eliminating index bias averages 13.7 points for White-Black comparisons and 26.9 points for White-Latino comparisons.
As for the greater prevalence of dispersed versus polarized displacement from even distribution, most previous studies do not acknowledge this aspect of segregation, so it is not surprising that its prevalence in White-Latino segregation is not appreciated. This leads to a situation where multiple factors contributed to the adoption of understandable, but unfortunately incorrect, assumptions about White-Latino segregation patterns. Didactic discussions of segregation measurement leading to high index scores invariably feature patterns characterized by polarized unevenness which produces clear group separation and prototypical segregation. In this situation, all index scores, including both D and S, take high values. Landmark studies such as Duncan and Duncan (1955) and Massey and Denton (1988) suggest D correlates closely with alternative indices and do not stress that D and S markedly differ. The few studies that did note the possibility that scores for D and S can diverge (e.g., Stearns & Logan, 1986) did not make the point as forcefully as might have been possible and thus had limited impact on segregation measurement practices. Thus, it was not until Fossett (2017) that a methodological study provided comprehensive evidence that scores on D and S not only can diverge, but also frequently do in studies that consider a wider range of communities and group comparisons than were considered in previous methodological studies.
This leads us to the present study where we report the finding, surprising to many for the reasons just reviewed, that divergence of scores for S and D characteristic of dispersed unevenness is much more common in White-Latino comparisons than in White-Black comparisons. Consequently, our use of the separation index to identify the extent to which groups occupy different neighborhoods in the community yields scores that are much lower than the scores of S we found for White-Black segregation. One reason why this finding is important, other than what it says about previous analyses of White-Latino segregation, is that it complicates analyses of trends in White-Latino segregation in new destination communities where Latino presence is relatively recent but is growing rapidly. We review the topic of new destinations in more detail in Chap. 5 and show that index choice turns out to be highly consequential for understanding segregation trends in new destinations, as S and D can lead to opposite conclusions if not understood correctly.
While While-Latino segregation was low across all three decades and across all community types, the trends over time varied by community type. White-Latino segregation slightly declined in noncore counties, remained stable in micropolitan areas, and increased in metropolitan areas. The highest levels of White-Latino segregation in 1990 are observed in noncore counties, many of which were experiencing the arrival of Latino migrants and immigrants in predominately White rural communities across the Midwest and South, which are more likely to be new destinations – again explored further in Chap. 5. By 2010, the average level of White-Latino segregation is highest in metropolitan areas due both to rising segregation in those areas and declining average segregation in noncore counties.
While White-Latino segregation is generally low across the communities in our study, segregation does reach medium levels (S ≥ 15) in many communities and high levels (S ≥ 35) in a smaller subset of communities. Not surprisingly, this includes large well-known metropolitan areas with S > 45 (e.g., Chicago, IL, Los Angeles, CA, and New York City, NY), smaller, less well-known metropolitan areas with S > 45 (e.g., Bakersfield, CA, Brownsville, TX, McAllen, TX, and Salinas, CA), micropolitan areas with S > 35 (e.g., Del Rio, TX, Dodge City, KS, Liberal, KS, Lumberton, NC, Nogales, AZ, and Uvalde, TX), and noncore counties with S > 35 (e.g., dozens of counties across states such as Arizona, Colorado, Georgia, Nebraska, New Mexico, North Carolina, and Texas). We point this out as reassurance that many basic patterns from past research carry forward when segregation is measured without bias and using S instead of D. And, of course, the much higher scores for White-Black segregation reviewed earlier also reinforce this point. In sum, our study finds White-Latino segregation to be lower than past studies might suggest because scores reported in past studies, especially outside of large metropolitan areas, are substantially inflated by index bias and because a pattern of dispersed unevenness, where scores for S are much lower than scores for D, is much more common for White-Latino segregation than for White-Black segregation.
The Latino population was growing rapidly at the national level from 1990 to 2010 and diffusing into areas of the country where Latino presence had previously been limited or absent altogether. This often played out as dramatic growth in new destination communities that met the criteria for inclusion in 1990 and it also led to new communities first meeting inclusion criteria in 2000 or 2010. This was most common in noncore counties where the number of communities meeting our minimum household count criteria more than doubled from 1990 to 2010. This raises the question of whether the addition of new qualifying cases in 2000 and 2010 impacts the findings we reviewed earlier. We addressed this question by performing our descriptive analysis using only the set of communities that met criteria for inclusion over the full time period from 1990 to 2010. We present these results in Table 3.5 and find that White-Latino segregation in noncore counties remained stable over the time period. This indicates that the newer areas of Latino settlement that emerge over this time frame appear to be driving the declines in White-Latino segregation in noncore counties seen in Table 3.4. Thus, Latino new destination communities that emerged most recently have lower levels of segregation than is seen in new destination communities where Latino migrants and immigrants began arriving in significant numbers at an earlier point in time. This suggests that segregation in the most recently emerging new destination communities will rise to the levels observed in the Latino new destinations that are further along in the process of transitioning to areas of established Latino presence. Patterns for micropolitan and metropolitan areas are more consistent between communities that were included in all three decades of analysis and communities that joined the analysis in later decades due to population growth.
We conclude our descriptive findings for White-Latino segregation by noting that there is a much larger set of communities outside of the most urban-populated metropolitan areas which have markedly lower levels of segregation with weaker spatial boundaries, particularly for non-Black groups. Thus, in the case of White-Latino segregation, the typical community does not reflect the levels of White-Latino segregation observed in metropolitan areas like Chicago or Los Angeles, where the separation index hovers between 40 and 50 and would indicate high levels of segregation. By moving beyond the context of segregation in a selective group of large metropolitan areas, which has had a major influence on how we think about residential segregation in the United States, we can develop a new narrative of the reality of Latino segregation patterns across the United States.
3.6.3 White-Asian Segregation
Previous studies consistently report that Asian households experience the lowest levels of segregation from White households, and this result is replicated in our findings with the stipulation that, as with our findings for White-Black and White-Latino segregation, the levels of White-Asian segregation we find are lower than those reported in the literature. Some, but not all, of the reasons for this pattern are the same that we noted earlier for White-Latino segregation. First, moving to using unbiased index scores leads to lower measured levels of segregation. Second, moving from using D to using S leads to lower measured levels of segregation as well because for White-Asian segregation, more so than any other of these group comparisons, the underlying pattern involves dispersed rather than polarized unevenness. This pattern produces high-D, low-S combinations as the norm, not as an exception. Thus, while average values of D for White-Asian segregation are actually similar to average values of D for White-Black and White-Latino segregation when scores are calculated using standard formulas, average values of S for White-Asian segregation are much lower than average values of S for the White-Black and White-Latino comparisons and the divergence across group comparisons is even greater when scores are calculated using formulas that yield unbiased scores.
