DATA AND METHODS

Our principal data source is the Decennial Census of Housing and Population for 1970, 1980, 1990, 2000, and 2010 and the 2008–2010 American Community Survey. We extracted data on measures of residential segregation and spatial isolation for 1980–2010 from Logan and Stults (Reference Logan and Stults2011) for all metropolitan areas and divisions (hereafter MSAs) in the United States as defined in 2009. For 1970 we used data from the professional version of Social Explorer^{Footnote 1} at the census tract level to compute segregation and isolation indices for MSAs as defined in 2009. Our dataset consists of a balanced panel of 287 consistently defined MSAs for which we were able to compute segregation indices for 1970–2010. The MSAs included in our analysis are listed in Appendix A.

Here we focus on two of segregation’s five constituent dimensions: unevenness and isolation (Massey and Denton, Reference Massey and Denton1988a). We measure unevenness using the well-known index of dissimilarity, which captures the degree to which the residential distribution of any two groups departs from the ideal of evenness. In an even distribution, each neighborhood has the same proportion of minority and majority members as the metropolitan area as a whole. Our indicator of neighborhood is the census tract and we consider three minority groups—non-Hispanic Blacks, non-Hispanic Asians, and Hispanics and compare their residential distribution to that of non-Hispanic Whites. Under these circumstances, the index of dissimilarity states the relative percentage of minority group members and non-Hispanic Whites who would have to exchange tracts to achieve an even residential distribution.

We measure a group’s spatial isolation using the P* index, which gives the minority percentage within the neighborhood of the average minority member. The Black isolation index, for example, gives the percentage Black in the neighborhood of the average African American residing in a particular metropolitan area. Whereas the dissimilarity index is invariant with respect to the minority-majority composition of a metropolitan area, the isolation index directly depends on the relative number of minority versus majority group members.

In order to consider the determinants of residential segregation and spatial isolation we assembled data on a variety of variables that prior studies have shown to be relevant in shaping residential outcomes in U.S. metropolitan areas, which are listed in Table 1. Until now investigators have been unable to measure variation in the extent of racial-ethnic prejudice across metropolitan areas, owing mainly to the cost of developing reliable survey estimates from probability samples of hundreds of different areas but also to the reluctance of respondents to admit to having prejudicial sentiments. Google Trends, however, offers new opportunities to assess topics that were previously difficult for survey researchers to tackle (Stephens-Davidowitz Reference Stephens-Davidowitz2013). For example, the most extreme expression of anti-Black sentiment and the harshest epithet one can apply to an African American is the word “nigger” and when we entered variations on this term into Google Trends we found it to be the subject of a large volume of internet searches that yielded a robust and quite variable distribution of frequencies across metropolitan areas since 2004.

Table 1.

Independent Variables Used to Predict Segregation Outcomes for Blacks, Hispanics, and Asians

Stephens-Davidowitz (Reference Stephens-Davidowitz2013) performed a similar operation using Google Trends and found that the resulting index correlated strongly with other known measures of racial prejudice at the aggregate level; the index strongly predicted voter turnout for Obama across market areas in the 2008 presidential election. Whereas he used market areas, we employed metropolitan areas, which are smaller, and found that in some the volume of searches was too small to support a reliable index, and in these cases we substituted the state-level search frequency. On this index, the five most racist metropolitan areas were Flint, MI, Altoona, PA, Charleston, WV, Scranton, PA, and Wheeling, WV. The five least racist were Salt Lake City, UT, Ogden, UT, Provo-Orem, UT, Honolulu, HI, and Napa, CA.

When we entered various pejorative terms for Asians (chink, gook, etc.) into Google Trends, we did not find a sufficient volume of searches to provide a reliable index of Anti-Asian bias across metropolitan areas, suggesting a much lower level of hostility against this group. Likewise, when we entered various pejorative terms for Latinos into the system (spic, beaner, etc.) we also came up empty. However, recent decades have seen the rise of a Latino threat narrative in the media and public discourse tied to the framing of Latino immigrants as “illegal” (owing to undocumented migration) and therefore by definition “criminals” and “lawbreakers” (Chavez Reference Chavez2001, Reference Chavez2008). When we entered variations on the term “illegal immigrant” into Google Trends we again encountered a rather large volume of searches that yielded a robust and variable distribution of frequencies across metropolitan areas. As before, we substituted the state-level frequency whenever the volume was too low to sustain measurement within a particular metropolitan area. According to this index the lowest levels of anti-Latino sentiment were observed in Honolulu, HI, Bangor, ME, Cleveland, OH, Detroit, MI, and Lewiston, ME, whereas the highest levels occurred in Santa Barbara, CA, El Paso, TX, Brownsville, TX, Phoenix, AZ, and Tucson, AZ.

