Bucking the Trend: Is Ethnoracial Diversity Declining in American Communities?

Abstract

Although increasing diversity at the national scale is a well-documented trend, substantial variation in patterns of ethnoracial change occurs across American communities. Our research considers one theoretically implied path: that some communities are ‘bucking the trend’, becoming more homogeneous over time. Using 1980 through 2010 decennial census data, we calculate panethnic (five-group) entropy index scores to measure the magnitude of diversity for nearly 11,000 census-defined places. Our results indicate that while certain places reach their diversity peak in 1980 or 1990, they are few in number. Moreover, they experience a variety of post-peak trajectories other than monotonic diversity decline. Decreasing diversity is concentrated in the South and West, among places with higher levels of diversity and larger proportions of Hispanic or black residents at the beginning of the study period. These places exhibit complex shifts in racial-ethnic structure, but Hispanic succession predominates.

A major transformation is underway in the racial and ethnic composition of the United States. Since 1980 Hispanics have more than doubled their share of the national population and Asians have tripled theirs. Immigration stands out as a key driver of minority growth, but significant roles are also played by the higher fertility and youthful age structures of some groups, by intermarriage and the resulting multiethnic offspring, and by shifts in racial-ethnic identity (Hirschman 2005; Lee & Bean 2010; Lichter 2013). The effect of these demographic processes has been to shrink the size of the white slice of the population pie while expanding the slice of every other panethnic group—blacks and Native Americans as well as Hispanics and Asians. The pie-slice analogy nicely captures the essence of what we take diversity to mean: how evenly people are divided among the ethnoracial segments or categories of a population.

Diversity has become especially visible at the local level, with certain communities approaching the maximum possible magnitude of diversity, i.e., equal proportions of residents in each racial-ethnic group (Lee et al. 2012). As of 2010, for example, whites (25%), blacks (21.5%), Hispanics (22.6%), and Asians (25.5%) constitute comparable shares of the Vallejo, CA population. On the East Coast, Jersey City exhibits a similarly diverse profile (21.5% white, 23.9% black, 27.6% Hispanic, 23.5% Asian). Even in places where diversity is less extreme, it can have important consequences for the economy, housing market, schools, politics, healthcare, social services, and taxes (Bean & Stevens 2003; Clark 1998; Lichter 2012, 2013). Other studies have examined links between city- or neighborhood-level diversity and social capital, trust, crime, and intergroup relations (Hou & Wu 2009; Kubrin & Ishizawa 2012; Oliver 2010; Putnam 2007). To date, the evidence remains mixed, yielding positive, neutral, and negative assessments of diversity's impact.

Few places look exactly like Vallejo or Jersey City; substantial variation occurs across communities in their diversity magnitude, underlying racial-ethnic structure (which groups are present), and pace of change. But the primary message from the recent literature, reviewed below, is that all communities are heading in the same direction. Large gateway cities, with long histories of incorporating immigrants, continue to lead the way toward a ‘rainbow’ future that includes not only white and black residents but many Hispanic and Asian newcomers. Joining the gateways are increasingly diverse suburbs and smaller metro areas. Finally, the dispersion of Mexicans and other Hispanics to new destinations throughout rural America has the potential to simultaneously revitalize and disrupt the nature of small-town life.

Our purpose is to subject conventional wisdom about the direction of diversity change to careful scrutiny. Is ethnoracial diversification ubiquitous, or are some places ‘bucking the trend’, becoming more homogeneous over time? Such a question is worth posing because both theory and evidence suggest that diverse residential environments at a range of geographic scales (from metropolitan area to neighborhood) are inherently unstable and thus represent temporary or transitional phenomena. Using decennial census data, we search for communities that have experienced a decrease in diversity between 1980 or 1990 and 2010. Our analytic aims are to determine (a) the number of communities undergoing such a decrease relative to other types of change, (b) the paths that they have followed (their decade-by-decade shifts in diversity magnitude), (c) whether the locations of these communities set them apart from their diversifying counterparts, and (d) how their racial-ethnic structures have evolved.

We pursue these aims for nearly 11,000 census-defined places, most of which are cities, towns, boroughs, or villages. The largest places, principal cities of metropolitan areas, approximate housing and labor markets. The vast majority of all places in our sample coincide with governmental jurisdictions that have legally prescribed powers and functions. Hence they are responsible for developing fiscal and policy responses to any diversity issues that arise inside their boundaries. Some places (e.g., affluent suburbs) have employed annexation, zoning, or other mechanisms to dilute if not deter minority growth, working against gains in diversity (Jackson 1985). Elsewhere immigrant-fueled diversification is sought as a strategy to stave off demographic and economic decline (Carr et al. 2012). More generally, a place's population shapes the ethnoracial mix of locally embedded institutions such as schools, work settings, and political and civic organizations. In a number of ways, then, places constitute real communities rather than just convenient statistical aggregations.

Background

Diversity at the Local Level

That diversity suffers from a degree of imprecision should come as no surprise, given the many meanings assigned to it. These meanings range from the abstract (diversity as an ideal or moral imperative) to the concrete (diversity as a planning goal) and can be applied to income, culture, land use, housing, or other dimensions of a community. Our focus on racial and ethnic diversity continues a longstanding tradition in the urban demography literature. Even among demographers, though, diversity is sometimes conflated with residential segregation, perhaps because physical proximity among members of different groups is central to both concepts. The former typically reflects the overall ethnoracial composition of a larger spatial unit—in our case, a place—while the latter indicates the extent to which racial groups are unevenly distributed across subunits (e.g., census tracts or blocks) composing the larger unit. Thus, a diverse community may be more or less segregated contingent on whether whites, blacks, Hispanics, and Asians live in isolation from one another or share the same neighborhoods. This analytic distinction between the two concepts is confirmed by their empirical association, which remains modest and inconsistent (Defina & Hannon 2009; Iceland 2004; Parisi et al. 2013).

