If Residential Segregation Persists, What Explains Widespread Increases in Residential Diversity?

Abstract

Recent studies have identified increasing residential diversity as a near-universal trend across the United States. At the same time, a wide range of scholarship notes the persistence of White flight and other mechanisms that reproduce residential segregation. In this article, we attempt to reconcile these findings by arguing that current trends toward increased residential diversity may sometimes mask population changes that are more consistent with racial turnover and eventual resegregation. Specifically, we show that increases in diversity occur nearly identically across neighborhoods where White populations remain stable or decline in the face of non-White population growth. Our findings demonstrate that, particularly in its early stages, racial turnover decouples diversity and integration, leading to increases in diversity without corresponding increases in residential integration. These results suggest that in many neighborhoods, diversity increases may be transitory phenomena driven primarily by a neighborhood's location in the racial turnover process. In the future, stalled or decreasing levels of diversity may become more common in these areas as segregation persists and the process of racial turnover continues.

Introduction

Never before have American communities exhibited greater levels of residential diversity. Over the last few decades, the number of all-White communities has declined steadily (Fowler et al. 2016; Glaeser and Vigdor 2012; Sharp and Lee 2017) as “global” or “quadrivial” neighborhoods with residents of all major racial/ethnic groups have become more common (Bader and Warkentien 2016; Logan and Zhang 2010; Zhang and Logan 2016).¹ These and other demographic shifts have established increasing residential diversity as a near-universal trend in the United States, affecting the racial/ethnic composition of neighborhoods across a broad range of urban, rural, and suburban areas (Hall et al. 2016; Lee and Hughes 2014; Lee et al. 2014). On its face, this trend seems encouraging. If increases in residential diversity lead predictably to the formation of stably integrated multiethnic communities, we should expect to see corresponding declines in residential segregation in the years to come. Some scholars have referred to this possibility as the demographic integration perspective (Farrell and Lee 2011).

Here, we build on prior theory and research to provide an alternative prediction. Namely, we argue that for many neighborhoods, current trends toward increased residential diversity may mask population changes that are more consistent with the reproduction of segregated White and non-White areas. The logic underlying our argument is straightforward. In addition to the growth of minority groups, increases in residential diversity may also result from net losses in the total number of White residents. Whites may become a more “equal” share of a neighborhood's population—increasing residential diversity, at least temporarily—but this occurs because these neighborhoods are in the midst of racial turnover (Bader and Warkentien 2016; Crowder 2000; Crowder and South 2008; Frey and Farley 1996; Hall 2013; Hartmann 1993; Lee 2007). In the extreme, this process of racial turnover could slow current trends toward increasing racial/ethnic residential diversity and integration. We refer to this alternative prediction as the bifurcation model.

In this article, we seek to test these predictions by parsing diversity trends between three neighborhood types: (1) those where White populations persist over time in the context of neighborhood diversity, (2) those where White populations are in the early stages of exit, and (3) those where White populations have entered turnover's final stages. Although the first neighborhood type reflects racial/ethnic stability indicative of integration (demographic integration), the other two are characterized by patterns of racial/ethnic turnover that have historically led to persistent residential segregation (bifurcation). Our results show that White persistence and decline can lead to increasing levels of residential diversity but that their ultimate effects on residential segregation depend critically on the stage of racial turnover. In the early stages of turnover, diversity increases rapidly despite slowing decreases in segregation. These trends continue until the final stages of turnover, when emergent decreases in diversity recouple with stagnant segregation declines. These findings suggest that ongoing patterns of racial turnover can help explain why segregation persists in many areas despite unprecedented increases in ethnoracial diversity.

Background

Much research has focused on the reverberating effects of America's ongoing “diversity explosion” (Frey 2015). Fueled by sustained Hispanic and Asian population increases, residential areas have dramatically increased in ethnoracial diversity in recent decades. Indeed, in their examination of nearly 11,000 communities from 1980 to 2010, Lee and Hughes (2014) found that less than 6% of census-designated places—individual cities, suburbs, towns, boroughs, and villages—peaked in diversity in either 1980 or 1990. By contrast, more than 90% of places exhibited their highest levels of diversity in 2010, suggesting a sharp and consistent upward trend (Lee and Hughes 2014). The proliferation of diversity as a dominant trend represents the outcome of both structural shifts in migration—the greater distribution of immigrants away from a handful of gateway cities—and processes of spatial assimilation that have allowed minority residents greater mobility into middle-class suburban spaces (Massey 1985; Singer 2009). These broader shifts have generated optimism that traditionally durable color lines may be blurring, including for Black populations who have historically experienced the most rigid barriers to integration (Massey and Denton 1988). As Frey (2015:189) noted:

