Residential Integration on the New Frontier: Immigrant Segregation in Established and New Destinations

In 2010, 13.6 % of Americans were foreign-born, and another 11.2 % had at least one parent who was born abroad.

The “group threat” literature almost exclusively focuses on the stock of ethnic/racial groups (e.g., Hood and Morris 1997; McDermott 2011; Oliver and Wong 2003; Quillian 1996; Rocha and Espino 2009; Taylor 1998) rather than recent changes in their populations posited here to influence perceptions and migration behavior (but see Hopkins 2010, 2011).

This analysis is not possible with recent American Community Survey data. Although dissimilarity scores for each of these groups can be estimated, most of the group-specific characteristics used in the analysis are unavailable. Table S2 in Online Resource 1 shows dissimilarity scores, by destination type, for the 10 groups from in 2005–2009, generally showing the same patterns as seen in 2000, although the levels are in some cases modestly different. I estimated models that regress 2000–2009 change in dissimilarity on 2000 group-level and metropolitan-level variables. These models yielded substantively similar results, showing that net of group and metropolitan characteristics, immigrant dissimilarity from native whites increased significantly more rapidly in new (b = 1.25, SE = .49) and minor (b = 2.35, SE = .50) destinations than in established areas.

In addition to including a relatively small (albeit growing) share of U.S. immigrants, smaller metropolitan and nonmetropolitan areas have too few tracts to capture residential spatial patterns. Lichter et al. (2010) bypassed this issue by using block data, a level of geographic detail not possible here because of suppression of place-of-birth data at that more-refined level. Research has shown, however, that although levels of segregation tend to be higher at lower geographic levels, metropolitan estimates based on tracts, block groups, and blocks are highly correlated (Iceland and Steinmetz 2003).

These groups are not necessarily representative of all American immigrants, but they are a broad cross-section of the foreign-born population, constituting 56 % of all immigrants in 2000 and making up 77 % of all foreign-born growth between 1990 and 2000.

In supplemental analysis, I tested several alternative destination-type operationalizations that either increased or decreased the stringency of being an established or new destination, including group-size restrictions and relaxing/increasing the extent to which population shares or growth exceed metropolitan averages. Although the size of the destination-type effects varies slightly depending on the approach used, the general interpretation is consistent and the magnitude of the coefficients shown here approximates the midpoint of all considered specifications.

Census 2000 does not tabulate income or housing tenure by country of birth. To circumvent this issue, I draw on metropolitan-level ethnic-group data (from Summary File 4). These data are imperfect because the subpopulations represent those identifying with specific ethnic groups on race, Hispanic origin, or (for Jamaicans and Haitians) ancestry questions, and thus include both the U.S.-born and foreign-born. What is essential is that these variables capture the variability in demographic, economic, and acculturation characteristics across immigrant groups and metropolitan areas. Correlation analyses suggest that they do: at the national level, the ethnic-origin data used here is strongly related to estimates specific only to immigrant-group members (white income parity, r = .99; homeowners, r = .98). Nevertheless, the percentage of each ethnic group that is foreign-born is included as an additional control.

Supplemental models explored the use of a destination typology based on the total immigrant population (i.e., “place-based”) rather than the group-based one employed here. When included alone or alongside the group-based typology, there were no significant differences between place-based established and non-established destinations, and this variable’s inclusion does not alter the statistical or substantive interpretation of the group-based destination-type coefficients. The interactions between the group-placed and place-based dummy variables do, however, indicate a somewhat heightened impact of being a new group in a non-established place, but these models also exhibit major signs of collinearity and instability, and are thus not shown. Complete results are available on request.

In additional analyses, I considered occupational concentration in four other sectors of labor markets: health, sales, construction, and government. The coefficients on each of these in the full model were small and nonsignificant.

Although these measures can take values anywhere between 0 and 100, their truncated range makes a linear model technically inappropriate. However, residual plots reveals no major violations of regression assumptions due to truncation, and skewness/kurtosis statistics suggest that the D values approximate a normal distribution (s = 0.31, k = 2.78).

Fixed-effects models that account for metropolitan characteristics that do not vary across groups produce results that are substantively similar to those presented here. The coefficients for new destinations (b = 2.44; SE = 0.93) and minor destinations (b = 4.77; SE = 0.95) under the fixed-effects approach are nearly identical to those shown in Table 3.

The substantive interpretation of the unweighted and weighted models is similar. In the full regression model with weights, the new destination coefficient is smaller than that in the unweighted models but remains positive and significant (b = 1.57; SE = 0.78); the coefficient for minor destinations is larger in the weighted results (b = 5.87; SE = 0.97) than the unweighted ones.

Modifying recent arrivals to include immigrants who entered the country between 1990 and 1994 does not change the interpretation of the results (arrived 1990–2000, Model 2: b = 0.14, SE = 0.05; Model 3: b = 0.09, SE = 0.04).

Substituting a measure of a group’s family does not alter the interpretation of the results (family income, in $1,000s, Model 2: b = 0.04, SE = 0.06; Model 3: b = 0.02, SE = 0.04).

When entered separately, both coefficients are significantly negative (percentage immigrant b = –0.21; SE = .04; top 5 gateway b = –1.69; SE = 1.29). Importantly, their inclusion does not alter the results of the group-specific destination-types in any meaningful way. With both excluded from the analysis, segregation in new (b = 2.53; SE = 1.05) and minor (b = 5.21; SE = 1.01) destinations remains significantly higher than in established areas.

Setting Mexicans as the referent in these models indicates that their adjusted level of segregation from native whites is significantly lower than all groups except Koreans (b = 3.12; SE = 2.82) and Chinese (b = 4.67; SE = 2.64).

Results generated partially from a forward-stepwise approach lead to similar conclusions.

Differences in the effects of destination type (both new and minor destinations) are significant only between Chinese immigrants and all other groups.

In supplemental analyses, I explored models that pool Jamaicans and Haitians and estimate the reduced set of variables in Table 4 on dissimilarity from native whites. The results show that these groups are no more nor less segregated from native whites in new destinations (b = .82, SE = 3.12) yet more segregated in minor areas, although not significantly (b = 1.78, SE = 2.63), than in established areas. However, the small N of 53, even when these groups are pooled, may contribute to model instability; thus, I do not present these estimates.

Supplemental models for Koreans reveal the importance of metropolitan black and retirement populations, both of which significantly increase segregation.