Recent research conducted at the local level—at geographic scales above the neighborhood but below the nation—provides little support for a homogenization counter-trend to increasing diversity. Hispanics and Asians, who historically have resided in a few gateway settings (Portes & Rumbaut 2006), are dispersing toward non-gateway states, metropolitan areas, and communities although both groups remain concentrated in the West and South (Ennis et al. 2011; Hoeffel et al. 2012; Lichter & Johnson 2006, 2009; Massey & Capoferro 2008). Within the metro context, spatial assimilation appears to be weakening the hold of inner-city enclaves. Just over one-half of all foreign-born persons nationwide now live in the suburbs of the 100 largest metro areas (Wilson & Singer 2011), and approximately three-fifths of the Asian and Hispanic residents of these areas are suburbanites (Frey 2011a). African Americans also exhibit an impressive level of suburbanization, though lower than either Asians or Hispanics.

The obvious implication of such group redistribution trends is that most kinds of communities are diversifying. Research highlights the diversity-enhancing changes underway in major metropolitan areas and their principal cities, changes often prompted by a combination of minority population gains and white losses (Berube 2003; Farrell 2005; Frey 2006, 2011b; Singer 2005; Suro & Singer 2003). In the biggest cities multigroup compositions rather than dominant white majorities are the norm, and Hispanics have supplanted blacks as the second largest group after whites (Lee et al. 2012). As Johnson and Lichter (2008) document, this Hispanic advantage is not dependent solely on the arrival of new immigrants or domestic migrants in a given destination. Instead, the higher fertility rates and the youthfulness of local Hispanic populations have made natural increase a significant component of their growth.

Because minority gains in metropolitan areas extend beyond the urban core, fewer suburbs today constitute the white bastions of popular stereotype. A number of investigations find a rise in black, Hispanic, and Asian proportions—and, by definition, a rise in diversity—in suburban rings and places (Farrell 2005; Frey 2011a; Logan 2001). In some instances, group-specific patterns of change have led to the emergence of ‘ethnoburbs’ such as those in the San Gabriel and Silicon valleys of California (Li 2009; Wen et al. 2009). Smaller metro areas and their constituent places exhibit increasing diversity as well, albeit not to the same extent that New York, Los Angeles, and other gateways do.

Rural communities have followed the national trend in direction, but they remain relatively homogeneous. Recent comparisons of racial-ethnic diversity across metropolitan and micropolitan settings show that the 2010 mean diversity levels of micro areas and rural counties (i.e., counties located in neither metro nor micro areas) are lower than those experienced by metro areas 30 years earlier (Lee et al. 2012, 2014). Mean patterns, however, hide the dramatic diversity increases that occasionally occur in rural communities and small towns. Postville, IA represents an extreme instance of how the demand for labor in meat-processing plants can attract large numbers of immigrants during a short period: Hispanics now make up a third of Postville's population (Grey et al. 2009; for other examples, see Carr et al. 2012). Sudden compositional changes of this sort may have a substantial impact on daily life when a place is not accustomed to integrating new arrivals and lacks the institutional infrastructure for doing so.

Theories and Dynamics

Despite abundant evidence that would appear to support the pervasiveness of community diversification, several studies of residential preferences and mobility suggest the potentially fragile nature of such a trend. The studies find that whites are unwilling to move into diverse neighborhoods, and they are more likely to move out of them when significant proportions of minorities or immigrants are present nearby (Charles 2006; Crowder et al. 2011, 2012; Friedman 2008; Pais et al. 2009). These dynamics can spur white population losses, eroding a setting's mixed character (Farrell & Lee 2011; Logan & Zhang 2010; but see Nyden et al. 1997). Whites’ avoidance of neighbors from other groups thus hints at an alternative scenario: a non-trivial number of communities as well as neighborhoods may be heading toward greater homogeneity.

This scenario is bolstered by three theoretical perspectives on racial-ethnic change. The first, invasion-succession, emphasizes the inevitability of white-to-minority transitions in residential environments across a range of geographic scales (Hartmann 1993; Lee 2007). The racial composition of a focal area is hypothesized to evolve gradually until the representation of blacks or Hispanics reaches some vague threshold, often referred to as the ‘tipping point’. Incumbent whites then begin to exit the area at a faster rate than normal, and few white homeseekers are willing to enter. Most vacancies wind up being filled by minority households eager to improve their residential circumstances. According to succession logic, diversity is a temporary state midway between predominantly white and predominantly minority occupancy.

The succession model has been used most often to describe racial-ethnic changes in neighborhoods (Duncan & Duncan 1957; Lee & Wood 1991; Taeuber & Taeuber 1965), but its fundamental prediction—of movement from one form of homogeneity to another—is not limited to that type of unit. Assessing redistribution trends during the 1990s, Frey (1995, 1996) foresaw a succession-like process that he labeled balkanization unfolding on a grander scale. He argued that as economic opportunities and social networks draw Hispanic and Asian immigrants to gateway communities, the working-class whites most threatened by the newcomers are likely to depart to a different metro or nonmetro area instead of to a different neighborhood in their current community. Over time these group-specific migratory flows could lead to a future marked by “demographic balkanization rather than an even increase in racial and ethnic diversity across all regions and metropolitan areas” (Frey 1995:755). Although aspects of the balkanization thesis have been criticized (Lichter & Johnson 2006; Wright et al. 1997), Frey deserves credit for recognizing that the master trend toward greater diversity might play out in checkerboard fashion at the subnational level.