The consequence is that high values of D are particularly misleading for White-Asian comparisons because they rarely occur in combination with high values of S, as occurs frequently for White-Latino segregation and is the norm for White-Black segregation. Thus, values of unbiased S above 35 are rare in our data for White-Asian segregation. Of the communities in our analysis, only three have scores of at least 35, in two of the three decades. These communities are the micropolitan areas of Bay City, TX, Garden City, KS, and Morgan City, LA. Each one has an atypical history involving refugee settlement of a single Asian nationality subgroup. Dispersed unevenness is the norm for White-Asian segregation. The quantitative signature of the pattern is the combination of high-D, low-S. This is clear from the fact that, across all White-Asian comparisons in our analysis, the averages for the unbiased versions of D and S are 36.9 and 7.7, respectively. The unbiased scores are more accurate and meaningful. But we also note the averages for the standard (biased) versions of D and S are 74.0 and 14.3, respectively, to establish that the dramatic D-S difference is not specific to unbiased scores. The distinction between patterns of dispersed and polarized unevenness has been overlooked in past research. But it has important substantive implications. The occurrence of high values of D for White-Asian segregation rarely occurs in combination with a high level of group separation wherein White and Asian households occupy different neighborhoods that are polarized on group composition, such that the two groups could experience systematically different location-based outcomes. Instead, higher values of D for White-Asian segregation result because Asian households generally live in neighborhoods where they have contact with White neighbors at levels that are quantitatively close to parity, but technically are below parity. The dissimilarity index is highly sensitive to these near misses on parity. But, since this pattern does not create the level of group difference in contact that can create prototypical segregation where neighborhoods are polarized on group composition, the separation index takes very low values.
The prevalence of the pattern of dispersed unevenness in White-Asian segregation is evident in other findings. One is that, across most communities where D is high, S is low. Consequently, few neighborhoods are predominantly Asian – a requisite for group separation and group disparity on location-based outcomes. This is true when assessed in relation to total population and even when assessed on just the combined pairwise White and Asian populations. This stands in stark contrast to the pattern of polarized unevenness that is the norm in White-Black segregation. This pattern involves a substantial portion of Black households residing in predominantly Black neighborhoods as occurs when groups occupy different neighborhoods, which can create the structural potential for group disparity on location-based outcomes. Additionally, in our review of micro-level attainment processes across selected metropolitan areas in Chap. 6, we document that Asian households attain contact with White households at near-parity levels because the average levels on relevant resources for attainment are similar across White and Asian households and, equally importantly, Asian households convert these resources into contact with White households at much higher rates than Black households.
In comparison to White-Black and White-Latino segregation, we find that White-Asian segregation is more uniform across community types and over time. Across all community types, White-Asian segregation on average remains at very low levels with slight fluctuations that do not suggest any important trend over time. Similarly, there are only very small and inconsistent differences in levels of White-Asian segregation between noncore counties, micropolitan areas, and metropolitan areas, although the dynamics that drive these patterns may be quite different. Large metropolitan areas are more likely to be home to established Asian enclaves that contribute to residential separation, while micropolitan and noncore counties may be experiencing new enclave formation but also potentially more conflict dynamics if they are predominately White areas that are adjusting to the arrival of new minoritized racial populations.
Our major finding that White-Asian segregation based on households is very low is markedly different from past research. Previous reporting on White-Asian segregation has often reported that White-Asian segregation, while the lowest in comparison to segregation between other groups and White households, is still at moderate levels (Frey, 2018; Iceland, 2014). One reason for this seeming discrepancy is that past research is often only looking at metropolitan areas, but even in metropolitan areas we find that White-Asian segregation is quite low. We bring up two methodological points to explain these low scores. First, removing the upward bias from the segregation index can reduce scores dramatically, particularly when one group in the analysis is disproportionately smaller. Measuring White-Asian segregation is fraught with issues of index bias. Second, our choice to use the separation index means that we are more reliably capturing the extent to which the two groups are actually living apart from one another. As we explained in Chap. 2, uneven distribution does not necessarily mean that the two groups in the analysis are living in meaningfully different neighborhoods. Compared to the oft used dissimilarity index, which is the measure behind the most cited findings in the literature, the separation index is more likely to reflect polarized unevenness. Uneven distribution where the minoritized racial group is in fact living in neighborhoods that approach parity with the overall area on proportion White, dispersed unevenness, will not result in high scores on the separation index, as it might with the dissimilarity index.
The role of new destinations which plays prominently in our understanding of trends of White-Latino segregation is also relevant for our analysis of White-Asian segregation. We note a less pronounced but similar pattern to that found in our analysis of White-Latino segregation, which is that the number of counties that meet our criteria for inclusion increased from 1990 to 2010. As in the case for White-Latino segregation, this is also because of population growth and migration. While the trend of migration to rural communities observed in the Latino population is not as prominent for the Asian population, there is evidence of some dispersal away from metropolitan areas, as reflected in the significant increases of micropolitan and noncore areas eligible for inclusion in our analysis from 1990 to 2010. This trend emphasizes the increasingly important but understudied question of Asian residential patterns in nonmetropolitan communities.
For the few noncore counties that remained consistently in the analysis from 1990 to 2010, segregation remained low but increased over time. In contrast, when we look at all noncore counties, including areas that emerged as cases for analysis in 2000 or 2010, White-Asian segregation appeared to remain stable. This is also true for micropolitan and metropolitan areas, with one difference being that segregation on average is declining for micropolitan areas when all areas are included, whereas it is slightly rising in areas that are in the analysis across all decades. Similar to what we found in our analysis of White-Latino segregation in noncore counties, newly emergent sites of Asian settlement may initially experience relatively lower levels of segregation but see segregation increase over time. The story of rising White-Asian segregation in noncore counties is exemplified at the extreme when we look at maximum scores. The maximum observed White-Asian segregation score for noncore counties was 50 in 1990 but reached a high of 76 by 2010, while for metropolitan areas the maximum score was 49 in 1990 and 45 in 2010. Asian segregation in nonmetropolitan counties will be a subject of deeper investigation in Chap. 4 as well as Chap. 5, where we focus on nonmetropolitan segregation and segregation in new destinations for minoritized racial groups, respectively.