This shift has been accompanied by several encouraging demographic trends. Research shows that the number of all-White or predominantly White neighborhoods has severely contracted since 1980 (Farrell and Lee 2011; Glaeser and Vigdor 2012; Lee et al. 2014). Likewise, multiethnic neighborhoods with diverse racial structures have proliferated (Denton and Massey 1991; Farrell and Lee 2011; Holloway et al. 2012; Logan and Zhang 2010). What allows these neighborhoods to avoid turnover and endure as sites that would, in the aggregate, promote integration? According to the buffering hypothesis, immigration has diminished long-standing color lines between White and Black residents (Farley and Frey 1994; Frey and Farley 1996). In an era of unprecedented diversity, the social and spatial buffer that immigrants provide considerably reduces the odds of traditional racial residential turnover (Logan and Zhang 2010; Parisi et al. 2015; Zhang and Logan 2016). In sum, the sheer magnitude of ethnoracial diversity—across the urban–rural continuum—suggests that stable integration has emerged as a more viable outcome for U.S. neighborhoods in the twenty-first century.

However, the prospect of stable residential integration runs counter to research findings on racial residential segregation (Krysan and Crowder 2017; Logan 2013; Massey and Tannen 2015). Nationwide, Black–White segregation has declined modestly but remains effectively unchanged in metropolitan areas with the largest Black populations (Krysan and Crowder 2017). As recently as 2010, less than 1% of Black residents lived in metropolitan areas with low levels of segregation (Massey and Tannen 2015). Meanwhile, Hispanic and Asian segregation levels have increased since 1980 (Charles 2003; Logan et al. 2004), and White flight has continued unabated “from neighborhoods where Hispanics and Asians are available as a social buffer between Whites and Blacks” (Logan and Zhang 2010:1103).² Several studies have found greater segregation for immigrants in new destinations (Hall 2013; Lichter et al. 2010), which have also become more segregated over time (Park and Iceland 2011). Taken as a whole, this body of work provides little evidence that greater diversity leads in lockstep to declines in residential segregation or that Whites' preferences for neighborhoods with a large Black or immigrant presence have radically changed (Crowder et al. 2011; Kye 2018; Parisi et al. 2019).

Like other researchers, we suspect that the inconsistent relationship between racial/ethnic diversity and residential segregation stems partly from basic technical distinctions between the two concepts and the indices used to measure them (Fowler et al. 2016; Iceland 2004; Lee and Hughes 2014; Sin and Krysan 2015). Diversity measures, such as the entropy index (E), focus on a single areal unit. For any given neighborhood, place, or metropolitan area, diversity increases as groups come to mirror absolute standards of equal group representation (e.g., if there are four groups, equal representation would occur when they each account for 25% of the population). By contrast, measures of residential segregation, such as the index of dissimilarity (D) or Theil's information theory index (H), summarize the group distributions across subunits (e.g., neighborhoods) that compose a broader geographic space (e.g., metropolitan areas). According to D or H, desegregation occurs as more neighborhoods approach the demographic composition of the metropolitan areas in which they are embedded (Massey and Denton 1988; Reardon and Firebaugh 2002). Thus, even if a subset of neighborhoods shows significant increases in diversity (as measured by E), there is no guarantee that segregation (as measured by D or H) will decline similarly. Greater homogeneity in other neighborhoods could prop up segregation for the metropolitan area as a whole, as others have routinely observed (Friedman 2008; Holloway et al. 2012; Iceland 2004).

It seems unlikely, however, that these conceptual and technical distinctions explain the substantial gulf between diversity and segregation trends. If residential segregation shows minimal decline, as research has shown (see, e.g., Krysan and Crowder 2017), then what accounts for widespread increases in residential diversity? Or conversely, what explains the continued growth of diverse neighborhoods in a landscape characterized by persistent or growing segregation? Despite what appears to be growing slippage between measures of diversity and segregation, we know relatively little about the mechanisms responsible for their incongruence. Our focus in this article is on White population losses, or White flight, as one such mechanism. We argue that the relationship between diversity and segregation can take two ideal-typical forms: (1) increasing diversity leads to a breakdown of residential segregation (through the processes already described), and (2) increasing diversity is an intermediate stage in a process that culminates in resegregation. In the first case, neighborhoods that diversify in the absence of White flight provide an opportunity for long-term multiethnic neighborhood formation. On balance, areas where these neighborhoods are more prevalent are more likely to move toward stable integration. In the second case, the presence of White flight is a part of a process of racial turnover that leads to fleeting increases in diversity. In areas where more neighborhoods diversify because of turnover, states of persistent or increased segregation are the more likely outcome.

The building blocks of our argument have been suggested by existing research. Several studies question the long-term stability of multiethnic neighborhoods (e.g., Friedman 2008; Kye and Halpern-Manners 2022). Others have identified evidence of White flight (Kye 2018), avoidance (Parisi et al. 2019), and institutional discrimination (Korver-Glenn 2021) that would appear strongly suggestive of enduring segregation despite national trends in residential diversity (Holloway et al. 2012; Wright et al. 2020). Our main contribution lies in empirically connecting these various strands of research by explicitly linking micro-level changes in diversity (i.e., within neighborhoods) to macro-level changes in residential segregation (i.e., within broader geographic spaces). We focus, in particular, on the role of White population change as a central mechanism that leads increases in diversity and integration to overlap (demographic integration) or decouple (bifurcation).³ Our central argument is that White population losses can trigger widespread but temporary increases in diversity that, in the aggregate, preserve or exacerbate segregation.