Both balkanization and succession partially overlap the broader ethnic stratification (or place stratification) perspective popular in residential segregation and attainment research (Alba & Logan 1991; Rosenbaum & Friedman 2007). This perspective offers the fullest account of the forces that work against diversity. More so than the other two models, it stresses how the actions of institutions shape the racial-ethnic landscape. One example is the discriminatory treatment that blacks, Hispanics and Asians receive in many housing markets at the hands of real estate agents, lenders, and insurers (Ross & Yinger 2002; Turner & Ross 2005; Turner et al. 2013). Similarly, density zoning and other land use policies have been shown to reduce minority access to whiter communities (Rothwell & Massey 2009). While proponents of the stratification perspective recognize the critical impact of whites’ desire to live in majority white environments, they further note that co-ethnic residential preferences are reported by all groups (Charles 2006, 2007; Clark 2002). Such preferences may be especially strong among immigrants attracted to the resources and support in enclave settings. Operating together, the forces identified in the stratification model should halt the advance of ethnoracial diversity or even diminish it, fostering homogeneous communities over the long run.

Despite differences in their particulars, the succession, balkanization, and stratification perspectives tell a similar story about racial-ethnic change. They are also in agreement, at least implicitly, about the mechanical aspects of the process. Places occupied by a single group experience a floor effect with regard to diversity. As homogeneous contexts, they have nowhere to go but up, achieving more mixed compositions with the passage of time. Race-specific growth rates, however, are rarely identical for extended periods; one group often outdistances the others. When that happens, an inflection point is reached on the diversity continuum beyond which the slices of the compositional pie start to become less equal again. We would therefore expect the magnitude of ethnoracial diversity to follow a curvilinear course across decades: (1) from low (e.g., an all-white community) (2) to high, with the arrival of substantial numbers of residents from multiple minority groups, then (3) returning to low as Hispanics (or blacks or Asians) supplant whites as the dominant majority in the local population.

A Quasi-Cohort Approach

The ability to capture this entire curvilinear pattern depends upon the length of one's observational period. Places that decline in diversity during a relatively brief period would be consistent with the counter-trend of interest, as would places that increase then decrease in diversity. To identify these cases, we pursue a quasi-cohort approach, defining categories of places by the census year (from 1980 through 2010) when each place reaches its peak level of racial-ethnic diversity. The diversity categories qualify as cohorts in a broad sense: their member places all experience a common event at about the same time, and that event is expected to have important implications for their future (Shryock & Siegel 1975). However, unlike most cohorts familiar to demographers—for example, people who are born, graduate from high school, or marry in the same year¹—the assignment of a place to a cohort may not be fixed. Because of data limitations, we are unable to determine if a place has already achieved its highest diversity prior to 1980 or will do so after 2010; either possibility would make its current cohort either inaccurate or temporary. In short, the truncated temporal window available to us necessitates the quasi-cohort label for our framework.

Within that framework, the evidence reviewed earlier indicates that almost all places should belong to the 2010 peak cohort, with many presumably continuing to diversify for decades to come. From a theoretical standpoint, however, the succession, balkanization, and ethnic stratification models anticipate that some communities will peak before 2010, becoming more homogeneous thereafter. The 1980 and 1990 peak cohorts are particularly strategic because they can be tracked for at least two decades beyond their point of greatest diversity. This longer follow-up provides more leverage for addressing four key research questions.

Our first question concerns the prevalence of the counter-trend represented by the 1980 and 1990 cohorts: is increasing ethnoracial homogeneity fairly common or rare? Put another way, how large are the two cohorts? The tendency to rely on mean results leaves the prevalence of different patterns unclear. Such results can also mask variation in the sequences of change that places exhibit beyond their diversity peak. Although the relevant theories hint at a monotonic diversity decline, some communities may experience a post-peak decade of stability or even a modest increase in diversity after their initial decline. Hence, our second research question is about the decade-by-decade trajectories detected for the 1980 and 1990 peak cohorts.

In addition to documenting variation in their trajectories, we consider whether the places that comprise the two cohorts are geographically distinctive. Our third question asks if they are overrepresented in the South and West, where historical settlement patterns of African Americans (in the South) and Hispanics and Asians (in the West) may have long ago produced diverse communities primed to become less so during the last 30 years. That is, any regional association might be accounted for by high diversity itself: just as initially homogeneous places respond to the floor of the ethnoracial diversity continuum, very diverse places would be expected to descend from the ceiling over time. Greater proportions of black or Hispanic residents in 1980 could reflect nascent stages of transition, with subsequent minority gains and white losses leading to a more homogeneous population.

The manner in which the 1980 and 1990 peak cohort communities are defined and the first three questions posed about them all emphasize the magnitude of diversity, i.e., how evenly divided members of a local population are among ethnoracial categories. But measures of magnitude such as the entropy index (introduced later) do not reveal the specific groups present. A hypothetical place in which whites and Asians each constitute one-half of the population would receive the same diversity score as a place where half of the residents are Hispanic and half are black, despite meaningful compositional differences between the two. This directs attention to another dimension of diversity: its underlying racial-ethnic structure. Our fourth question raises the issue of changes in that structure for the 1980 and 1990 peak cohort places. We seek to determine which groups are present in what proportions during a place's year of maximum diversity and how such proportions compare to those observed 20 or 30 years later.

Methodology

Place Sample

Our study examines places, most of which are chartered by their respective states as cities, towns, boroughs, or villages. To satisfy the Census Bureau's definition, a place must have a concentration of population, be recognized as a named entity, and not overlap spatially with any other places (U.S. Census Bureau 1994). Places are nested within larger census geographies such as states, metropolitan areas, or nonmetropolitan counties. We have identified all places with at least 1,000 residents in the 1980, 1990, 2000, or 2010 decennial censuses. Such places are big enough to avoid major swings in racial-ethnic diversity induced by the arrival or departure of a few households from the same group. The number of places meeting this size threshold has increased from 11,774 to 15,042 during the last three decades.