3.6.4 Segregation Between Minoritized Racial Groups
In this section of the chapter we review patterns of segregation between minoritized racial groups: Black-Asian, Black-Latino, and Latino-Asian segregation (Table 3.6). One remarkable finding from these results is that in general average levels of Black-Asian and Latino-Asian segregation are significantly higher than average levels of White-Latino or White-Asian segregation, reflecting medium segregation levels that we more often expect to find when looking at segregation from White households. Black-Asian and Latino-Asian segregation has remained low-to-medium and stable, with the exception of Black-Asian segregation in noncore counties where fluctuations are likely a result of the small number of cases included in the analysis. Generally, Black-Latino segregation has been on the decline across all areas, beginning at medium levels in 1990 and moving towards low levels by 2010. The overall declines in Black-Latino segregation fit with the patterns where Latino households have lower levels of separation from White households (albeit trending upward) than do Black households, and where White-Black segregation is trending down more strongly than any group comparison we consider.
Previous research has largely neglected review of segregation among nonwhite groups. In part this has been due to concerns about index bias when measuring segregation of small subpopulations. This problem has been addressed and is no longer a constraint on research. Neglect of this topic also may have been due to the fact that until recent decades, most communities were closer to being mono-ethnic or bi-ethnic than multiethnic in group composition. However, as the Latino and Asian populations have grown in both absolute and relative size and as these populations have increasingly diffused spatially beyond an initially smaller set of regionally concentrated locations, multiethnic communities are more common and will steadily grow in prevalence.
We hope the patterns we document here will be incorporated into future research because they are potentially valuable for providing a more complete description of how spatial residential distributions vary by race and ethnicity. It also may be valuable for thoughtfully reviewing and potentially refining theories of racial-ethnic segregation. Simply put, the dominant prevailing perspectives guiding segregation research have not been applied to the analysis of segregation among nonwhite groups. The fact that our descriptive analysis documents levels of segregation among these subpopulations that sometimes approach or equal White-nonwhite segregation raises questions about whether theories of segregation primarily crafted to explain White-nonwhite segregation will need revision to explain a wider range of segregation patterns.
3.6.5 Where Is Segregation Rising? Where Is It Declining?
Because segregation is primarily shaped by the context of the given community, which includes migration patterns, local zoning and housing policies, patterns of residential development, dominant racial ideologies, and local history, changes in levels of segregation do not happen uniformly across the United States. To understand how segregation is shifting in any single community would call for a deeper and more qualitative analysis. But from a more macro-level demographic vantagepoint, we can ask a basic question: how are changes in segregation varying across the United States? In Tables 3.7, 3.8, and 3.9 we tabulate communities by community type and by broad categories of segregation change from 1990 to 2010 based on the unbiased separation index. Areas were categorized as “stable” if segregation changed by no more than 2 points in either direction.
For White-Black segregation, the vast majority of communities have experienced small but steady declines in segregation, regardless of community type. This includes two-thirds of micropolitan and metropolitan areas and over three-quarters of noncore counties. Less than 10 percent of areas have experienced increases in White-Black segregation. Given how high White-Black segregation typically is, these results are not necessarily surprising. With weakening effects of institutional discrimination in the housing market and an increasing amount of housing stock built after fair housing laws were enacted, and the emergence of a significant Black middle class in many communities, we would expect some reductions in White-Black segregation from levels that are initially very high. This is made most poignantly clear in Fig. 3.1, where we can see that White-Black segregation is declining across wide swaths of the United States. Increases in segregation are occurring in only scattered pockets along the Midwest and in the Northeast. Nevertheless, our previous results show that White-Black segregation is persisting at the highest levels, implying that these small reductions over time are indicators of slow progress toward White-Black integration.
Changes in White-Black segregation, 1990–2010
Changes in White-Latino segregation are more varied, but one clear finding is that declining segregation is the least common scenario across all community types. Over a quarter of communities have had stable levels of White-Latino segregation from 1990 to 2010 while roughly a third of communities have seen large increases (over 5 points) in White-Latino segregation over the decades. White-Latino segregation as measured using the separation index, corrected for index bias and using data for households, is markedly lower than previous studies have suggested, but it is rising and, in many communities, it is rising quickly. This trend is most pronounced in metropolitan areas, where 45 percent have had separation index score increases by more than 5 points over the decades. Unlike White-Black segregation, which has historically been high in metropolitan settings but declining, over half of metropolitan areas are seeing greater residential separation between White and Latino households. Furthermore, nearly a third of micropolitan areas and over a third of noncore counties are also seeing large increases (over 5 points) in White-Latino segregation. These communities may be less likely to have historically established Latino populations, but as Latino migrants and immigrants continue to spread outward across the United States to smaller, nonmetropolitan communities, segregation patterns are emerging. Indeed, Fig. 3.2 shows that areas of rising White-Latino segregation appear to be most concentrated in the South and parts of the Midwest. Declining White-Latino segregation is primarily occurring in the Southwest along the U.S.-Mexico border in Texas, New Mexico, and Arizona (Fig. 3.3).
Changes in White-Latino segregation, 1990–2010
Changes in White-Asian segregation, 1990–2010
Decline is also the most unlikely scenario for White-Asian segregation, with only 12 percent of all communities experiencing declines in White-Asian segregation from 1990 to 2010. Though, to be clear, this is partly due to the fact that White-Asian segregation is generally at such low levels it would not be easy for S to decline by 5 or more points in many communities. An equal portion of communities are experiencing either stable or increasing White-Asian segregation with a fifth of all communities seeing White-Asian segregation increase by more than 5 points in two decades. Although there is only a small number of noncore counties included in our analysis for White-Asian segregation, their patterns mirror those of micropolitan and metropolitan areas in that White-Asian segregation is most likely to be stable, with the second likely outcome being rising segregation. In sum, we find that White-Asian segregation is generally quite low, with Asian households on average having high levels of residential contact with White households, and these patterns appear to be either holding steady or trending towards increasing segregation.