In addition to White animus toward certain minority groups, as noted above researchers have also demonstrated a strong causal connection between restrictive density zoning and both class and racial segregation (Rothwell and Massey, Reference Rothwell and Massey2009, Reference Rothwell and Massey2010). In their survey of local land regulations prevailing in forty-nine metropolitan areas, Pendall et al. (Reference Pendall, Puentes and Martin2006) asked 1,677 governmental units to report the maximum allowable density permitted in the jurisdictions they controlled. Localities that allowed less than four units per acre were coded 1; those that allowed four to seven units were coded 2; those permitting eight to fifteen units were coded 3; those allowing sixteen to thirty were coded 4; and those permitting thirty or more units per acre were coded 5. Rothwell and Massey (Reference Rothwell and Massey2009) computed the average density score across jurisdictions for each metropolitan area and found that it strongly predicted segregation, even after the application of rigorous controls.

In order to establish the causal effect of density zoning on segregation, they derived a prediction equation to estimate the density score as an instrumental variable, using year of statehood as the principal exogenous identifier. We borrowed their prediction equation and used it here to estimate density instruments for all metropolitan areas in our data set according to the following formula (1). Following Rothwell and Massey (Reference Rothwell and Massey2009), all predictors are defined as of 1980 except for the intergovernmental transfer variable, which was defined as of 1967 according to their formula.

The relative size of minority groups has long been recognized as a key determinant of segregation and discrimination. Sociologists have long argued that larger minority groups pose a greater threat to majority interests than smaller ones, in both symbolic and practical terms (Blalock Reference Blalock1967; Blau Reference Blau1977; Blauner 1972; Lieberson 1980). Symbolically, a larger minority population increases the visibility and salience of group members in public. Practically, more minority group members intensify competition for scarce public and private resources and can be expected to yield higher levels of residential dissimilarity.

In addition, the minority percentage has a direct mathematical relationship with the P* isolation index, essentially setting its lower bound. In a metropolitan area that is 20% Black and 80% White, for example, the minimum possible isolation index for African Americans is twenty, which would occur when the two groups are evenly distributed across neighborhoods (yielding a dissimilarity index of zero), thereby producing a distribution where every neighborhood is 20% Black. The upper bound, of course, would be 100, which would occur when all Blacks lived in neighborhoods that were 100% Black and all Whites lived in neighborhoods that were 100% White.

To a large extent, therefore, residential isolation is produced in urban areas by the confluence of a high level of dissimilarity with a large percentage of minority group members. Apart from these structural influences, the degree of segregation and isolation has been found to vary according to the socioeconomic status of the group in question (Denton and Massey, Reference Denton and Massey1988; Iceland Reference Iceland and Goering2007; Iceland et al., Reference Iceland, Sharpe and Steinmetz2005). Given that U.S. housing markets are segmented by price, and that wealth and income continue to vary sharply by race and ethnicity, intergroup differences in socioeconomic status translate into differences in residential status, leading to segregation (Massey and Denton, Reference Massey and Denton1985, Reference Massey and Denton1993). In general, the greater the gap in socioeconomic status between Whites and minorities, the greater the level of minority segregation and spatial isolation (Alba and Logan, Reference Alba and Logan1991, Reference Alba and Logan1993; Massey and Denton, Reference Massey and Denton1987). Hence our data set includes two indicators of relative status: the ratio of minority to White household income and the ratio of minority to White college graduates. In each case, we used census data to compare minority group members (Blacks, Hispanics, or Asians) to Whites living in the same metropolitan area. In addition, since homeowners have a greater stake in the status of a neighborhood than do renters and thus tend to behave more conservatively in response to perceived threats (Hirsch Reference Hirsch1983; Sugrue Reference Sugrue1996), we also control for the relative number of homeowners in each metropolitan area.