To facilitate comparisons over time, we limit our analysis to the 10,924 places with populations that equal or exceed 1,000 in all four census years.² Four-fifths of these (82.2%) are municipalities, incorporated places that have primary fiscal and policy responsibility for the local consequences of diversity (e.g., offering ESL instruction in schools, providing health and social services).³ Because they are governmental units rather than mere statistical aggregations, we rely on the meaningful place boundaries in effect at each time point rather than imposing constant ones. The rationale here is that any changes in the racial-ethnic composition of a place's population caused by annexation or loss of land would matter more for the community as constituted at that moment than for its previous or future territory.⁴

The 10,924 places in the analytic sample range widely in size. New York City, with over 8.1 million inhabitants in 2010, is the largest place every census year. The smallest place differs from one year to the next, never surpassing the 1,000 mark by more than two residents. Cities of a half-million or more constitute barely 2% of our sample. They do, however, house nearly one-fifth of all people living in the sample places. At the other extreme, communities with populations under 10,000 represent about three-fourths of all places but only capture between 14% and 20% of the total place population, depending upon the year. More places are embedded in metropolitan than nonmetropolitan settings (64.3% versus 35.7%). They are also distributed unevenly across regions, with greater proportions in the South (34.4%) and Midwest (30.8%) than in the Northeast (19.3%) or West (15.4%).

Measuring Diversity

As input for our diversity measures, we have extracted census summary file (SF) data on the 1980, 1990, 2000, and 2010 ethnoracial composition of each place in the sample. The race by Hispanic origin crosstabulation in SF1 (for 1990-2010) and SF2A (for 1980) yields counts of Hispanics of any race and of non-Hispanic whites, blacks, Asians, Pacific Islanders (tabulated separately from Asians since 2000), Native Americans (American Indians and Alaska Natives), multi-race individuals (since 2000), and those reporting some other racial identity. For all years we combine Asians and Pacific Islanders into a single category, hereafter labeled Asian. Small numbers justify the creation of an ‘other’ category consisting of Native Americans and multi-race and other-race persons. These recodes leave us with five panethnic groups—Hispanics and non-Hispanic whites, blacks, Asians, and ‘others’—that are exhaustive, mutually exclusive, and largely comparable throughout the study period.⁵

Two properties of a place's population contribute to its diversity: the number of different racial-ethnic groups it comprises and the sizes of the groups relative to each other. On intuitive grounds, a place whose population is marked by evenness—made up of many groups identical in size—would be considered highly diverse, while a place with only a single group present would have no diversity. These hypothetical places differ in their magnitude or level of diversity, which we measure with the entropy index, symbolized by E (for more detailed treatments, see Reardon and Firebaugh 2002; White 1986). Once the appropriate number of groups is decided upon, E provides an absolute standard against which diversity can be judged. Such a standard facilitates over-time and across-place analyses, avoiding complications that arise when relative benchmarks are employed (e.g., the year-specific ethnoracial composition of the nation, state, or metropolitan area in which a place is located).

The entropy index can be formally defined as follows:

$$E=\sum _{r=1}^R{p}_r\ln \left(\frac{1}{p_r}\right)$$

where p_r refers to racial-ethnic group r's proportion of the population in a place and R signifies the number of such groups. The maximum value of the index (the natural log of R) occurs only when all groups are the same size. Because there is no fixed upper limit, a place consisting of more equal-sized groups will produce a higher E score than one with fewer equal-sized groups. To standardize E, we divide it by its maximum (1.609 in our five-group scenario) then multiply by 100. Thus, a value of 100 indicates complete heterogeneity, with each of the five panethnic groups representing one-fifth of a place's population. The two places mentioned in the introduction, Vallejo and Jersey City, approach this level of diversity, with 2010 E scores of 94.5 and 92.4 respectively. An E of 0 denotes complete homogeneity, when all inhabitants belong to the same group (for examples of homogeneous places, see Lee et al. 2012.)

The Simpson interaction index (denoted by S) is another popular measure of diversity. In our case, it can be interpreted as the probability that two residents randomly drawn from the population of a place will belong to different ethnoracial groups. Supplementary analysis reveals that E and S scores are correlated at .985 or greater across the 10,924 sample places in each year of the study period. Given this strong association, the very similar diversity patterns documented by the two measures come as no surprise. We show results only for E to avoid redundancy.

Ideally, the mean entropy scores should reflect the experiences of most Americans. Due to the disproportionate number of small places in our sample, however, the ‘average’ diversity picture is likely to be distorted if all places are treated as equivalent cases. Weighting offers a simple solution to this problem. Throughout the aggregate portions of the analysis, we weight E by the ratio of a place's population to the summed population of all places in its reference category. The weighting procedure produces a series of means depicting the diversity magnitudes that average residents of places in different regions, types of areas (metro/nonmetro), and size classes encounter in their respective communities. Weighted or unweighted, E does not provide a detailed breakdown of the racial-ethnic structure of a place, i.e., the specific groups constituting the local population. To capture this dimension of diversity, we employ pie charts, bar graphs, and a ‘majority rule’ typology (introduced later) that summarize the group percentages underlying the entropy scores. The percentages are, once again, weighted by place population size. They reveal the proportions of the five panethnic groups to which the average person living in a particular kind of place is exposed.

Other Variables

At various stages in the analysis, places are classified by region, area status, and population size. We use Census Bureau definitions of region to situate places in the Northeast, Midwest, South, or West. We code places as metropolitan if they fall within the boundaries of one of the 366 metro areas defined by the U.S. Office of Management and Budget in 2009. The 2009 boundaries have been imposed throughout the study period, creating a geographically constant measure of metropolitan status at all four time points. (Any places outside of metro areas are coded as nonmetro in status.) Place population size is treated as a continuous variable when calculating correlation coefficients. For other purposes, though, we assign places to five size categories ranging from 500,000 or more to under 10,000. Some places gain or lose enough residents in a decade to switch categories, which they are allowed to do.

The region, area, and size variables are joined in the correlational analysis by a place's 1980 entropy index score and its proportions of blacks and Hispanics in that year. These characteristics, chosen to identify communities closer to the diversity ceiling or inflection point at the start of the study period, may prove helpful in understanding regional differences in the likelihood of racial-ethnic homogenization.