3.7 Community-Level Analysis of Segregation Patterns
In this section of the chapter we attempt to further clarify the segregation patterns we have documented in the preceding descriptive analyses by reviewing community-level regression analyses we estimated to explore how variation in segregation across communities may correspond with variation in other characteristics of communities. We are cautious, for both methodological and theoretical reasons, in our approach to specifying community-level regressions predicting segregation. In particular, we have concerns about including aggregate-level predictors that measure group differences on characteristics that are relevant for segregation based on the role they are hypothesized to play in micro-level attainment models. Past research investigating community-level variation in segregation has sometimes tended in the direction of, perhaps inadvertently, framing segregation as an aggregate-level phenomenon. In part, this was due to the fact, noted as early as Duncan and Duncan’s (1955) landmark article on segregation measurement, that the specific quantitative links between segregation indices and micro-level processes that gave rise to segregation were unclear. The introduction of the difference-of-means framework in Fossett (2017) changed this state of affairs. After first establishing that all popular segregation indices can be given as group differences on average levels of scaled contact with the reference group, Fossett (2017) then took the next logical step of pointing out that segregation index scores reflect group disparities and thus can be mathematically equated to the effect of (regression coefficient for) group membership in a micro-level regression predicting scaled contact with the reference group.
The difference-of-means formulation of segregation index scores clarifies how such effects can be directly estimated in micro-level analyses of the type we review in Chap. 6 and in Crowell and Fossett (2018, 2020, 2022). These analyses permit direct estimation of how group differences on micro-level characteristics ultimately impact segregation and the results document that the impact is often negligible even when the group disparity on the characteristic is large. Simply put, this is because the very communities where group differences on characteristics potentially relevant for locational attainment – for example, education, income, nativity, English-language ability, etc. – are largest also tend to be communities where minoritized racial groups are less able to translate these resources into more residential contact with White households. In such communities, eliminating inequalities in resources will have minimal to no impact on group differences in locational outcomes. The implication of this is that the correlation of group inequalities at the aggregate level is primarily a spurious relationship resulting from the multiple inequalities that are all shaped by a broad pattern of constrained and stratified opportunities (Fossett, 1988, 2017; Fossett & Crowell, 2018). Understanding this also opens the door to a wide range of analysis possibilities including detailed micro-level regressions investigating the factors contributing to segregation in a single community as we review in Chap. 6. Another new possibility is to estimate contextual and multi-level regression analyses exploring how the effect of group membership (race) on contact with the reference group varies across communities.
Given all this, we use conservative specifications of aggregate-level regressions that avoid including variables that should instead be taken into account in more complex multi-level models. And we also avoid drawing overly confident conclusions regarding how segregation patterns are determined by community-level factors based on aggregate-level regressions that cannot accurately control for the impact of group differences on individual characteristics. We estimated fractional regression models predicting community-level segregation measured by the separation index, pooling all communities, group comparisons, and years with dummy variables for community type, pairing, and time. For community-level covariates we draw on previous research and include the predictors of population size, region, percent Black, percent in armed forces, percent of housing units built in the last 10 years, and a set of workforce characteristics based on industry of occupation, including percent in government, percent in manufacturing, percent in retail, and percent in service. We studiously avoid including community-level predictors whose relevance derives from the hypothesized role individual-level characteristics play in micro-level attainment processes within individual communities. The covariates used in this model are described in Table 3.10.
The results from this pooled model in Table 3.11 indicate that segregation overall has dropped significantly from 1990 to 2010, although we know from our descriptive tables that White-Latino and White-Asian segregation outcomes are more varied with many communities experiencing stable or rising segregation for these groups. We also find that segregation does not significantly differ by community type when population size, which has a positive effect on segregation, is included as a predictor. Taking communities in the Northeast as the point of comparison, we find that segregation is on average higher in communities in the South. This is perhaps surprising given the high levels of segregation often observed in metropolitan areas of the Northeast, but the South has seen widespread increases in White-Latino segregation as a result of increasing Latino migration to the South.
The results we report show several other factors have significant associations with levels of segregation when holding other contextual characteristics constant, including percent Black, which is a positive predictor of segregation, and percent in the armed forces, which is a negative predictor. The integrating effect of military presence is one that has been suggested in the literature before, particularly with regards to intermarriage. Communities with a larger military presence appear to also have more integrated neighborhoods even when looking specifically at household data (which excludes persons residing in military barracks). Our previous research on locational attainments, and our extensions of that work which we report in Chap. 6, have also shown that nonwhite householders who have served in the armed forces are more likely to have greater residential contact with White households, while White householders who have served in the armed forces have less residential contact with other White households (Crowell & Fossett, 2018, 2020, 2022).
These findings together indicate that individuals who have served in the armed forces and therefore have been placed in more diverse settings are more likely to seek out and feel more comfortable in integrated neighborhood environments.Footnote 3 Finally we find that the percentages of the workforce who are in the manufacturing, retail, or service industries are negative predictors of segregation, with areas where the industrial composition of the workforce is more diverse perhaps being those that are likely to be included in our analysis of White-Latino and White-Asian segregation, which is often quite lower than White-Black segregation in large metropolitan areas where the manufacturing industry no longer dominates.
The purpose of these models is to identify the contextual characteristics of areas that are associated with cross-community variation in levels of segregation and describe the nature of those relationships. We find, as previous research has, that segregation is associated with community racial composition, population size, military presence, and industrial composition. This analysis should be seen as a step toward a more satisfactory analysis that investigates the impact of community characteristics in a modeling framework that can correctly take account of individual- and household-level characteristics that are relevant in micro-level processes of household locational attainments.
3.7.1 Aggregate-Level Predictors Not Considered
To elaborate on our point regarding how to model segregation outcomes at macro-and micro-levels, we add more discussion here on a category of variables that we intentionally excluded from the candidate list of community-level predictor/control variables we considered for inclusion in the aggregate-level regressions we estimated to explore cross-community variation in segregation. Specifically, we excluded predictors that measured group differences on resources (e.g., group inequality on income) and social characteristics (e.g., English language ability, foreign born status, etc.). Our reason for excluding these predictors is not because they are irrelevant to segregation. To the contrary, our own analyses presented in Chap. 6 document that these variables can in some cases be highly relevant to shaping the level of segregation in a community. Instead, we excluded these predictors because their true impact on segregation cannot be accurately estimated using aggregate-level regressions. The practice of including such measures of this type in aggregate regressions predicting segregation is widespread. Accordingly, we could cite many examples, but we do not wish to call attention to a few studies when in fact the practice is common and in general is not seen as controversial. In light of this, we explain our basis for viewing this practice as flawed and likely to lead to erroneous conclusions about the determinants of segregation. We view the practice in question as a specific example of a broader flawed practice where researchers estimate aggregate-level regressions that use one or more measures of group disparity to predict a particular measure of group disparity of interest. In the interest of economy of discussion, we focus on the example of aggregate-level regressions that use measures of White-Black income inequality to predict measures of White-Black segregation.
