The analyses rely on data from the 2001 British and 2000 U.S. censuses that contain the size and distribution of different ethnic groups needed to calculate residential segregation indexes. Residential segregation typically describes the distribution of different groups across smaller areal units within a larger area, such as a metropolitan area (Massey & Denton, 1988). Thus, to measure residential segregation, one usually has to define both the appropriate larger area and its component parts, or neighborhoods. The most common larger geographic unit chosen is in fact the metropolitan area, which is a reasonable approximation of a housing and job market. Using housing and job markets for the computation of segregation indexes is based on the notion that a person (or household) who works in a given commutable area can potentially choose to live in any community within the housing market. In the United States, metropolitan areas were designed to represent housing and labor markets and generally contain at least 50,000 people. For the 2000 U.S. census data, we use 1999 county-based Metropolitan Statistical Area (MSA)/Primary Statistical Area definitions, which yield 318 metropolitan areas in total.
For Great Britain, the analyses are based on 2007 Travel to Work Areas (TTWA) (Bond & Coombes, 2007), an official geographical unit produced by the Office for National Statistics (ONS) based on work journeys between Census Super Output Areas, that divide the entire territory of Great Britain into 232 TTWAs. This choice differs from those of many previous segregation studies in Britain. Previous studies typically used local government administrative boundaries (Local Authorities or Districts) that do not necessarily resemble the housing and job market areas within which residential decisions are made, as they often detach many metropolitan urban cores from their commuter hinterland. An alternative geography would have been the definition of Urban Areas produced by the ONS for the 2001 Census (Bibby & Shepherd, 2004; ONS, 2004). However, these Urban Areas comprise contiguous built-up areas with a high population density, and they therefore purposely exclude adjacent rural areas in the urban fringe and within the metropolitan commuter belt. However, it is precisely in these fringe areas where contemporary residential segregation processes are likely most acute in Great Britain, through the suburbanization of affluent groups into sparse rural commuter belts where urban expansion is largely forbidden by strict planning regulations. This phenomenon is commonly ignored by segregation studies in Great Britain that repeatedly concentrate on urban cores and administrative areas.
Because of this, we decided to use TTWAs as the only available standard geography that is closest in definition to U.S. metropolitan areas—functional regions. However, there is an important difference between the two. U.S. metropolitan areas are defined above a minimum population threshold that excludes half of the counties and approximately 20% of the population, while TTWAs are drawn without any population thresholds and cover the whole territory. Hence, although in the United States some rural areas in the urban fringe are included within the constituent metropolitan counties, TTWAs contain a much higher proportion of rural areas. To account for the difference in the proportion of “rurality” between TTWAs and MSAs in the United States, we could have reengineered TTWAs through applying population and distance thresholds to create an exact match to MSAs, but we discarded this option since we would end up with nonstandard regions that will make future comparisons difficult. We finally decided to eliminate from the analysis those TTWAs with a total population below 50,000 people (the U.S. minimum threshold for a metropolitan area), resulting in 179 TTWAs for the analyses below. The results of the paper do not significantly differ with or without these small areas. For ease of interpretation, in both countries we refer to these housing market units as “metropolitan areas.”
The next geographic consideration is the basic unit of analysis—the neighborhood. In U.S. studies of segregation, the “census tract” is the most common unit chosen (e.g., Logan, Stults, & Farley, 2004; Iceland, Weinberg, & Steinmetz, 2002; Massey & Denton, 1993). Census tracts typically have between 2,000 and 8,000 people, with an average size of about 4,000. The comparable unit used with the British data are Middle-Level Super Output Areas (MSOAs: Mean = 7,200 people) in England and Wales, and Intermediate Geography in Scotland (Mean = 4,000 people), both referred to as MSOAs in this paper (overall Mean = 6,600 people) (ONS, 2010). These geographical units are an aggregation of the Census smallest areas (Output Areas) for the collection and dissemination of government statistics using units of similar population size across the country compared to, for example, electoral wards. We used MSOAs as the most similar unit to U.S. census tracts in terms of population size and much smaller variance than electoral wards (with a large variance around Mean = 5,500 people).