Research has generally shown that minority segregation and isolation vary systematically by metropolitan demographic circumstances, being greater in metropolitan areas that have more inhabitants (Massey and Denton, Reference Massey and Denton1987, Reference Massey and Denton1988b), more foreign born (Denton and Massey, Reference Denton and Massey1988; Iceland and Scopilliti, Reference Iceland and Scopilliti2008), more female-headed households (Massey et al., Reference Massey, Denton and Eggers1994), and more elderly (Farley and Frey, Reference Farley and Frey1994; Massey Reference Massey2006). Older Whites tend to be more prejudiced than younger persons (Maykovich Reference Maykovich1975; Schuman et al., 1998; St. John Reference St. John1996) and so are more resistant to residential integration.

Segregation has also been hypothesized to vary according to patterns of industrial organization, with large manufacturing sectors and high rates of unionization pushing segregation levels upward (Lieberson Reference Lieberson1980; Massey and Denton, Reference Massey and Denton1993). In contrast, metropolitan areas dominated by “creative class” service industries, such as finance, insurance, and real estate, tend to be more diverse (Florida Reference Florida2002). Another indicator of a creative class economy is the rate of patent production, which we also include in the model. Sectoral employment data was drawn from the census whereas the unionization data came from Hirsch and Mcpherson (Reference Hirsch and Macpherson2012). Patent production was computed as the number of patents per 100,000 persons in the MSA using data on utility patents from the U.S. Patent and Trademark Office.

Researchers have also shown that metropolitan areas dominated by colleges and universities and large military populations are more integrated than others (Farley and Frey, Reference Farley and Frey1994). The military is the most integrated institution in American society and its influence apparently shapes race relations and housing patterns in metropolitan areas that contain them (Moskos and Butler, Reference Moskos and Butler1996). Using census data we computed the number of persons living in military quarters per 100,000 persons. Expressions of prejudice decline sharply with education, and in order to assess the dominance of the educational sector in each metropolitan area we computed the proportion of workers employed in education.

We also consider the influence of several facets of urbanism. Since metropolitan areas are constructed from multiple counties, many of which contain significant non-urban populations, we measure the percentage of metropolitan inhabitants who are actually urban residents (living in census tracts with greater than 1,000 persons per square mile). Moreover, given that Whites, on average, continue to associate African Americans with higher rates of crime (Quillian and Pager, Reference Quillian and Pager2001, Reference Quillian and Pager2010) and resist integration based on this perception (Emerson et al., Reference Emerson, Chai and Yancey2001) we also constructed violent crime rates for MSA’s using count data from the U.S. Federal Bureau of Investigation (2012). Metropolitan areas with newer housing stocks built after the civil rights era tend to be less segregated than those areas built up in earlier periods (Farley and Frey, Reference Farley and Frey1994), we drew upon census data to compute the median year in which an MSA’s housing was built.

Finally, we control for geographic location by including dummy variables for region (with the Midwest serving as the reference category), a dummy variable for whether the MSA is located on a coast (given that coastal areas tend to have more restrictive building codes and higher housing costs (Glaeser and Gyourko, Reference Glaeser and Gyourko2008)), and a final dummy for location in a border state, where levels of Hispanic segregation have historically been higher (Grebler et al., 1970; Massey Reference Massey1979).

TRENDS IN RESIDENTIAL SEGREGATION AND SPATIAL ISOLATION

As noted above, group size strongly affects the potential for segregation and isolation experienced by minority members in metropolitan America, so we begin our analysis by showing trends in the number of Hispanics, Blacks, and Asians residing in our panel of 287 metropolitan statistical areas from 1970 through 2010. As seen in Figure 1, the greatest change in the nation’s racial-ethnic makeup over past four decades has been the remarkable growth of the Hispanic population, which surged from eight million to forty-five million persons and went from 4.7% to 16.3% of the population of these MSAs. The number of Asians also grew very rapidly, expanding by a factor of more than ten, but from a much smaller base, going from 1.5 million in 1970 to sixteen million in 2010 and raising their share of the population from 0.8% to 4.7%. Over the same period, the number African Americans roughly doubled in size, going from 17.4 to 34.2 million, but their share of the population climbed upward only slowly, going from 11.1% to 12.6% owing mainly to immigration from Africa and the Caribbean. In strictly demographic terms, then, we can say that the potential for segregation and isolation in metropolitan America increased sharply for Hispanics and to a lesser extent for Asians, but shifted little for African Americans.