The practice of estimating and “controlling for” the impact of income inequality on segregation in aggregate-level regressions is inherently flawed and prone to yield results that grossly overestimate the impact of income inequality on segregation. The core issue is that one cannot accurately estimate the impact of income inequality on segregation based solely on knowing the level of income inequality. An accurate estimate requires detailed knowledge of how locational attainments for each group vary with income separately in each community in the analysis. There are other methodological discussions which review the point in more careful detail (Fossett, 1988, 2017; Fossett & Crowell, 2018), but here we highlight two fatal problems with aggregate regression analyses of segregation that include measures of income inequality as predictors. The first is that the strategy implicitly assumes that co-residence with White households varies significantly for Black households by level of income, and that this relationship is uniform across communities. To put simply, these assumptions are untenable. Analyses of detailed microdata for individual communities indicates Black co-residence with White households tends to be low across all levels of income, and the pattern is consistent across communities. This fact leads to the inescapable conclusion that White-Black income differences on segregation are inconsequential.
The second fatal problem with the aggregate-regression specification is that there are compelling reasons to conclude that White-Black disparities across different domains of social and economic attainment will be spuriously correlated across communities. This is because theory predicts racial stratification dynamics in a community will have broad impacts across all attainment processes. As a result, measures of White-Black disparity across different domains of social and economic attainment will be strongly and spuriously correlated because they all have a common cause; their values rise and fall together depending on the intensity of the racial stratification system that constrains Black opportunities and attainments in the community. Thus, for example, it would be utterly implausible to suggest that White-Black residential segregation in a community in the Jim Crow South was due to White-Black income inequality. Income inequality and residential segregation would both be high under the Jim Crow racial caste system. But in this context, an intervention that increased Black incomes (but otherwise did not change the local racial stratification regime) would not lead to a reduction in White-Black segregation.
Once segregation is equated to a group inequality on a micro-level attainment outcome (per Fossett, 2017), it immediately follows that the correct way to take account of the effects of group differences on individual-level characteristics is within contextual or multi-level models that directly estimate the impact of the relevant covariate on the attainment outcome across individuals while allowing the effect to vary across communities (where it will be minimal in some and stronger in others) and also including community-level characteristics as predictors (Fossett, 1988, 2017). Unfortunately, the correct models are not easily implemented because they require large samples of detailed microdata across a large number of communities. Relevant data are available so the task is in fact feasible, but it is a major undertaking one or two orders of magnitude more difficult than estimating an aggregate-level regression. This is the unfortunate but hard reality of the situation.
In conclusion, we challenge researchers who include measures of White-nonwhite income inequality in aggregate regressions predicting White-nonwhite segregation to (a) specify and substantiate the assumptions that must be met for this method to yield correct estimates of the impact of income inequality on segregation and (b) provide a basis for setting aside the strong, theory-based presumption that White-nonwhite disparities across multiple domains of attainment will be spuriously correlated because they rise and fall together depending on the intensity of racial stratification dynamics in different communities. We do believe community-level regressions can provide useful insights, but we view the results as revealing community-level correlates of segregation, which is a preliminary, not definitive, step toward establishing the determinants of segregation. In Chap. 6 we use models of locational attainments to frame segregation as a form of inequality and demonstrate how segregation is driven by micro-level processes. In that chapter, we argue for a more methodologically appropriate modeling approach for understanding segregation as a product of micro-level factors.
3.8 Consequences of Index Choice for Understanding Trends in Segregation
In Chap. 2 we explained in detail the considerations that must be made when deciding which segregation measure to use for specifically analyzing the dimension of segregation known as evenness. The most widely used measure is the dissimilarity index, or D, which was first popularized by Duncan and Duncan (1955) many decades ago and continues to be the dominant choice in the literature on residential segregation. However, as we discussed in the previous chapter, methodological studies (e.g., Winship, 1977) have established that the dissimilarity index is especially susceptible to the problem of distortion by intrinsic upward bias in index scores and the problems can be alarming under certain conditions. Most notably, the issue arises when group counts for spatial units are small. This is the case with block-level data needed to study segregation in small communities and when one group in the comparison is disproportionately larger than the other group – a common occurrence in predominately White rural communities, micropolitan areas, and even many metropolitan areas.
Knowing that the conditions that create problems for using the dissimilarity index likely do occur in our comprehensive analysis of segregation across all areas of the United States, we chose in the previous sections to limit our substantive interpretations to the separation index, which is far less susceptible to the same issues that affect the dissimilarity index and more reliably reflects prototypical segregation, or patterns of polarized unevenness where the two groups in the analysis are living in substantively different neighborhoods with little residential contact with one another. In contrast, the dissimilarity index may react to uneven distribution but can register high scores even when the magnitude in the difference between the amount of residential contact that each group has with the reference group is small, i.e., dispersed unevenness. Over time, this may affect how we observe and interpret changing patterns of segregation within a community. In this final, brief methodological section of the chapter, we consider the separation index alongside the dissimilarity index to empirically demonstrate where S and D are most likely to deviate from one another over time.
Strictly speaking, one only needs to review the value of S to know whether the pattern of prototypical segregation and polarized unevenness is present. If the value of S is high, it is present; if the value of S is low, it is not. Since this is the aspect of segregation that motivates most concerns about segregation, one could stop at this point. However, when S is low, one must examine the value of D to know whether dispersed unevenness is present. If D is high while S is low, it is present; if D is low, it is not. Knowledge of the presence of dispersed unevenness might be of interest because it can be a precursor to the emergence of polarized unevenness, or it can be a vestige of declining polarized unevenness. The basis for characterizing the combination of high-D, low-S as a precursor to polarized unevenness is grounded in understanding how values of D and S can change in relation to each other when D and S are at intermediate and low levels and uneven distribution increases. When D and S are both low, all aspects of uneven distribution will be low. If uneven distribution increases, values of D and S will take paths at or between two possible extremes as follows:
-
If emerging unevenness is maximally dispersed, values of D will rise and values of S also will rise but by much smaller increments.