We evaluate the merits of this argument through a series of interrelated analyses. We begin by establishing that increases in diversity are not the sole result of stable multiethnic neighborhood formation. In particular, we show that although most commonly associated with patterns of non-White entry, diverse neighborhoods also emerge through processes of White depopulation. In the second part of our analysis, we examine how these different pathways to diversity affect residential segregation among neighborhoods nested within a given area. Our findings suggest that trends in diversity can be unreliable predictors of segregation, especially in places where neighborhoods—via White population exit—are approaching the final stages of turnover. This raises reasonable doubts about whether increases in diversity will lead to lasting and widespread integration in the contemporary United States. If anything, our results suggest that decreases in diversity and further movement away from residential integration may loom as an emerging trend in many areas going forward.

Data

Data for our analyses come from the 1990 to 2010 U.S. decennial censuses and the American Community Survey (ACS).⁴ Our analytic strategy requires two levels of analysis: (1) a macro geography to measure changes in segregation; and (2) a micro geography nested within these units, whose population changes (and diversity changes) might affect segregation outcomes measured for the broader geographic space. Previous studies, for example, have examined how metropolitan area segregation is affected by population changes across the census tracts that compose them (e.g., Hall 2013; Iceland 2004). We follow a similar approach here by examining census places composed of census blocks.

At the macro level, we use census places as proxies for local communities (e.g., Faber 2020; Fowler et al. 2016; Hall et al. 2016). Census places represent individual cities, suburbs, towns, boroughs, and villages across the metropolitan United States. Although places may vary in size, most retain some level of independent government authority or service because of their function as political boundaries. As a result, they are a useful proxy for local labor and housing markets (Hall et al. 2016). We begin by sampling all places that had at least 1,000 residents in 1990 and were classified as places throughout the study period (i.e., 1990–2010). From this initial universe of places, we then sampled all places from any metropolitan area where the share of non-Whites in 2010 was greater than the 2010 nationwide average of 28.6% (N = 150 metropolitan areas; the full list of sample areas can be found in the online appendix).⁵ These criteria reflect our interest in examining places where diversity has grown enough that mechanisms of buffering or racial turnover are potential outcomes.

By invoking a threshold pertaining to non-White groups more broadly (i.e., Blacks, Hispanics, and Asians, collectively), our metropolitan sample ranges from areas colloquially considered “multiethnic” (e.g., Houston, Texas) or “immigrant” (e.g., Riverside, California) to those considered “White–Black” (e.g., Memphis, Tennessee). This approach provides greater protection from the omission of important edge cases that might otherwise be excluded because a specific racial/ethnic group falls just under a designated threshold. The sample primarily excludes places in predominantly White metropolitan areas across the Great Plains and less densely populated portions of the Midwest because even though prior research suggests that diversity in these places has increased, minority population growth remains a relatively new phenomenon. The sample also excludes places in nonmetropolitan and micropolitan areas for this same reason, as well as the lower reliability of the ACS's block-sampling rates in nonmetropolitan areas. Our final sample includes 4,434 places, capturing 61% of U.S. Blacks, 75% of Hispanics, 74% of Asians, and 36% of all non-Hispanic Whites in 2010.⁶Table 1 provides a descriptive overview of these places, including sociodemographic characteristics and information on racial segregation and diversity observed in 1990 and 2010.

At the micro level, we use census blocks as proxies for the neighborhoods composing local communities. We use census blocks rather than the more commonly used census tracts for several reasons.⁷ Conceptually, our interest in diverse neighborhoods is motivated by the rich body of work on intergroup contact, which has consistently demonstrated that contact between members of different racial/ethnic groups can reduce prejudice and build trust under the proper conditions (Allport 1954; Dixon 2006; Pettigrew et al. 2011). For our purposes, capturing this dynamic means identifying a neighborhood level of geography where entropy can reasonably be expected to capture intergroup interactions. All else being equal, we view smaller, diverse units as more reliable proxies for such contact than larger units (e.g., census tracts), where pockets of homogeneity may preclude meaningful interracial exposure (Allen and Turner 1995). Although census blocks are not precluded from being affected by similar issues of clustering, their smaller scale provides tighter geographic constraints on the degree to which intergroup contact can be avoided (e.g., in schools, parks, grocers, and other shared public spaces).