Results

Basic Patterns: 1980-2010

As a preliminary step, we briefly describe the magnitude and structure of ethnoracial diversity over time for all places in the sample, not just for those ‘bucking the trend’. The top row of Table 1 reveals regular gains in average diversity magnitude (weighted) from one census year to the next, with the entropy index registering a 20-point increase between 1980 and 2010 and its largest percentage change between 1990 and 2000. Diversity levels are greatest in the West, although Northeastern and Southern places have narrowed the gap. Midwestern places remain the least racially and ethnically diverse; their mean 2010 E score is roughly two-thirds that of their Western counterparts. Metropolitan places now hold a larger advantage in diversity over nonmetropolitan places than they did three decades earlier. With respect to population size, the bigger the place, the more diverse it tends to be. This generalization holds despite the smaller change in diversity magnitude observed for major cities with 500,000 or more residents.

Table 1.

Mean Place Diversity, Total and by Region, Area Status, and Population Size

Type of Context	Mean E				2010–1980 Difference	N of Places
	1980	1990	2000	2010
Total	37.16	42.25	51.80	57.31	20.15	10,924
Region
Northeast	38.01	44.80	55.84	61.48	23.47	2,111
Midwest	26.07	28.95	38.21	45.25	19.18	3,367
South	39.51	43.32	52.57	58.74	19.23	3,759
West	46.41	52.00	60.18	62.87	16.47	1,687
Area Status
Metro	39.22	44.47	54.13	59.48	20.26	7,020
Nonmetro	22.87	25.24	32.69	38.47	15.60	3,904
Population Size
500k+	64.56	68.72	75.44	76.57	12.00	24*
100-499k	45.73	51.28	61.18	65.82	20.09	189*
50-99k	34.13	41.47	51.89	57.59	23.46	354*
10-49k	28.00	32.74	42.01	48.98	20.97	2,377*
<10k	19.45	22.32	29.35	35.18	15.73	7,979*

^*Mean number of places across 4 time points.

Table 1 also provides an opportunity to summarize three sensitivity tests (not shown). First, when we replicate the table for places excluded from our analytic sample (see note 2), the general findings stay the same. Mean diversity rises over time among places in each region and type of area. However, the E scores for the excluded places are much lower, consistent with their small populations and nonmetropolitan locations. In the second sensitivity test, the results in the table hold when places are defined in constant 2010 boundaries at each time point (note 4). Despite some data quality problems with the 1980 Geolytics file, the mean E scores based on constant and variable boundaries mirror each other, with ratios of the former to the latter typically falling in the .95 to 1.03 range for all categories of places.

Finally, we have calculated entropy index values using only the four largest panethnic populations (whites, blacks, Hispanics, and Asians) to determine to what extent shifts in diversity are affected by the addition of multi-race individuals to the ‘other’ category in 2000 (note 5). This exercise yields Es that are a few points higher than those in Table 1 but that are otherwise similar with respect to trend line and variation by region, area status, and size. One difference in the recalculated results is that the mean 1990-2000 percentage changes in E scores no longer exceed the 1980-1990 and 2000-2010 changes for every category of place, as they do in the original table. Our inference is that the expanded ‘other’ category modestly boosts diversity.

Shifts in racial-ethnic structure associated with the upward trend in diversity are visually conveyed in Figure 1. Across all regions, types of areas, and size categories, whites comprise a shrinking share of place populations. The mean black share has been fairly stable except in major cities (500,000+), where it has declined from 29% in 1980 to 21.5% in 2010. The key diversity driver is the impressive growth of Hispanics. By 2010, they represent almost one-third of the populations of Western places (31.6%) and the largest cities (31%), on average. Asian representation has also increased dramatically, more than doubling during the study period across most contexts in the figure. However, Asians continue to have a minimal presence in nonmetropolitan communities and in small towns under 10,000, where members of the ‘other’ panethnic category—primarily Native Americans—exceed them in 2010.

Figure 1.

Mean Racial-Ethnic Structure of Places, Total and by Region, Area Status, and Population

Diversity Changes in Peak Cohorts

It is tempting to infer from the changes for the full sample that all places have experienced steady diversity gains during the last three decades. Such an inference would be inaccurate. The curves in Figure 2 distinguish among four peak quasi-cohorts of places. The two of primary interest to us, the 1980 and 1990 peak cohorts, both exhibit mean E values of approximately 50 in 1980. Then the Es for the 1980 cohort decline to 37.3 by 2010 while those for the 1990 cohort rise to 60.5 in 1990 before returning to the near-50 level 20 years later. The communities in these two cohorts seem to be bucking the trend toward greater diversity: they have become more racially and ethnically homogeneous over an extended time period. But they are also few in number. Only 452 places belong to the 1980 peak cohort and 145 to the 1990 cohort; together they make up just 5.5% of the sample. By contrast, 9,422 places reach their highest diversity magnitude in 2010 (the curve anchored by solid circles), with another 905 peaking in 2000 (the uppermost curve). We thus conclude that the expected counter-trend to diversification exists, but as of 2010 it can hardly be considered pervasive.

Figure 2.

Mean Place Diversity (E) by Peak Cohort

The theoretical perspectives that anticipate this counter-trend further suggest a uniform pattern of decline once communities have passed their presumed diversity apex. Although the curves for the 1980 and 1990 peak cohorts in Figure 2 appear consistent with the notion of uniformity, their aggregate nature—they connect means across years—may obscure variation in the trajectories followed by individual places within each cohort. To address the issue of uniform versus heterogeneous trajectories, we inspect decade- and place-specific changes in diversity magnitude. If the E score for a place rises or falls by less than half a point from one census year to the next, ethnoracial diversity has been coded as stable for that decade. Changes that exceed the .5 limit are coded as diversity increases or decreases, depending upon direction. Table 2 summarizes the results from this exercise.⁶

Table 2.