-
If emerging unevenness is maximally polarized, values of D and S will rise in equal increments.
-
If emerging unevenness is intermediate on dispersal/polarization, values of D will rise and values of S also will rise but by smaller increments.
If D and S are at intermediate levels with D > S, the value of S can always potentially rise to match the value of D if unevenness shifts from being dispersed to being polarized. If uneven distribution increases while D and S are at intermediate levels with D > S, the value of S will lag behind D if increased unevenness is dispersed or, alternatively, it will move toward D if increased unevenness is polarized. As unevenness progresses from an intermediate level where D > S to its maximum level, unevenness must eventually become fully polarized, so the value of S must eventually rise to match the value of D. From this, it is logically possible that the emergence of uneven distribution might start first with dispersed unevenness (high D, low S) and then continue and progress toward polarized unevenness (high D, high S) and prototypical segregation. In this scenario, the combination of high-D, low-S is a precursor to polarized unevenness.
The basis for characterizing the combination of high-D, low-S as a vestige of declining polarized unevenness is similarly grounded in understanding how values of D and S can change in relation to each other when both D and S are at intermediate and high levels. If both S and D are high, all aspects of uneven distribution will be high. If uneven distribution then declines, values of D and S will take paths at or between two possible extremes as follows:
-
If declining unevenness involves shifting from maximum polarization to intermediate or maximum dispersal, values of S will decline rapidly, and values of D also will decline but by much smaller increments.
-
If declining unevenness remains maximally polarized, values of S and D will decline in equal increments.
-
If declining unevenness leads to an intermediate mix on dispersal/polarization, values of S will decline, and values of D also will decline but by smaller increments.
At any intermediate level of unevenness where values of D and S are concordant, the value of S can decline more rapidly than the value of D if unevenness transitions from being polarized to being dispersed. If unevenness progresses from an intermediate level to its minimum level, the value of D must eventually decline to match the value of S. Thus, it is logically possible that the elimination of uneven distribution might start first with polarized unevenness (high D, high S) and then progress toward dispersed unevenness (high D, low S) before ultimately going to zero on both D and S. In this scenario, the combination of high-D, low-S would be the last vestige of prototypical segregation going away.
Reviewing these scenarios calls attention to the possibility that trends over time can differ by index and it may be interesting to see whether D and S move in unison, or in different sequences. In Table 3.12, we describe initial and changing patterns of unevenness by area type and group comparison using both the separation index and the dissimilarity index as described above to identify the extent to which unevenness is polarized or dispersed. For White-Black segregation, the story is quite simple. In all area types, the typical initial pattern is one of polarized unevenness which is indicative of prototypical segregation. Both indices are initially at medium to high levels and White and Black households are largely living in different neighborhoods. However, we find that over time these levels of polarization are declining, leading towards more dispersed patterns of unevenness where Black households may still have less residential contact with White households than White households do, but their overall residential contact with White households is increasing. Across all community types, White-Asian segregation has initial patterns of dispersed unevenness where the separation index is low in absolute terms and much lower than the dissimilarity index. For the most part, this pattern is holding steady with both indices changing in only negligible amounts.
While the dissimilarity index for White-Latino segregation remains relatively steady over time across all community types, the separation index shows more complicated patterns. In all community types, initial patterns of unevenness are dispersed because the separation index is considerably lower than the dissimilarity index. However, over time these communities are trending in different directions. In noncore counties, the separation index is declining, which indicates that these communities continue to shift towards more dispersed patterns of unevenness. In micropolitan areas, the separation index is holding steady and therefore patterns of dispersed unevenness are also holding steady. Finally, in metropolitan areas, the separation index is rising. This is indicative of patterns of unevenness that are polarizing, leading to higher levels of residential separation between White and Latino households. In contrast to the simplicity of White-Black and White-Asian segregation trends, White-Latino segregation trends demonstrate how using both indices can also provide more nuanced insights into complex patterns of unevenness.
3.9 Summary
This chapter provides a broad overview of racial and ethnic residential segregation trends across the United States from 1990 to 2010. The analysis we conducted is one of the most comprehensive performed to date based on: (a) covering a wide range of group comparisons, (b) covering metropolitan areas, micropolitan areas, and noncore counties, and (c) including many more communities. Additionally, this is the first major analysis of trends and patterns of residential segregation in the U.S. to use segregation indices that are free of the problem of index bias that has troubled researchers in the past and has forced undesirable restrictions on the segregation comparisons included for analysis. Some of the findings presented in this chapter may not be surprising, nor should they have been, but others are new and important. One less surprising finding is that, when we look at segregation involving comparisons and contexts common to previous studies, such as White-Black segregation in metropolitan areas, we are analyzing cases where index bias is less likely to distort segregation scores and therefore, we replicate previous findings. The reason this finding is not surprising is that, in cases where bias truly is negligible, scores for the unbiased versions of segregation indices we use in our study will closely replicate the scores of standard versions of segregation indices used in previous studies. The problem, of course, is that index bias is far from negligible for most cases in our analysis. Thus, our study reports the new and important finding that results for the unbiased versions of segregation indices are very different from results from previously reported standard versions of segregation indices because the availability of valid and reliable unbiased versions of measures of uneven distribution allows our study to perform analyses using a more comprehensive and more representative analysis sample.
The reason for the differences in results is simple. For most of the cases in our analysis, the impact of index bias on standard scores is not negligible; to the contrary, it is typical for bias to inflate standard index scores by large amounts. This is true when we assess segregation using data for persons, as is typical in most prior research. And the impact of bias takes on even greater importance when we assess segregation using data for households to eliminate bias that results from persons locating with and having contact with same-race members of their households. Accordingly, eliminating the impact of bias on index scores and measuring segregation of households rather than persons leads our results to differ from findings reported in any earlier studies that adopted case selection criteria that allowed analysis datasets to be larger and more comprehensive. One major difference, of course is that we find significantly lower levels of segregation. Partly this is because we focus attention on scores for the separation index (which we discuss next,) but our results for D also are much lower than scores reported in previous studies, especially for comparisons where groups are imbalanced in size. The main reason for this is that the cases we are able to include using new methods tend to have high standard scores but low unbiased scores because they are especially affected by bias. Previous studies have no effective method for working with these cases.