Fig. 1.

Growth of Minorities in U.S. Metropolitan Areas 1970–2010

Figure 2 shows average values of residential dissimilarity with respect to non-Hispanic Whites for Blacks, Hispanics, and Asians in the 287 metropolitan areas from 1970 to 2010. The average values were computed by weighting the dissimilarity for each metropolitan area by the size of the minority population. Thus the average index for Blacks in 2010 gives the degree of neighborhood segregation experienced by the average Black inhabitant of the 287 metropolitan areas in that year. In considering tract-level residential dissimilarities, index values below 30 are conventionally considered “low;” those between 30 and 60 are viewed as “moderate;” those above 60 are seen as “high;” while any value above 75 is labeled “extreme” (Massey and Denton, Reference Massey and Denton1988a).

Fig. 2.

Trends in Neighborhood-Level Dissimilarity from Whites 1970–2010

According to these standards, over the past four decades African American segregation has fallen from extreme to merely high levels. The pace of the decline was roughly linear, with the index dropping from an average value of 78 in 1970 to reach 60, the lower boundary of the high range, in 2010, a decline of about 4.5 index points per decade. The prevailing trend in Black-White segregation is thus one of moderate but steady decline. Nonetheless, at the current rate of change, average Black-White residential dissimilarity would not reach the upper threshold of the low range for another sixty-seven years.

In contrast to the pattern of steady decline observed for Blacks, the prevailing pattern of residential dissimilarity for Hispanics and Asians is one of relative stasis. Despite the massive increase in the size of both groups, dissimilarity indices for Hispanics and Asians rose only slightly over the past four decades. Average Hispanic-White dissimilarity rose from a value of 46 in 1970 to 49 in 2010 while average Asian-White dissimilarity only rose from 39 to 41 between 1980 and 2010. Such stability in residential dissimilarity in the face of rapid population growth is remarkable. Whereas in the twentieth century the growth of Black urban populations spurred Whites to practice more intense exclusion and discrimination that yielded sharply rising Black-White dissimilarities (Lieberson Reference Lieberson1980; Massey and Denton, Reference Massey and Denton1993; Sugrue Reference Sugrue1996), the rapid growth of metropolitan Hispanic and Asian populations from 1970 to 2010 does not appear to have triggered a comparable surge in Hispanic-White or Asian-White segregation. The contrasting experience of African Americans in the early twentieth century compared with that of Hispanics and Asians today implies that rising numbers of African Americans then posed a much greater threat to Whites than rising numbers of Hispanics and Asians do today.

When it comes to spatial isolation, both substantive and mathematical effects are relevant in determining index values because the P* index is a mathematical function of the minority percentage. It necessarily increases as the minority percentage rises no matter how Whites react to a growing minority presence or how dissimilarity changes. Indeed, even if residential dissimilarity were to remain constant, the isolation index would increase if the minority percentage were to rise. For African Americans, then, the trend in spatial isolation is expected to depend on which process has more force in determining index values: the slight increase in the demographic potential for isolation attributable to the slow increase in the Black percentage over the four decades or the faster decline in the structural potential for segregation brought about the slow but steady decrease in residential dissimilarity.

As Figure 3 shows, the effect of the steadily decline in Black dissimilarity appears to outweigh the modest increase in the relative size of the urban Black population in determining Black isolation. As before, we computed size-weighted isolation indices for Blacks, Hispanics, and Asians from 1970 through 2010. The degree of spatial isolation experienced by African Americans decreased linearly at roughly the same rate as the decline in dissimilarity, falling by an average of 4.8 points per decade and dropping from a value of 65 in 1970 to 46 in 2010. In contrast, Hispanic isolation increased quite sharply, as we would expect given the large increase in group size combined with slowly rising dissimilarity. The Hispanic isolation index rose from 27 to 47 over the period, an increase of around 5 points per decade. We also observe an increase in Asian spatial isolation, which went from an index value of 10 in 1980 to reach 21 in 2010. Given their small share of most metropolitan populations, however, the pace of the increase was only about 3.7 points per decade for Asians, and even the value of 21 in 2010 indicates a relatively low level of Asian spatial isolation.

Fig. 3.