Diversity Trajectories of Places in 1980 and 1990 Peak Cohorts

Cohort/Pattern	Direction of Change in E			Percentage of Cohort
	1980–1990	1990–2000	2000–2010
1980 Peak
1	Decrease	Decrease	Decrease	41.37
2	Decrease	Decrease	Increase	16.15
3	Decrease	Increase	Decrease	13.94
4	Decrease	Increase	Increase	11.50
5	Decrease	Stable	Decrease	6.42
6	Decrease	Increase	Stable	3.10
7	Decrease	Decrease	Stable	1.99
8	Decrease	Stable	Increase	1.77
9	Stable	Decrease	Decrease	1.55
10	Decrease	Stable	Stable	1.11
11	Stable	Stable	Decrease	.66
12	Stable	Decrease	Stable	.22
13	Stable	Decrease	Increase	.22
1990 Peak
1	Increase	Decrease	Decrease	62.07
2	Increase	Decrease	Increase	16.55
3	Increase	Decrease	Stable	6.21
4	Increase	Stable	Decrease	5.52
5	Stable	Decrease	Decrease	3.45
6	Increase	Stable	Stable	2.76
7	Stable	Decrease	Stable	1.38
8	Stable	Stable	Decrease	1.38
9	Stable	Decrease	Increase	.69

* Change defined as +/− .5 point in E score.

According to the top panel, the modal path for a member of the 1980 peak cohort is to undergo diversity declines during each of the three post-peak decades (pattern 1). However, this trajectory holds for a plurality (41.4%) of all places in the cohort, not a majority. In fact, more places (46.7%) experience an increase in diversity at some point after their 1980 peak, as reflected in patterns 2, 3, 4, 6, 8, and 13. The theoretically implied trajectory of declining E scores in both 1990-2000 and 2000-2010 occurs for two-thirds of 1990 peak cohort places (patterns 1 and 5 in the lower panel), but one-third display some other pattern. What stands out about Table 2, then, is the wide range of trajectories identified. Spells of stable or increasing diversity are not uncommon among places past their supposed historic highs.

The Geography of Diversity Decline

Heterogeneous trajectories notwithstanding, the 1980 and 1990 peak cohort communities remain distinct from the vast majority of places in our sample: they all have experienced a non-trivial and, in some instances, prolonged decrease in the level of racial-ethnic heterogeneity. We now ask whether they are geographically distinctive as well. That is, are a disproportionate share of them located in the two regions of the country known for historical concentrations of blacks (the South) and Hispanics and Asians (the West)? Given this limited question, an answer can be sought via the simple procedures of mapping, difference-of-proportions tests, and correlational analysis. For all three procedures, we combine the 1980 and 1990 peak cohort places into a single category since preliminary work reveals quite similar results when the cohorts are examined separately. Combining the two cohorts also maximizes the number of cases available.

The maps in Figure 3 highlight the contrast in spatial distribution between the 1980-1990 peak cohort places and their counterparts in the much larger 2010 cohort. While the latter places are ubiquitous, blanketing all portions of the nation, the former exhibit a Sunbelt proclivity. Most of the 1980-1990 places can be found in a lower tier that stretches from the Pacific to the South Atlantic seaboard. Clusters of these places are evident in the Central Valley and coastal areas of California, in the Panhandle and southern region of Texas, and across the Black Belt, especially near the Mississippi River. Difference-of-proportions tests confirm that 1980-1990 peak cohort members are significantly more likely to be located in the South (53.3% of peak cohort places vs. 33.3% of all other places in the sample) or West (29.8% vs. 14.6%). The inclusion of Texas and Florida in the Census-defined South hints at the early boosts to community diversity provided not only by African Americans but by Hispanics in that region.

Figure 3.

Location of 1980-1990 and 2010 Peak Cohort Places

Despite the coarse nature of regional comparisons, significant zero-order correlations exist between our measure of diversity decline—whether a place belongs to the 1980-1990 peak cohort (coded 1 if yes, 0 if no)—and location in the South or the West (r = .10 in both cases). But these already modest coefficients diminish in magnitude when we calculate a series of partial correlations, which indicate the independent association of peak cohort membership with a single place characteristic (such as regional location) while controlling other relevant 1980 characteristics of places.⁷ Perhaps the most noteworthy relationship uncovered by this analysis is for ethnoracial diversity in 1980 (r_p = .27): a high E score at the beginning of the period significantly increases the chances of a community becoming less diverse. Stronger findings emerge when black (r_p = .30) and Hispanic (r_p = .45) proportions are substituted for the entropy index. The size of the 1980 Hispanic coefficient, in combination with rapid Hispanic growth during subsequent decades, suggests that the nontrivial representation of this group at an early time point may portend homogenization down the road. More generally, region serves as a crude proxy for the ‘premature’ diversity promoted by historical minority settlement patterns.

Unpacking Racial-Ethnic Structure

The disproportionate representation of Hispanics and African Americans in the 1980 and 1990 peak cohort places underscores the potential importance of the structure of ethnoracial diversity in addition to its magnitude. To obtain a snapshot of how that structure has changed, we compare the weighted mean racial composition of the places in both of these quasi-cohorts at the beginning and end of the study period. The pie charts for the 1980 cohort in the top panel of Figure 4 document a transition from a no-majority structure in 1980—where none of the panethnic groups constitutes more than half of the local population—to Hispanic dominance by 2010. As the average Hispanic slice of the pie has expanded, the black and especially the white slices have shrunken. A somewhat different story emerges for the 1990 peak cohort places in the bottom panel. These places exhibit majority white populations in 1980 and majority Hispanic populations in 2010, but the Asian and black shares have grown over three decades. Once again, the percentage of white residents decreases in dramatic fashion.

Figure 4.