While previous studies underestimate the magnitude of the problem of index bias, they acknowledge it is a serious problem they must deal with. The main method they use is to minimize the impact of worrisome cases by discounting them (through differential weighting) or excluding them outright. These methods not only do not solve the problem (scores for these cases remain distorted and have undesirable effects on results), they also make the analysis less representative. Our methods allow us to include these numerous cases and obtain a more comprehensive and representative analysis dataset. Having these cases in the analysis is crucial because they are highly relevant for understanding the level and form segregation takes when groups are small in size and how the level of segregation may (or may not) change as groups grow in absolute and relative size. The method of differential weighting simply discounts the distorted scores to minimize their impact. Thus, to the extent the cases are actually allowed to influence results, their inclusion drives average scores up when using standard index scores.
In addition to using unbiased indices, our results also differ from past research because we give more attention to the separation index rather than the dissimilarity index. The reason, as we explained throughout, is that S provides a more accurate measure of the aspect of uneven distribution that motivates most segregation research – namely, identifying communities where groups occupy different neighborhoods and are at risk of inequality on location-based outcomes, which we refer to as polarized unevenness. We note that D cannot identify these communities, as high values of D will often identify communities where this pattern is absent and instead there is a pattern of dispersed unevenness. We also note it may be interesting to more closely compare D and S to gain a more nuanced understanding of certain kinds of patterns of uneven distribution. But that is not our main focus in this chapter. Our findings for S are important for showing that group separation is lower than previous research would suggest, especially for White-Latino segregation and White-Asian segregation.
Our study raises a question about how we should characterize variations in segregation across communities documented in our study in comparison with findings reported in previous research. The central issue is the analysis dataset we use in our study is larger and more representative than the analysis datasets used in previous studies and this has a nontrivial impact on findings about variations in segregation across communities. All else equal, the additional communities we are able to bring into the analysis tend to have lower levels of segregation, so their inclusion shifts the distribution of index scores to lower values for measures of central tendency and also lower values for percentile locations such as quartiles and deciles. The resulting changes in descriptive statistics represent technical improvements on previous research. But some may find the changes jarring because they depart from previous findings that are more familiar.
What consequences flow from documenting segregation in a broader, more representative set of communities? One key outcome is that the distribution of index scores shifts toward lower values. And the next question we must ask is how we should think about findings from previous studies. First, using standard scores in nonmetropolitan settings is no longer defensible. Index scores computed using smaller spatial units appropriate for measuring segregation in nonmetropolitan settings are always significantly inflated by index bias and the problem is severe for areas where group size is imbalanced and/or one group is small in absolute size. The patterns are stark, and they cannot be overcome. Excluding cases offers poor protection from the distortions of index bias. Many of the cases that are excluded are absolutely of legitimate sociological and demographic interest. Therefore, the loss in coverage and representativeness skews results and distorts findings. Furthermore, the non-excluded cases are not free from bias. When we apply conventional sample restriction constraints in a sequence of increasingly conservative steps, the problem of scores being significantly distorted by index bias never disappears even as the analysis sample becomes increasingly non-representative.
The index scores obtained using unbiased versions of index calculation formulas provide the clearly superior solution. Bootstrap simulation analysis establishes that the unbiased scores perform exactly as desired. In particular, they take average values of zero across every subset or grouping of cases in the study and thus, in dramatic and superior contrast to standard index scores, unbiased index scores have no intrinsic associations with any characteristics of communities. And, while new and not yet familiar to many researchers, the unbiased index scores have simple, intuitive interpretations as group differences in average contact with White households among neighbors that can be easily explained to broad audiences as well as to seasoned researchers. Furthermore, focusing on the separation index rather than the dissimilarity index ensures that researchers can accurately identify cases where polarized unevenness is occurring, the pattern of uneven distribution most consequential for creating the conditions of unequal outcomes.
Our intention for this chapter is that it will provide an exemplar for what is possible with new methods and establish benchmarks for evaluating segregation patterns in the future. Following this chapter, in which we also empirically explored some of the measurement issues that can be overcome using our measurement approaches, we begin to focus in on specific contexts of segregation, including nonmetropolitan communities and Latino and Asian new destinations, that have been understudied due to the limitations of conventional segregation measurement. In addition to considering nonmetropolitan contexts that are often left out of the literature, understanding the complexities of racial segregation will require considering the role of micro-level, individual-based characteristics such as immigration and acculturation as well as socioeconomic diversity, which we do in Chap. 6. With these new methods of measurement and analysis at our disposal, we can proceed to advance our understanding of the dynamics and patterns of racial and ethnic residential segregation across the United States.
Notes
- 1.
Technically, valid unbiased scores can be calculated using person data if one has access to detailed tabulations of persons by race and size of household for the relevant spatial units. The calculations are not feasible for the study period considered here because relevant tables are not available in all years. We conducted methodological studies (not reported here) that confirmed, we would say not surprisingly, that unbiased index scores computed using data for households correlated near-perfectly with unbiased index scores computed using relevant detailed data for persons.
- 2.
This is not an issue if tabulations for persons exclude persons residing in institutions, group quarters, and other settings that are not relevant to measuring residential segregation. But this is not the case in all decades. In that situation, cases must be reviewed, flagged, and excluded if index scores are potentially subject to distortion by the presence of these subpopulations.
- 3.
The theory here has both community-level and micro-level components. The community-level component is that military policies directly and indirectly exert influence on the local housing dynamics and systematically promote equal-status contact between groups across a wide range of social domains. The micro-level component is that serving in the military leads to changes in the behavior of the individuals who served. Taking our own advice, we must acknowledge that the aggregate-level regression is not ideal for evaluating the micro-level component.
References
Alba, R. D., & Logan, J. R. (1993). Minority proximity to whites in suburbs: An individual-level analysis of segregation. American Journal of Sociology, 98, 1388–1427.
Allen, J. P., & Turner, E. (2012). Black-white and Hispanic-white segregation in U.S. counties. The Professional Geographer, 64(4), 503–520.
Charles, C. Z. (2000). Neighborhood racial-composition preferences: Evidence from a multiethnic Metropolis. Social Problems, 47(3), 379–407.
Charles, C. Z. (2003). The dynamics of racial residential segregation. Annual Review of Sociology, 29, 167–207.