As the basis for all census geographies, blocks also have the methodological advantage of nesting cleanly within places, which cannot be said of census tracts or block groups. Like tracts, blocks covered the entire nation as of the 1990 census, providing full coverage of the places in our analysis. Census blocks in 1990 and 2000 were standardized to their 2010 boundaries using crosswalks provided by the National Historical Geographic Information System (Manson et al. 2022). This step ensures that trends in segregation reflect changes in the unequal distribution of groups across census blocks rather than technical changes in block boundaries.⁸ We drop all census blocks that were unpopulated in both 1990 and 2010 because unlike census tracts, census blocks are not delineated based on population and can contain no residents. We also drop census blocks not located in a census-designated place, consolidated city, or incorporated place because they could not be linked to broader units necessary for assessing trends in segregation. These criteria result in a final sample of 1.7 million blocks nested within 4,434 census-designated places.

Analytical Strategy and Measurement

Our analytical approach begins at the block level, where we distinguish three types of neighborhood change: (1) White population persistence, (2) the early turnover stages, and (3) the later turnover stages. When our analyses shift to the place level, we identify places where blocks disproportionately exhibit persistence versus various stages of turnover (we detail these place-level criteria in greater detail later). At several points in our analyses, we present descriptive statistics that more fully characterize the changes each block or place type experienced. These figures serve as checks to ensure that our operationalizations of the preceding concepts appropriately capture blocks and places located in (or are disproportionately affected by) distinct stages of racial turnover.

Our main analyses use these block and place types to examine how racial turnover (or its absence) can affect diversity at a micro level and how racial turnover can aggregate to affect segregation at a macro level. In the rest of this section, we provide details on our (1) classification of different stages of racial turnover and (2) measurement of residential diversity and segregation.

Identifying White Population Persistence, Decline, and the Stages of Racial Turnover

We hypothesize that the nature of White population change can mediate the relationship between diversity and segregation. A detailed consideration of this argument requires that we define neighborhood types experiencing theoretically meaningful differences in both the direction (i.e., whether Whites exit) and magnitude (i.e., the extent to which Whites have exited) of population change. To achieve these aims, we adopt a two-step classification approach that eventually leads to three block types of interest. We begin by defining two broad neighborhood types: those experiencing White population decline and those experiencing White population persistence.⁹ Distinguishing between these neighborhood types rests on our ability to define and reasonably measure significant White population loss over time. To accomplish this objective, we employ Kye and Halpern-Manners' (2022) multicomponent approach for detecting patterns of White flight. The key benefit of this approach is the recognition that White population loss is a multidimensional phenomenon (see the online appendix for full methodological details). To qualify as experiencing White population decline, a census block must exhibit significant losses in both absolute (i.e., the size of the White population must decline) and relative (i.e., Whites' share of the population must decline) terms from 1990 to 2010 (see the online appendix for details on the thresholds used to define significant losses). In addition, the declines must be sufficiently large after standardizing for differences in White population size (i.e., what is considered a meaningful loss will vary according to the initial size of the White population). Census blocks that satisfy each component of this measure are classified as having experienced White population decline; all other blocks are considered to have experienced White population persistence.¹⁰

In the second step of classification, we further stratify the subsample of blocks with White population decline to distinguish between those in the early or later stages of racial turnover. Conceptually, White decline blocks are differentiated by the share of Whites remaining in 2010 (T₂) after experiencing White population losses from 1990 to 2010 (T₁ – T₂). Given that the overwhelming number of blocks began the transition period with a sizable majority-White population (mean value of 77.4% White), we consider White decline blocks to be in early turnover stages if they retained a White population share greater than 50% by 2010. By contrast, blocks are considered to be in later turnover stages if the White share fell below 50% by 2010. We choose this threshold because of the historical significance of majority-White neighborhoods, but conclusions were unchanged when we used alternative thresholds that ranged from 35% to 65%.

In contrast to White decline blocks, our approach does not further stratify blocks in the broader category of White population persistence. As a result, White populations in this block type may exhibit stability or growth. We collapse these neighborhood changes into a single persistence category because White population stability and growth both avoid progressing blocks (further) through the stages of racial residential turnover. This approach may overlook instances in which the absence of White losses does not imply integration, particularly in cases of gentrification. In additional analyses not shown here, we created a fuller typology of neighborhood transitions to examine the scope of such instances. We found that blocks fitting the general profile of gentrification accounted for less than 3% of persistence blocks (full tabulations available upon request), a figure consistent with previous research (Bader and Warkentien 2016). Given the relative paucity of these block types in our sample, we proceed without further stratifying White population persistence.

Our two-step approach leads to three blocks of interest: those where sizable White populations persist (N = 1,425,010) and those where racial turnover is in the early turnover stages (N = 73,752) or later turnover stages (N = 243,450). In our analysis, we use these block types as a device to demonstrate how divergent population processes can lead to the same trend: increasing racial/ethnic diversity at the census block level. We then explicitly examine the relationship between ethnoracial diversity and residential segregation.