Trends in Neighborhood-Level Isolation Indices 1970–2010

In sum, Black dissimilarity and isolation indices have both steadily declined over the past four decades at a rate of around 5 points per decade and the average level of Black-White dissimilarity now lies at the frontier between “high” and “moderate” segregation with an index value of 60. In contrast, dissimilarity and isolation have both increased for Hispanics and Asians. Although the increase in dissimilarity has been slight and in 2010 both remained well within the “moderate” range of segregation, increases in the size of the Asian and Hispanic populations have pushed isolation indices upward. Although the rate of Asian growth has been sharp, the relative size of the population has remained small in most metropolitan areas and average isolation remains quite low with an index value of 20. Among Hispanics, however, rapid rates of urban population growth has combined with very large absolute numbers have increased spatial isolation to levels comparable to those of African Americans. As of 2010, the average isolation index stood at roughly 46 for both groups.

Despite the impressive shifts toward integration by African Americans overall, the declines have been quite uneven across metropolitan areas. In general, levels of segregation and isolation are much higher and the declines considerably slower among large metropolitan areas with large Black populations. Indeed, in their analysis of residential segregation in the fifty largest metropolitan areas of 1980, Massey and Denton (Reference Massey and Denton1989, Reference Massey and Denton1993) identified sixteen areas in which African Americans experienced a uniquely intense form of segregation across multiple geographic dimensions, a condition they labeled “hypersegregation.” Neither Hispanics nor Asians experienced hypersegregation in the metropolitan areas they examined. Wilkes and Iceland (Reference Wilkes and Iceland2004) updated the analysis using 2000 census data for all metropolitan areas and identified twenty-nine metropolitan areas in which African Americans were hypersegregated. Although Asians still did not experience hypersegregation in any metropolitan area, by 2000 hypersegregation for Hispanics had emerged in the Los Angeles and New York metropolitan areas.

In Figure 4 we consider trends in residential segregation and spatial isolation for Hispanics and Blacks in those areas identified as hypersegregated in 2000. These metropolitan areas contain roughly a third of all African Americans in the United States and nearly half of those who live in metropolitan areas, and they account for 16% of all Latinos and close to 20% of metropolitan Latinos. Obviously in these metropolitan areas that house a disproportionate share of Blacks and Hispanics levels of segregation and isolation are considerably higher and downward trends are far more limited than among metropolitan areas in general. Among hypersegregated Black areas dissimilarity falls from 81 to 67 and isolation drops from 74 to 59 for a decline of only 3.5–3.7 points per year, leaving Black-White dissimilarity well within the high range in 2010 and isolation at a very elevated level. In the two hypersegregated Hispanic areas, moreover, levels of dissimilarity and isolation rose such that both came to roughly equal those observed in Black hypersegregated areas.

Fig. 4.

Trends in Segregation and Isolation within Hypersegregated Metropolitan Areas

Finally, we consider what neighborhood circumstances look like from the viewpoint of non-Hispanic Whites. To this point our analyses of neighborhood composition have relied strictly on the P* isolation index, but P* actually refers to a family of isolation and contact indices. The isolation index gives the average likelihood of within-neighborhood contact with members of one’s own group whereas contact indices give the likelihood of within-neighborhood contact with members of other groups. Within any metropolitan area, the isolation and contact indices for any given group must sum up to 100. Figure 5 examines how the world has changed for Whites by presenting weighted averages computed to measure the degree of own-group isolation and intergroup contact for Whites residing in neighborhoods of the 287 metropolitan areas between 1970 and 2010.

Fig. 5.

Trends in White Isolation and Contact Indices

These figures underscore the asymmetric nature residential experiences for minority and majority group members in metropolitan areas, for despite the steady declines in Black segregation, falling Black isolation, the massive increase of Hispanics, and the large increase of Asians within metropolitan areas, Whites still inhabit overwhelmingly White neighborhoods. Although the White isolation index dropped from a value of 91 in 1970 to 74 in 2010—some 4.4 points per decade—the starting point constituted a very extreme level of isolation. Moreover, despite the shift in White isolation, the average White American still lived in a neighborhood that was roughly three-quarters White in 2010, a time when the United States as a whole was only 63% White and that share was rapidly falling.