To test the comparability and sensitivity of using different neighborhood definitions, we also examined block group areas in the United States, which have an average of 1,000 people, and Lower-Level Super Output Areas (LSOAs: mean = 1,500 people) in Great Britain. Smaller areas tend to be more homogeneous, and thus display higher levels of ethnic segregation (Voas & Williamson, 2000). Overall, as is shown in the results section, segregation scores for most groups are considerably higher in the United States than in Great Britain, and would remain higher even if we were to use census tracts (the larger unit) in the United States and LSOAs (the smaller unit) in Great Britain.2 Since data on detailed groups in the United States are much more limited at the block group level, the analyses focus on patterns using census tracts and the nearest unit in Great Britain—the MSOA.
This study compares the residential segregation of ethnic groups present in both Great Britain and the United States, and examines the role of nativity in explaining patterns. For this kind of analysis, one would ideally like to have data available on the residential patterns of the foreign- and native-born for a wide variety of specific ethnic groups, such as Pakistanis, Indians, and so forth (Mateos, Singleton, and Longley, 2009). These detailed data are partially available in the British Census (for the largest ethnic groups the ethnicity question can be cross-tabulated by the country of birth question), but not in the United States; the race question focuses on physical appearance (not ethnocultural groups), and the quality of the data from the ancestry question is uneven.3 This means that with the U.S. data, one can obtain counts of foreign-born Pakistanis (from the place of birth question), but not necessarily reliable counts of U.S.-born people of Pakistani origin.
This limitation is addressed in this study with the following strategy. First, for both the United States and Great Britain one can calculate reliable numbers of native- and foreign-born people of panethnic groups. In the U.S. context, common panethnic groups used in segregation studies include non-Hispanic whites, blacks, and Asians (e.g., Logan et al., 2004) (we do not include Hispanics in this study because of the small number of Hispanics in Great Britain). The British census data contain counts of these groups by nativity as well. Thus, we examine three panethnic groups that are present in large numbers in both countries4—non-Hispanic whites/white British, blacks, and Asians—and the role that nativity plays among them.
Since “race” and “ethnicity” are socially constructed concepts, a challenge confronting this kind of cross-national comparison is that these ethnic categories do not have precisely the same meaning in Great Britain and the United States. In addition, the composition of the panethnic groups varies in the two countries. For example, within the Asian panethnic group there are more South Asians in Great Britain and more East Asians in the United States. These groups likely differ in their settlement patterns in each of the countries. To address these issues, we examine the residential patterns of specific foreign-born groups, such as Bangladeshis, Jamaicans, Indians, Germans, and those born in Hong Kong. This provides insight on how residential patterns differ across specific national groups, as well as within the panethnic groups defined above.
Segregation indexes are calculated for a given group in a metropolitan area only if there are at least 1,000 group members living there, as segregation indexes for metropolitan areas with small minority populations are less reliable than those with larger ones.5 In order to make meaningful comparisons of particular groups across the United States and the United Kingdom, we narrowed the set of countries to those present in at least 10 metropolitan areas in both the United States and the United Kingdom that met the criterion of a minimum population of 1,000 group members per area. This procedure results in the inclusion of eight countries of birth located in various world regions: Bangladesh, France, Germany, Hong Kong, India, Italy, Jamaica, and Pakistan. Our conclusions about the variation in segregation patterns within panethnic groups and across the United State and Great Britain do not differ if we adjust the selection criteria, such as by allowing a greater number of countries of origin to be included in the analyses (these results are available upon request).
While these countries are located in a variety of continents, they provide information on only a slice of the foreign-born populations in the United States and Great Britain. Thus, the analyses below also include information on the foreign-born by world region. As above, indexes are calculated in metropolitan areas where at least 1,000 group members were present in at least 10 metropolitan areas in both the United States and Great Britain. The world regions included are: sub-Saharan Africa, Middle East, Eastern Europe, South Asia, Far East, Southern Europe, North Western Europe, North America, and the Caribbean. Table 1 contains descriptive statistics about how many people are in each group and how many metropolitan areas for each group are in the analyses.