Measuring Residential Diversity and Segregation

where p_r indicates the proportion of each racial/ethnic group at the block or place level, and R indicates the total number of groups. We measure diversity with respect to the four major racial/ethnic groups in the United States: non-Hispanic Whites, Blacks, Asians, and Hispanics of all racial identifications. With four groups, maximum diversity is achieved with an entropy score of log 4, or 1.386. To facilitate interpretation, we standardize E by dividing by the maximum and multiplying by 100. As a result, entropy scores of 100 indicate the highest possible diversity level, with each of the four racial/ethnic groups representing a 25% share of the population. Minimum diversity is achieved with an entropy score of 0, which would occur in neighborhoods where all residents belong to one racial/ethnic group.

where b_i refers to the total population of block i; B is the place population; n is the number of blocks; and E_i and E represent block i's entropy and place diversity, respectively. As with E, we multiply H by 100 so that the final measure ranges from 0 to 100; 0 indicates that all blocks have the same percentage of Whites, Blacks, Hispanics, and Asians as the entire place (maximum integration), and 100 indicates that each block contains only one group (maximum segregation).¹¹

Results

Entropy, White Decline, and White Persistence: Block-Level Summary Trends

In this section, we demonstrate that increases in residential diversity can result from stable multiethnic neighborhood formation or White population losses suggestive of racial residential turnover. Figure 1 summarizes entropy changes for all census blocks in our sample. The panels indicate how entropy changed from 1990 to 2010 for neighborhoods defined as experiencing White persistence or decline, with the latter distinguished by stage of racial turnover.

The graphs—corresponding to our definition of each neighborhood type—indicate three distinct patterns of non-White change. In persistence blocks, non-White populations grew modestly, and they did so at rates comparable to those for White residents. For blocks in the early stages of neighborhood turnover, non-Whites exhibited lower levels of absolute growth but showed the greatest increase in percentage terms, nearly tripling from 1990 to 2010. Finally, blocks in later stages of turnover experienced the most accelerated non-White increases, adding to what was already a sizable population presence in 1990. These nuances aside, the general pattern of non-White increase is consistent across all neighborhood types. By contrast, White persistence and decline blocks showed qualitatively different patterns of White population change. In White persistence blocks, the number of Whites increased by approximately 65 residents, on average. In both early and later turnover blocks, the number of Whites decreased by approximately 70 and 100, respectively.

Importantly, the distinction between growing and shrinking White populations is hardly reflected in entropy scores. Because entropy measures provide a point-in-time snapshot of neighborhood diversity and do not consider the direction or nature of ongoing population change, they cannot distinguish between compositional changes stemming from racial turnover and trends more indicative of stable coresidence. As a result, increases in diversity can arise in blocks where White populations persist or decline in the face of non-White population growth. In fact, increases in diversity tend to be most pronounced among early turnover blocks where both White and non-White groups—through divergent processes of neighborhood exit and entry—converged more rapidly toward an equal share of the overall population. If these trends continue, transitory increases will eventually lead to decreasing diversity as minority groups grow to a predominant share of the population. Indeed, this is precisely the pattern found among later turnover blocks over this period, with a modal trend of decreasing diversity from 2000 to 2010. By contrast, diversity in persistence blocks increased more gradually because of a pattern of White increase coinciding with even greater levels of non-White population growth. These results demonstrate that in the context of ongoing White flight, diversity increases can be poor predictors of stable White and non-White coresidence. That is, greater diversity may not necessarily translate to greater integration. On the contrary, these results suggest that even surging diversity may be an initial symptom of persistent residential segregation.

Entropy and Residential Segregation: Place-Level Trends

We now shift our unit of analysis to places—the macro level at which our multigroup measure of segregation is measured. The block-level results demonstrate that diversity can surge at a micro level despite population changes that lead to blocks with a homogeneous (and majority–minority) racial composition. This finding suggests that racial turnover could help to explain the tenuous connection between diversity and segregation trends. If diversity gains are driven by White population loss in a given area, the expected relationship between diversity and segregation could break down. In this section, we formally test this hypothesis. Specifically, we examine whether trends toward increased diversity and decreased segregation decouple in places with higher racial turnover. We would like to know whether racial turnover triggers increases in diversity without generating corresponding decreases in segregation.

Answering this question requires that we (1) identify places that demonstrated meaningful differences in racial turnover and then (2) examine the relationship between diversity and segregation within them. To do so, we focus on the distribution of early and later turnover blocks within census places, as shown in Figure 2. If the percentage of early (later) turnover blocks within a place was higher than the mean percentage across all places in our sample, we classified the place as experiencing “more early (later) turnover.” If the percentage was less than the mean, we classified the place as experiencing “less early (later) turnover.”¹² This strategy produces a simple 2 × 2 matrix, with places grouped according to the amount of early turnover (more or less) and later turnover (more or less) from 1990 to 2010. Each cell in the matrix maps onto a distinct place type: (1) those experiencing White population persistence (N = 2,133; i.e., places with less early turnover and less later turnover), with minimal evidence of White decline; (2) those in the early turnover stages (N = 876), where most White-decline blocks are in the beginning stages of turnover; (3) places experiencing intermediate turnover (N = 783), with a mixture of early and later turnover stages (i.e., places with more early turnover and more later turnover); and (4) those in the later turnover stage (N = 642), where most White-decline blocks fell below a majority-White share by 2010.