Mixing within neighborhoods likewise has not kept pace with the changing racial-ethnic composition of the nation for the shift away from White isolation involved relatively small changes in the degree of contact with minority group members. In the four decades from 1970 to 2010, for example, the likelihood of White neighborhood contact with African Americans rose by less than 1 point per decade to yield an index value of 8 in 2010; contact with Asians grew by just 1.4 point per decade to reach a value of 6 in 2010; and contact with Hispanics rose by only 2 points per decade to achieve a value of 11. It is only through the combined effect of rising contact with Hispanics, Blacks, and Asians together that White isolation fell to the extent observed. Considered on a group-by-group basis, contact probabilities with minority groups remain minuscule. In the end, the vast majority of Whites do not experience the rising racial-ethnic diversity of contemporary America.

EXPLAINING CONTEMPORARY SEGREGATION AND ISOLATION

In order to isolate the principal drivers of segregation and integration we estimated the effect of the independent variables described in Table 1 on levels of residential dissimilarity and spatial isolation in 2010. We also assess the determinants of change from 1980 to 2010 by regressing changes in dissimilarity and isolation scores on changes in the values of the various independent variables over the period. Rather than estimating separate models for African Americans, Hispanics, and Asians, for economy of presentation we concatenate data for the three groups into a single file and include dummy variables to indicate which group’s segregation is being measured as the dependent variable. All independent variables pertain to the metropolitan area population as a whole except for the indices of racism and illegal immigration, which are specific to Blacks and Hispanics, respectively, and the ratios of minority/White income and minority/White college graduates which are defined with respect to Blacks, Hispanics, and Asians when regressed on dissimilarities defined for the same three groups.

As of 2010, the group dummy variables in the first panel indicate that if other factors in the model were equal, Hispanics would exhibit a residential dissimilarity from Whites that was 20.3 points below that of Asians (p < 0.001) whereas African Americans would display a dissimilarity 5.5 points lower (n.s.). In fact, Asians are far less segregated from Whites than either Blacks or Hispanics, which implies that other things are decidedly not equal. In the second panel of Table 2 we interact these indices of anti-Black and anti-Latino sentiment with the dummy variables for membership in each group. Results indicate that anti-Black racism has a powerful and highly significant effect in predicting the level of Black-White segregation. Each point increase in the anti-Black index raises the expected level of Black-White dissimilarity by 2.09 points, potentially raising the dissimilarity index by 15 points over the observed range of the scale. Among Latinos, the effect is even more pronounced. Each point increase in the anti-Latino index raises Hispanic-White dissimilarity by 3.4 points, potentially raising Hispanic dissimilarity by 23 points over the observed range of the index.

Table 2.

Effect of Selected Variables on Minority-White Residential Dissimilarity in 2010 and 1980–2010 Change

*** p < .001 **p < .01 *p < .05 + p < .10 (Note: Number of MSAs = 275)

As expected, the density instrument also strongly predicts the degree of racial-ethnic segregation across metropolitan areas. In this case, a higher density score indicates a more permissive zoning regime, allowing a higher average level of residential density. As can be seen, higher allowable densities are associated with significantly lower levels of residential dissimilarity. Moving the index from its observed minimum to its observed maxim would reduce residential dissimilarity by 16 points.

Also as expected, residential dissimilarity is predicted by the relative size of the minority population. Each point increase in the percentage of Blacks raises dissimilarity by 0.235 points, yielding a potential increase of 12 points as the percentage goes from its observed minimum to maximum. Likewise, each point increase in the percentage of Hispanics raises dissimilarity by 0.163 points, yielding a potential shift of 15 points from its minimum to maximum. Although the coefficient is only marginally significant (p < 0.10), each point increase in the percentage of Asians raises dissimilarity by 0.301 points, yielding a potential shift of 21 points over the observed range of percentages.

Aside from their relative numbers, the relative income of each minority group strongly predicts the residential dissimilarity they experience. Each point increase in the ratio of minority to White household income lowers residential dissimilarity by 6.65 points. Among African Americans the income ratio varies from 0.17 to 1.14, yielding a potential drop in dissimilarity of 6.5 points over the range of the index. Among Hispanics the range is from 0.31 to 1.43, yielding a potential drop of 7.5 points; but among Asians the range goes from 0.35 to 2.81 yielding a potential decline of 16.4 points. Thus, one important reason for the lower level of Asian-White dissimilarity is the relatively high incomes earned by Asians, which in most metropolitan areas top those of Whites. The ratio of college-educated minority members to college educated Whites works in the same direction—lowering levels of residential dissimilarity—but the effect is much weaker and only marginally significant. Racial-ethnic segregation, however, is quite strongly and positively associated with the degree of class segregation. The residential dissimilarity between the affluent and the poor is potentially responsible for an upward shift of 6 points in the level of racial-ethnic dissimilarity as it goes from its observed minimum to maximum.