Table 1. Descriptive Statistics for the United States and Great Britain, 2000–2001
| ||Number of U.S. Metro Areas||Number of G.B. Metro Areas||U.S. Population||G.B. Population|
|Native non-Hispanic white/white British||318||179||142,779,553||49,347,979|
| Native-born||181|| 82||2,938,434||1,320,813|
| Foreign-born||241|| 90||6,719,237||1,447,004|
| Native-born||293|| 59||27,546,288||816,113|
| Foreign-born||118|| 42||2,003,274||593,044|
|Country of Birth among the Foreign-Born|
|Bangladesh|| 13|| 16||72,431||133,022|
|France|| 31|| 15||96,174||61,525|
|Hong Kong|| 28|| 20||171,350||55,873|
|Italy|| 58|| 22||412,041||73,338|
|Jamaica|| 43|| 14||509,452||129,428|
|Pakistan|| 34|| 38||179,381||316,032|
|World Region of Birth among the Foreign-Born|
|Caribbean|| 96|| 26||2,858,535||232,567|
|Eastern Europe||136|| 37||1,787,861||154,609|
|Far East||228|| 70||5,512,511||349,830|
|Middle East & North Africa||101|| 35||1,097,686||247,914|
|North America||122|| 44||654,464||173,608|
|Northern & Western Europe||201||149||1,808,551||1,044,595|
|South Asia||120|| 77||1,304,909||1,006,186|
|Southern Europe|| 94|| 47||857,439||263,695|
|Sub-Saharan Africa|| 72|| 73||559,539||717,542|
|Total metropolitan population in census year||318||179||281,421,906||57,075,654|
Measures of Segregation
The analyses use two common segregation measures, the information theory index (or Theil's H) and the isolation (Pxx) index. The information theory index is a measure of evenness. Evenness refers to the differential distribution of the subject population across neighborhoods in a metropolitan area. More specifically, H is the weighted average deviation of each unit's “entropy” (or diversity) from the metropolitan-wide entropy, expressed as a fraction of the metropolitan area's total entropy. The analysis includes both dual-group and multigroup versions of the information theory index (Iceland, 2004; Reardon & Firebaugh, 2002). The former compares the segregation of one particular group from another, while the latter measures the joint distribution of several groups simultaneously. Additional analyses were run with another measure of evenness, dissimilarity (D), but results were for the most part similar to results with H, and we thus include only H (rather than dissimilarity) because of the advantageous attributes of H (see Reardon & Firebaugh, 2002). Of particular interest here is the ability to calculate the multigroup version of H, which allows one to look at the joint distribution of several groups simultaneously. In this way, one does not have to rely only on two-group measures, which typically involve picking a specific reference group (such as “whites”). As such, H has been recently used with increasing frequency in segregation studies (e.g. Farrell, 2008; Fischer, 2008; Lee et al., 2008).
The information theory index is specifically calculated as follows. Since H is the weighted average deviation of each unit's entropy (or diversity) from the metropolitan-wide entropy, one first calculates each metropolitan area's entropy score as
where Πr refers to a particular racial/ethnic group's proportion of the total metropolitan area population. All logarithmic calculations use the natural log. The higher the number, the more diverse an area is. The maximum level of entropy is given by the natural log of the number of groups used in the calculations. The multigroup H used in this study calculates the segregation of three panethnic groups (described above) from each other: whites, Asians, and blacks. In order to include the entire U.S. and British population in these segregation calculations (i.e., in order to have a set of mutually exclusive and exhaustive groups, as is common in calculations of multigroup H), we add a fourth group that contains the residual population (termed “other”).6 With four racial/ethnic groups, the maximum entropy is log 4 or 1.39.