Figure 3 summarizes the average composition of block types in each of the four places defined above. By design, each place type tends to have a disproportionate share of blocks in a different stage of the racial turnover process. Persistence places, by definition, have very few blocks where White populations have declined. In contrast, early turnover, intermediate turnover, and later turnover places demonstrate a further progression of White loss toward (or reaching) later turnover in greater numbers with each place type.

A similar pattern is evident in Figure 4, which plots the average racial/ethnic change in composition for blocks within each of the four place types over the two decades. The average block in a persistence place (the plot on the far left) tended to experience a slight decline in the White population share, falling from roughly 80% in 1990 to 73% in 2010. These declines were successively more pronounced in the average block for early and intermediate turnover places (the middle two plots), where racial turnover led to more dramatic decreases in the White population share. Finally, in later turnover places, White population losses slowed over time as the final stages of White flight waned and non-Whites came to hold a majority share of the population (plot on the far right).

We use these place types as a tool for directly addressing the question at the heart of our study: if residential segregation persists, what explains widespread increases in residential diversity? Earlier, we showed that increasing diversity can result from population processes that look like resegregation (i.e., White population losses). Here, we explicitly consider that possibility by shifting to the place level and directly observing patterns of segregation change. We begin by examining places where Whites tend to persist. These results serve as a baseline for understanding how diversity generally affects segregation in places where racial turnover tends to be absent. We then use our remaining place types to examine precisely how the relationship between diversity and segregation changes as places progress through the turnover cycle.

Figure 5 summarizes changes in segregation (H) and the mean level of block entropy across each place type from 1990 to 2010. In persistence places, the relationship between diversity and segregation tends to be consistent with the demographic integration perspective: as diversity (the red line) increased, places experienced steady decreases in segregation (the black line) over time. The remaining panels suggest that this general relationship holds in all place types but not to the same extent. Although blocks in places with early and intermediate turnover experienced large increases in entropy, these increases did not, on average, translate into comparably large decreases in segregation. In fact, the rate of desegregation was lower in these areas relative to persistence places despite more rapid increases in diversity. By comparison, changes in diversity and segregation show greater correspondence in places where blocks tended to be in the later stages of racial turnover. If these trends were to continue, diversity would plateau before eventually decreasing with further White population decline. Likewise, trends toward decreased segregation appear likely to slow before possibly reversing.

Multivariable Analyses

The preceding results suggest that the diversity–segregation relationship changes as places progress through the various stages of turnover. When White population decline is absent, diversity and segregation are coupled: increases in block-level diversity translate to decreases in place-level segregation. The onset of racial turnover and White population losses stall integration. Diversity increases as Whites relocate, but the relationship between diversity and segregation decouples. This pattern continues through the height of turnover; intermediate turnover places with a mixture of blocks in the early and later turnover stages experience the greatest increases in entropy without comparable decreases in segregation. Finally, when later turnover becomes more prevalent in the blocks embedded within a census place, recoupling starts, and integration slows. Thus, the relationship between diversity and segregation is curvilinear: changes in diversity and segregation are coupled where persistence or the final stages of turnover tend to be common but are decoupled during the middle stages, at the beginning stages of turnover. Importantly, in these intervening states, segregation and diversity emerge as non-zero-sum entities: residential segregation can remain relatively unchanged even as the level of diversity accelerates. This paradox provides a basis for residential segregation to persist, despite widespread increases in residential diversity.

To evaluate this argument further, we fit multivariable models regressing changes in segregation (H) onto place type, changes in mean block diversity (E), and an interaction between E and place type. The key parameters are the interactions, which allow for inferences about variation in the relationship between E and H across stages of racial turnover. Statistically significant interaction terms indicate differences in the effects of E relative to the baseline effect observed in persistence places (the reference category).¹³ We estimate these effects net of two key time-varying place-level controls—the non-White entropy level and the share of residents aged 65 and older—to test our argument against two alternative explanations. First, although we have argued that diversity decouples from integration because of patterns of White population loss, it could also be the case that trends in diversity primarily reflect the residential preferences of racial/ethnic minority groups (and not Whites). If non-Whites prefer to reside alongside fellow coethnics or other non-Whites, as some have suggested (Brown and Chung 2008; Ihlanfeldt and Sacfidi 2002; Wen et al. 2009), diversity could increase across places without parallel declines in residential segregation within them. Second, deaths among Whites rather than their out-mobility might drive racial turnover (Johnson 2020; Johnson and Lichter 2016). Evidence that the age structure accounts for decoupling would shift the onus of segregation from Whites' racial residential preferences to more general discussions about the challenges of establishing integration with a naturally waning U.S. White population.