Changes in Mean Racial-Ethnic Structure of 1980 and 1990 Peak Cohort Places

The general pattern of Hispanic succession for the 1980 and 1990 peak cohorts diverges from the compositional changes experienced by the rest of the places in our sample (i.e., the 2000 and 2010 peak cohort places combined). Although the average racial-ethnic structure of the latter places has shifted, it has done so more gradually. By 2010 the mean percentages of Hispanics (18.3%), Asians (5.9%), and ‘others’ (3%) are roughly three times greater than 30 years earlier. Blacks make up 14% of the 2000/2010 peak cohort places in 2010, about the same as their 1980 proportion (13.2%) but falling behind Hispanics. The greatest contrast with the 1980 and 1990 peak cohorts is the persistence of whites as the majority group in the more recent cohorts. How long they will keep that status is uncertain, given their slippage in mean representation from a little over three-fourths (77.3% in 1980) to a little under three-fifths (58.9% in 2010) of the 2000/2010 peak cohort populations.

As with mean diversity magnitudes, mean group proportions may mask the frequency of different kinds of changes in racial-ethnic structure. A simple ‘majority rule’ typology can be employed to overcome this problem. Following prior research (Farrell & Lee 2011; Lee et al. 2012, 2014), we classify places in the 1980 and 1990 peak cohorts into white, black, Hispanic, Asian, or ‘other’ majority types based on which group captures more than 50% of the population in a particular year. White majority places are then divided into dominant (90%+ white) and shared (51-89% white) subtypes. Lastly, we define no-majority places as those in which multiple groups are present but none achieves more than a plurality. The majority rule approach uses an absolute standard (over one-half of all community residents belong to the same group) that is easy to understand and that carries meaning ‘on the ground’ in terms of group visibility and potential consequences for local politics, schools, businesses, and other institutional domains. Such a standard also facilitates comparisons, which become difficult to interpret when types of racial-ethnic structure are defined in temporally or geographically variable ways.

An examination of the 1980 and 2010 frequency distributions for types of places in the 1980 peak cohort and of the 1990 and 2010 distributions for types of places in the 1990 cohort (not shown) yields three main findings. During the 30- and 20-year observation periods, (1) the percentage of dominant white places increases while the percentage of shared white places declines sharply, (2) majority Hispanic places become more common, as do majority black and Asian places to a lesser extent, and (3) virtually all no-majority places transition to some sort of majority group status. These results differ in important respects from the changes for the rest of our sample, where dominant white communities are now much rarer but the percentages of shared white and no-majority places (i.e., more diverse settings) have increased substantially (Lee et al. 2012, 2014). In short, members of the 1980 and 1990 peak cohorts are bucking the trend with respect to the structure as well as the magnitude of diversity.

Table 3 supports a more nuanced version of the same conclusion. The top panel of the table is a 30-year transition matrix for 1980 peak cohort places; the bottom panel is a 20-year matrix for the 1990 cohort. Each row identifies a different type of ethnoracial structure (based on our majority rule scheme) at t₁, while the columns capture the type of structure at t₂. Note the high values along the diagonals of the matrices: nearly all peak cohort places that begin as dominant white or majority black, Hispanic, or ‘other’ stay that way. The proportion of the majority group in these places actually increases over time, contrary to the larger diversity trend. Only shared white and no-majority places are likely to exhibit a different t₂ than t₁ structure. In both types, a minority group becomes the majority or, in the former, whites become dominant.

Table 3.

Type-of-Place Transitions for 1980 and 1990 Peak Cohorts

Sample/Type of Place	2010 Type of Place							N of Places
	Dominant White	Shared White	Majority Black	Majority Hispanic	Majority Asian	Majority Other	No Majority
1980 Peak Cohort
1980 Type of Place
Dominant White	100.00	-	-	-	-	-	-	(35)
Shared White	21.51	36.56	19.35	21.51	1.08	-	-	(93)
Majority Black	-	-	98.28	1.72	-	-	-	(116)
Majority Hispanic	-	-	-	100.00	-	-	-	(142)
Majority Asian	-	-	-	-	100.00	-	-	(1)
Majority Other	-	-	-	-	-	100.00	-	(14)
No Majority	-	1.96	27.45	66.67	3.92	-	-	(51)
N of Places	(55)	(35)	(146)	(198)	(4)	(14)	(0)
1990 Peak Cohort
1990 Type of Place
Dominant White	100.00	-	-	-	-	-	-	(19)
Shared White	5.00	32.50	17.50	42.50	-	-	2.50	(40)
Majority Black	-	-	91.30	8.70	-	-	-	(23)
Majority Hispanic	-	-	-	96.15	-	-	3.85	(26)
Majority Asian	-	-	-	-	-	-	-	(0)
Majority Other	-	-	-	-	-	100.00	-	(4)
No Majority	-	-	9.09	72.73	18.18	-	-	(33)
N of Places	(21)	(13)	(31)	(68)	(6)	(4)	(2)

Profiles of Homogenizing Places

Illustrations of the kinds of changes in racial-ethnic structure that accompany declining diversity are shown in Figure 5. Among places in the 1980 quasi-cohort, the Chicago suburb of Calumet Park adheres to the classic succession model. Once home to families of Polish and Irish ancestry, Calumet Park began to attract African Americans at an accelerating rate in the late 1970s, fueling racial tensions (Bigott 2005). The community has since transitioned from a shared white structure to a majority black one over three decades, the proportion of black residents nearly tripling to 88% by 2010. In densely settled Hialeah, adjacent to Miami, diversity decreases as an already substantial Hispanic majority—dominated by Cubans but with growing populations of Nicaraguans, Colombians, and other Hispanic origin groups—reaches the 95% mark, up from 75% in 1980. The Los Angeles suburbs of Lynwood and Monterey Park both lack a majority group in 1980, but their three-group compositions erode in response to Hispanic (Lynwood) and Asian (Monterey Park) growth. Monterey Park has been heavily studied because of the social, political, and economic consequences of its emergence as a Chinese-oriented ethnoburb that also houses non-trivial numbers of Japanese, Vietnamese, Pacific Islanders, Hispanics, and whites (Fong 1994; Horton 1995; Li 2009).