As other researchers have found, we also find that racial-ethnic segregation is greater in large metropolitan areas and lower in areas with a sizeable military presence and in those with a newer housing stock. Each point increase in the log of metropolitan population increases residential dissimilarity by 3.47 points potentially accounting for a shift of 17 points from the smallest to the largest metropolitan area. However, each point increase in the log of the military population per 100,000 persons reduces racial-ethnic dissimilarity by 0.503 points whereas each year increase in the median year of housing lowers it by 0.271 points, yielding potential reductions of 4 points and 12 points, respectively, as these variables range from minimum to maximum. A few of our other substantive predictors—percent foreign born, percent urban, and the violent crime rate—also predict racial-ethnic segregation but the effects are weak and usually only of marginal significance. Among geographic effects, we see that levels or racial and ethnic segregation are systematically lower in the West (by 4.412 points) compared with the Midwest and systematically higher in border states (by 2.617 points) than elsewhere in the country.

The right-hand columns show a model predicting change in racial-ethnic dissimilarity between 1980 and 2010. The dummy variables indicating group membership, the prejudice indices, the density instrument, and the geographic indicators are all time-invariant but all other variables are defined in terms of their change between 1980 and 2010. As we would expect given trends already reported, the coefficient for Hispanics is positive and significant (7.828), the coefficient for Blacks is negative and significant, (-21.335) indicating that, other things equal, Hispanic segregation increased over the three decades while Black segregation decreased. In both cases, however, shifts toward integration were impeded by higher levels of anti-Black or anti-Latino sentiment, and across metropolitan areas by more restrictive regimes of density zoning. Holding constant anti-Latino sentiment, however, an increase in the percentage Hispanic was associated with a lower level of residential dissimilarity. Rising minority income relative to Whites was strongly associated with declining residential dissimilarity, as was rising education relative to Whites, though as in the cross-sectional model the latter effect was much weaker an only marginally significant. Not surprisingly, immigration, as measured by a rising share of foreign born, was strongly associated with rising residential dissimilarity as was the metropolitan rate of unionization.

Table 3 presents regressions to predict the degree of spatial isolation in 2010 as well as changes in isolation between 1980 and 2010, conditional on the observed level of residential dissimilarity. The degree of spatial isolation experienced by a group largely follows from its relative size and the degree of residential dissimilarity it experiences. It is no surprise, therefore to discover that residential dissimilarity and minority percentages very powerfully predict the level of isolation in 2010 as well as changes from 1980 to 2010. These findings are not substantively interesting since they follow directly from the definition and properties of the P* index. What is of interest here is the other factors that contribute to spatial isolation above and beyond these mathematical determinants.

Table 3.

Effect of Selected Variables on Spatial Isolation in 2010 and 1980–2010 Change

*** p < .001 **p < .01 *p < .05 + p < .10 (Note: Number of MSAs = 275)

As can be seen, the principal determinant of spatial isolation beyond dissimilarity and minority composition is the extent of racial prejudice, which acts to intensify isolation in the cross section and forestall reductions over time. Each point increase in anti-Black sentiment increases the spatial isolation of African Americans by 6.049 points and is associated with an increase of 1.103 points in isolation over time, yielding potential shifts of 42 points in the degree of isolation and 8 points of change in isolation over the observed range of anti-Black prejudice. Likewise, each point increase in anti-Latino sentiment is associated with an increase of 9.084 points in the degree of Hispanic spatial isolation and a 2.152 point increase in isolation over time, yielding potential shifts of 61 points of isolation and 14 points of change in isolation over the observed range of the prejudice index. Beyond these effects, not much else affects the level of spatial isolation. The only strong effect is the effect of a higher relative level of minority education, which promotes a decline in racial-ethnic isolation over time. The zoning instrument has no effect, implying that its influence on spatial isolation occurs entirely through its strong effect on residential dissimilarity.