We also include a battery of conventional measures to account for several well-known correlates of diversity and segregation, in addition to fixed effects for place (to adjust for time-invariant place characteristics) and year (to adjust for unobserved variables that are constant within year and across places). Because group threat and buffering levels may vary according to places' racial/ethnic and socioeconomic composition (Iceland and Wilkes 2006; Zhang and Logan 2016), we include controls for percentage Black, percentage Hispanic, percentage Asian, percentage foreign-born, and the White-to-non-White median household income (MHI) ratio. Previous research also suggests that turnover patterns may vary in the extent to which the number of places proliferates—for example, because of a complex array of zoning laws or restrictions on multifamily units (Rothwell and Massey 2009). We therefore include a measure of political fragmentation, defined as the total number of places per 100,000 metropolitan residents. Finally, we include several standard ecological measures shown to be associated with segregation in previous research (Logan et al. 2004): total population, percentage employed in the manufacturing sector, percentage employed in government, and the percentage of housing stock vacancies.

Regression results are presented in Table 2. We use mean-centered measures in all models so that the intercept indicates the predicted change in segregation (H) for persistence places when all continuous covariates are held at their means. The general patterns are consistent across model specifications and robust to the inclusion of fixed effects. In persistence places, increases in entropy are associated with decreases in segregation. For example, a 10-point increase in mean block entropy is expected to produce a segregation decline of −1.7 points (= −0.17 × 10) according to our fully specified model. The same is not true for early and intermediate turnover places, where increases in diversity have significantly weaker (i.e., less negative) effects. In these places, the relationship between diversity and segregation is essentially broken to the point that increases in diversity yield no meaningful gains in integration. Although the effect of diversity for later turnover places is also weaker (E × later turnover is 0.07 in the fully specified model), the discrepancy is significantly smaller than in places with early (0.12) and intermediate (0.24) turnover (p < .05 for a Wald test of equal coefficients), indicating a recoupling between diversity and segregation. In sum, these results strongly suggest that even as U.S. neighborhoods become more diverse, racial turnover via White population losses is a key barrier preventing otherwise greater integration.

Diversity and Integration: Decoupled for Whom?

In the last part of our analysis, we consider the implications of our findings within the broader context of White and non-White migration. A lingering question is whether non-White population growth is primarily occurring in persistence places or places experiencing racial/ethnic turnover. Answering this question provides a sense of the ramifications of decoupling. A finding of recent waves of non-White population growth concentrated in persistence places would suggest that communities of growing diversity and integration are increasingly within reach for mobile non-White households (e.g., among middle-class racial/ethnic minorities). Likewise, a finding that non-Whites are becoming less concentrated in places experiencing turnover could mean that cycles of White flight will disrupt integration for a waning share of non-Whites over the next several decades.

Figure 6 summarizes the changing total share of Whites and non-Whites in our sample residing in each place type from 1990 to 2010. The results in the middle panels indicate that over the past two decades, non-White populations shifted primarily to places experiencing racial/ethnic turnover (early and intermediate turnover)—areas where diversity tends to decouple from integration.¹⁴ In other words, greater diversity retains a tenuous link to integration outcomes in precisely those places where minority households have increasingly settled. By contrast, White populations increased nearly exclusively in persistence places. Ironically, this finding indicates that the link between diversity and segregation is inequitably distributed. Despite being the driving force behind the U.S. diversity explosion, minority households appear less likely than Whites to reside in places where neighborhood diversity leads to stable integration.

Discussion and Conclusion

In this article, we have argued that prevailing trends in ethnoracial diversity may conflate divergent population processes. The most salient takeaway from our results is that increasing diversity should not be viewed as a universal indicator of stable demographic integration. Although prior work has acknowledged that White decline or other sources of White population loss (e.g., natural decrease) may lead to fleeting increases in diversity or fragile states of integration, the focus has typically been on homogenizing neighborhoods (Farrell and Lee 2011; Lee and Hughes 2014). By contrast, our results suggest that many neighborhoods experiencing White decline—especially during the early stages of turnover—exhibit rapid increases in diversity while slowing what would otherwise be greater decreases in segregation. For many places, greater diversity emerges as a symptom of racial turnover rather than evidence of its decline.

Although these findings challenge optimistic assumptions about the association between diversity and integration, they are not wholly inconsistent with past research. Previous work has found that Whiter suburbs have historically contributed relatively more—and diverse principal cities have contributed relatively less—to overall trends in increased diversity (Fowler et al. 2016; Hall et al. 2016). Recent work also suggests that declines in the contemporary White population occur more in suburban neighborhoods in ways that can increase segregation within suburbs and new immigrant destinations (Farrell 2016; Hall 2013; Kye 2018). Collectively, these findings suggest new sources of non-White population growth and White decline in traditionally less diverse regions of the metropolis. Given their initial homogeneity, such neighborhoods can (and will) endure periods of increasing entropy before exhibiting the declines in diversity symptomatic of the final stages of racial turnover.