Figure 5.

Changes in Racial-Ethnic Structure of Selected Places in 1980 and 1990 Peak Cohorts

The lower half of the figure shows that no-majority structures in two 1990 peak cohort places, Passaic, NJ, and Chamblee, GA, shift to the Hispanic majority type between 1990 and 2010. The relatively affordable housing available in both places and their proximity to major employment hubs (New York City and Atlanta, respectively) have proven attractive to immigrants. Jennings, MO, in suburban St. Louis, resembles Calumet Park, trending from a 50-50 white-black mix toward black dominance (89.5%) during the 20-year period. Finally, Pine Ridge, SD, shows how an already homogeneous community (1990 E = 16.0) can become even more so (2010 E = 11.3). As the capital of the Oglala Sioux, Pine Ridge is dominated by residents in the ‘other’ racial-ethnic category. These residents, virtually all Native American, increase by over 700 while miniscule white and black groups decrease in absolute size and the Hispanic and Asian categories add eight members combined over two decades.

Pine Ridge and the other communities in Figure 5 remind us of the complex demographic dynamics that underlie changes in the structure and magnitude of ethnoracial diversity. A place may experience a diversity decline due to one group increasing in size as the others decrease, all groups increasing but one at a much faster rate, or some other combination of group-specific population gains and losses. One combination that is rare among the 597 places in the 1980 and 1990 quasi-cohorts is the pattern of white growth and minority decline consistent with large-scale gentrification. Only 34 places exhibit this pattern, and only two of them have total and minority (black and Hispanic) populations large enough to merit comment.⁸ Hoboken, across the Hudson River from Manhattan and part of New Jersey's emergent Gold Coast, has made the transformation from industrial town to ‘hip’ community over the last three decades but has become less diverse, gaining roughly 14,700 whites and losing 9,500 Hispanics. St. Augustine, FL, a centuries-old city with a history of racial conflict (Colburn 1985), features architecturally distinctive—and increasingly expensive—housing. Since 1980, its total population has fluctuated between 9,500 and 13,000. Nearly 1,400 more whites live in St. Augustine now than at the start of the period, and 1,000 fewer blacks.

Conclusion

The details of specific peak cohort cases should not obscure the most obvious conclusion to be drawn from the evidence at hand: that ethnoracial homogenization is a minor phenomenon, with barely one in 20 communities manifesting lower diversity now than they did in 1980 or 1990. Put another way, the vast majority of places throughout the U.S. are following rather than bucking the national trend, growing more diverse over time. The reach of diversification can be seen in its occurrence across regions, in nonmetropolitan as well as metropolitan settings, and among places of all population sizes. Increasing racial-ethnic diversity is pervasive with respect to direction if not pace or magnitude, visible far beyond the boundaries of large immigrant gateways such as New York or Los Angeles. In backhanded fashion, then, our results reinforce the narrative of diversity as demographic master trend.

Our results would also seem to pose a challenge to theory. The succession, balkanization, and ethnic stratification perspectives all predict a slowdown or reversal of ethnoracial diversification at the local level. The 1980 and 1990 peak cohort places on which we have focused conform to this prediction, but very few communities overall are trending toward a more homogeneous state. Of course, any conclusion about theoretical support (or lack thereof) must be accompanied by a pair of important caveats concerning the limited period of observation. First, a three-decade window strikes us as too narrow to rule out the possibility that some places may change cohort membership, surpassing what were assumed to be their all-time diversity highs. Second, the narrow window precludes a definitive estimate of how many places will eventually follow paths similar to those documented for the members of the 1980 and 1990 peak cohorts. The aggregate pattern for the 2000 peak cohort is potentially instructive (see Figure 2). Of the roughly 900 places achieving maximum E scores in 2000, one-half display a drop of two or more points during the next 10 years. What proportion of the 2010 cohort, which comprises over 9,400 cities, suburbs, towns, and villages, will undergo a comparable or larger decline in diversity magnitude two or three decades downstream?

If the 1980 and 1990 quasi-cohorts are harbingers rather than anomalies, we can venture an educated guess about the future racial-ethnic structure of the typical community. Assuming a continuation of immigrant flows from Mexico and other Latin American origins, that structure will take on a decidedly Hispanic character. Such a compositional shift remains probable even in the unlikely event that a significant policy initiative shuts down immigration. The current size of the Hispanic population, along with its young age and high rate of natural increase, has already infused Hispanic growth with sufficient momentum to drive local demographic change indefinitely. Based on our analysis, some highly diverse communities, particularly smaller ones, will transition toward greater homogeneity, culminating in a Hispanic majority. This scenario could become more common if whites, already a declining share of the national population, enter a sustained period of natural decrease (Roberts 2013).

We do not, however, envision a future in which ethnoracial homogenization is the universal trend among American communities. The variety of trajectories experienced by the 1980 and 1990 peak cohort places suggests an uneven course after the presumed apex in diversity has been passed. Indeed, increasing homogeneity and increasing diversity can be thought of as complementary rather than incompatible processes once a hierarchical conception of community is adopted. At the scale of the metropolitan area, the growth of multiple minority groups in the presence of a still-substantial white majority means that diversity will continue to climb for an extended period. But the central cities, suburban municipalities, and other places that make up the metropolis are likely to exhibit divergent patterns of change, including a turn toward greater homogeneity. A key question moving forward is whether place-level homogenization (or, at a minimum, a deceleration of the diversity trend) occurs more often in larger metropolitan areas, where the sheer number of local jurisdictions could facilitate the operation of the sorting mechanisms identified by theory: housing market discrimination, zoning ordinances, own-group residential preferences, and the like. If so, the hallmark of a highly diverse metropolitan region may be the segregation of whites, blacks, Hispanics, and Asians across the places nested within its boundaries.