Chetty, R., Hendren, N., Jones, M. R., & Porter, S. R. (2020). Race and economic opportunity in the United States: An intergenerational perspective. The Quarterly Journal of Economics, 135(3), 1567–1633.
Crowell, A. R., & Fossett, M. (2018). White and Latino locational attainments: Assessing the role of race and resources in U.S. metropolitan residential segregation. Sociology of Race and Ethnicity, 4(4), 491–507.
Crowell, A. R., & Fossett, M. (2020). The unique case of Minneapolis-St. Paul, MN: Locational attainments and segregation in the twin cities. Spatial Demography, 8(1), 1–31.
Crowell, A. R., & Fossett, M. (2022). Metropolitan racial residential segregation in the United States: A microlevel and cross-context analysis of black, Latino, and Asian segregation. Demographic Research, 46, 217–260.
Duncan, O. D., & Duncan, B. (1955). A methodological analysis of segregation indices. American Sociological Review, 20, 210–217.
Firebaugh, G., & Farrell, C. R. (2016). Still large, but narrowing: The sizable decline in racial neighborhood inequality in metropolitan America, 1980–2010. Demography, 53(1), 139–164.
Flippen, C., & Kim, E. (2015). Immigrant context and opportunity: New destinations and socioeconomic attainment among Asians in the United States. The Annals of the American Academy of Political and Social Science., 660, 175–198.
Fossett, M. (1988). Community-level analyses of racial socioeconomic inequality: A cautionary note. Sociological Methods and Research, 16, 454–491.
Fossett, M. (2017). New methods for measuring and analyzing segregation. Springer Publications.
Fossett, M., & Crowell, A. R. (2018). Assessing the impact of group income differences on cross-city variation in segregation: Do aggregate regressions provide valid estimates? Poster presented at the Annual Meetings of the Population Association of America, Denver.
Frey, W. (2018). Diversity explosion: How new racial demographics are remaking America. The Brookings Institution.
Hoeffel, E., Rastogi, S., Kim, M. O., & Shahid, H. (2012). The Asian Population: 2010. 2010 Census Briefs.
Iceland, J. (2014). A portrait of America: The demographic perspective. University of California Press.
Iceland, J., Weinberg, D., & Steinmetz, E. (2002). Racial and ethnic residential segregation in the United States: 1980–2002. U.S. Government Printing Office.
Iceland, J., Weinberg, D., & Hughes, L. (2014). The residential segregation of detailed Hispanic and Asian groups in the United States: 1980-2010. Demographic Research, 31, 593–624.
Krysan, M., & Crowder, K. (2017). Cycle of segregation: Social processes and residential stratification. Russell Sage Foundation.
Lichter, D. T., Parisi, D., Grice, S. M., & Taquino, M. C. (2007). National Estimates of racial segregation in rural and small-town America. Demography, 44(3), 563–581.
Lichter, D. T., Parisi, D., Taquino, M., & Grice, S. M. (2010). Residential segregation in new Hispanic destinations: Cities, suburbs, and rural communities compared. Social Science Research, 39, 215–230.
Logan, J. R., & Stults, B. J. (2011). The persistence of segregation in the metropolis: New findings from the 2010 census. Census Brief prepared for Project US2010.
Massey, D. (2020). Still the linchpin: Segregation and stratification in the USA. Race and Social Problems, 12, 1–12.
Massey, D., & Denton, N. (1987). Trends in the residential segregation of blacks, Hispanics, and Asians: 1970–1980. American Sociological Review, 52(6), 802–825.
Massey, D., & Denton, N. (1988). The dimensions of residential segregation. Social Forces, 67(2), 281–315.
Massey, D., & Denton, N. (1989). Hypersegregation in U.S. metropolitan areas: Black and Hispanic segregation along five dimensions. Demography, 26(3), 373–391.
Massey, D., & Denton, N. (1993). American apartheid: Segregation and the making of the underclass. Harvard University Press.
Massey, D., & Fong, E. (1990). Segregation and neighborhood quality: Blacks, Hispanics, and Asians in the San Francisco metropolitan area. Social Forces, 69(1), 15–32.
Massey, D., & Tannen, J. (2015). A research note on trends in black Hypersegregation. Demography, 52(3), 1025–1034.
Saenz, R. (2004). Latinos and the changing face of America. In R. Farley & J. Haaga (Eds.), The American people: Census 2000. Russell Sage Foundation.
Sakamoto, A., Goyette, K. A., & Kim, C. H. (2009). Socioeconomic attainments of Asian Americans. Annual Review of Sociology, 35, 255–276.
Sharkey, P. (2013). Stuck in place: Urban neighborhoods and the end of Progress toward racial equality. University of Chicago Press.
Stearns, L. B., & Logan, J. R. (1986). Measuring segregation: Three dimensions, three measures. Urban Affairs Quarterly, 22, 124–150.
Vásquez, M. A., Seales, C. E., & Marquardt, M. F. (2008). New Latino destinations. In H. Rodríguez, R. Sáenz, & C. Menjívar (Eds.), Latinas/os in the United States: Changing the face of América (pp. 19–35). Springer.
Wilkes, R., & Iceland, J. (2004). Hypersegregation in the twenty-first century. Demography, 41(1), 23–36.
Winship, C. (1977). A revaluation of indexes of residential segregation. Social Forces, 55(4), 1058–1066.
Zhou, M., & Logan, J. R. (1991). In and out of Chinatown: Residential mobility and segregation of New York City’s Chinese. Social Forces, 70(2), 387–407.
Author information
Authors and Affiliations
Rights and permissions
Open Access This chapter is licensed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license and indicate if changes were made.
The images or other third party material in this chapter are included in the chapter's Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the chapter's Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder.
Copyright information
© 2023 The Author(s)
About this chapter
Cite this chapter
Crowell, A.R., Fossett, M.A. (2023). National Trends in Racial and Ethnic Residential Segregation. In: Racial and Ethnic Residential Segregation Across the United States. The Springer Series on Demographic Methods and Population Analysis, vol 54. Springer, Cham. https://doi.org/10.1007/978-3-031-38371-7_3
Download citation
DOI: https://doi.org/10.1007/978-3-031-38371-7_3
Published:
Publisher Name: Springer, Cham
Print ISBN: 978-3-031-38369-4
Online ISBN: 978-3-031-38371-7
eBook Packages: Social SciencesSocial Sciences (R0)