Our findings do not dispute the basic fact that increases in diversity represent the modal trend across much of the contemporary United States, but they do suggest the need for greater scrutiny of the underlying mechanisms generating this phenomenon. U.S. neighborhoods may be more diverse now than in previous decades, but this situation does not necessarily make stable integration more likely. In some cases, diversity has increased because of the rise of multiethnic neighborhoods where non-Whites reside alongside a stable White population presence. Such outcomes represent the clearest manifestation of the demographic integration perspective. In other cases, diversity has increased as a function of persistent segregation as Whites leave neighborhoods experiencing non-White entry. On the whole, our findings suggest that it would be unwise to project that current diversity trends will eventually mitigate residential segregation. To the contrary, White flight and racial turnover may very well drive continued increases in diversity as residential segregation endures in many areas.

Correctly identifying the proximate sources of stratification among diversifying neighborhoods is all the more important because a considerable number of persistence neighborhoods appear to be experiencing steady increases in diversity that lead effectively to integration. In places with prevalent persistence, the effects of increasing diversity may vary considerably for minority groups who have managed to achieve residence alongside White populations rather than replace them (e.g., as seems to be the case in places where turnover is more common). Here, our findings suggest a particularly concerning cleavage: White populations have grown most commonly in places where diversity leads stably to integration, but the reverse appears true for non-Whites, whose numbers have grown most in places experiencing racial turnover. To the extent that stable diversity can enhance trust and cohesion between groups—as might occur through multiethnic schools, multicultural spaces, and interracial coalitions (Tropp et al. 2018)—the bulk of these benefits appear to be inherited by Whites in places where they persist but remain the (numerically) dominant group. Non-Whites, on the other hand, appear to be inheriting communities where declines in diversity loom. Examining the implications and consequences of this divide is an important avenue for future research.

We hesitate to make definitive projections about future racial residential diversity given the limits of our research design and the complex array of underlying household, institutional, and political factors. At most, our results provide useful clues about what might be expected. Diversity decreases were already the modal outcome for our sample of later turnover blocks from 2000 to 2010, and they will likely be the modal outcome by 2030 for blocks currently in the early stages of turnover if White population losses continue. Thus, although countertrends of decreasing diversity have appeared almost nonexistent relative to dominant trends, they may become more common over time. In some cases, growing diversity within non-White populations may help stabilize or prevent such declines (Hall et al. 2016). Nevertheless, the homogenization patterns that prior research anticipated—and that were driven in our sample by White population losses—could still emerge. Whether these trends ultimately lead to patterns of balkanization in the aggregate is something to watch closely.

There are several other useful directions for future research. First, researchers should examine decade-by-decade changes in diversity and segregation using data covering a longer time horizon than the one we considered here. Doing so would allow for (1) analyses of neighborhoods (or cohorts of neighborhoods) whose racial composition tipped at different points in time (Lee and Hughes 2014); and (2) comparisons to earlier periods, when the relationship between diversity and segregation may have looked different than it does now. Second, we would like to see studies dedicate greater attention to racial/ethnic majority–minority neighborhoods, gentrifying areas, and metropolitan areas with small racial/ethnic minority populations. As discussed earlier, we focused on patterns in majority-White neighborhoods because they have been the most common sources of turnover in urban and suburban areas (Wright et al. 2020) while also generating the bulk of diversity increases over the past several decades (Hall et al. 2016). A more expansive approach could offer new insights into patterns of neighborhood change that look different from classic models of invasion–succession but increasingly compromise the link between diversity and integration (Hwang and Ding 2020).

Finally, this study's findings would be better contextualized with additional explanatory work at the individual level. Although the patterns of White loss we identified are consistent with historic patterns of racial turnover, our analyses do not explicitly examine whether the processes underlying these patterns were (primarily) racially motivated. Likewise, White households that persist may have done so for reasons other than racial amity (e.g., when natural barriers or gated communities preserve more granular levels of segregation). Geocoded data capable of accounting for neighborhood, household, and individual characteristics—and the timing of people's moves into and out of a given area—are best suited for these types of questions. Our data show that neighborhoods can exhibit remarkably similar trends in diversity, even when the changes they are undergoing point to qualitatively different demographic futures (i.e., racial turnover vs. stable coresidence). Under these conditions, diversifying neighborhoods could be mistaken for those in the process of stably integrating, and new cycles of racial turnover could be confused with progress more generally. Although greater residential diversity can blur traditional racial boundaries, it is the processes generating diversity that determine its effect on segregation.

Acknowledgments

We thank Dina Okamoto, Jennifer C. Lee, Brian Powell, John Logan, and Matthew Hall, as well as the editors and anonymous reviewers, for their thoughtful comments and helpful suggestions on earlier drafts of this manuscript.

Notes

References