A unit within the metropolitan area, such as a census tract, would analogously have its entropy score, or diversity, defined as
where Πri refers to a particular racial/ethnic group's proportion of the population in tract i. The information theory index is the weighted average deviation of each unit's entropy from the metropolitan-wide entropy, expressed as a fraction of the metropolitan area's total entropy:
where ti refers to the total population of tract i, T is the is the metropolitan area population, n is the number of neighborhoods, and Ei and E represent neighborhood i's diversity (entropy) and metropolitan area diversity, respectively. The information theory index varies between 0, when all areas have the same composition as the entire metropolitan area (i.e., maximum integration), to a high of 1, when all areas contain one group only (maximum segregation).
The second index used in the analysis, isolation, is the most widely used measure of “exposure” (one of the dimensions of segregation defined by Massey and Denton (1988)). The isolation index indicates the average percentage of group members (of the group of interest) in the neighborhood where the typical group member lives. The index varies from 0 to 1, with 1 indicating the highest level of segregation. A black isolation score of 0.60 in a particular metropolitan area, for example, would indicate that the typical African American lives in a neighborhood that is 60% African American. The formula for isolation is as follows:
where Pxx is the usual notation for the isolation index, xi is the population of the minority group of interest in neighborhood i, X is that group's population in the metropolitan area as a whole, and ti refers to the sum of the minority and reference group populations in neighborhood i.
When comparing the two indexes (H and Pxx), the information theory index has the advantage of not being sensitive to the relative size of the groups in question. It merely provides information on how evenly group members are distributed across neighborhoods. In contrast, the isolation index is sensitive to the relative size of the groups being studied. Other factors being equal, larger ethnic groups will be more isolated than smaller ones simply because there are more coethnics present with which to share neighborhoods. For example, isolation is generally higher for blacks in the U.S. metropolitan areas with many blacks (e.g., Detroit) rather than ones with fewer blacks (e.g., Salt Lake City). This is not necessarily a negative feature of the index; from a sociological point of view, it can be useful to know to what extent a person of a particular ethnic group lives with coethnics, and hence in an ethnic community. There are some cases, for example, where a group is fairly evenly distributed across neighborhoods (where H would be low), but if that group comprises a large proportion of the overall metropolitan population, its isolation (Pxx) may be relatively high—indicating that the group lives in neighborhoods with mostly coethnics. The reason we use two measures in this study (H and Pxx) is to capture these distinct and meaningful dimensions of segregation (Massey & Denton, 1988; Massey, White, & Phua, 1997; White, 1986). Nevertheless, the different features of the two indexes need to be kept in mind when interpreting cross-group differences in segregation, as we do below.
A final methodological issue that arises when implementing dual-group indexes revolves around whom to use as the “reference group” in dual-group segregation calculations. In the United States, studies have typically measured the segregation of various minority groups from non-Hispanic whites, and in Great Britain vis-à-vis white British.7 Specifically, we use the native-born of these groups in our study, sometimes termed “white” or “ethnic majority” for short. A sensitivity analysis was also conducted where we calculated the segregation of different ethnic groups from a reference group consisting of all people who are not of the ethnic group in question; these scores were similar, if a little lower, and highly correlated to those calculated with the white majority as the reference group, so we present results only with native-born whites as the reference group.
To answer the first two research questions posed in the introduction: (1) how do levels of ethnic segregation compare in the United States and Great Britain? and (2) to what extent do these patterns vary by the ethnic group being considered?, the analysis begins with a descriptive examination of segregation scores (information theory and isolation) by country (Great Britain and United States), ethnicity, nativity, and, among the foreign-born, by country and world region of birth. The prime interest here is to examine the extent to which segregation differs in the two countries, and among which ethnic and immigrant groups differences appear greater.
The subsequent regressions then provide more detailed analyses to answer our third and fourth questions: (3) what is the role of nativity in explaining these patterns? and (4) does the role of nativity differ in the United States and Great Britain, and is this comparison affected by the ethnic group being considered? We run regressions using metropolitan area information theory index and the isolation index scores as dependent variables.
Model 1 of each of the regressions includes two basic independent variables. First is a dummy variable indicating if the metropolitan area observation is in the United States or Great Britain. The coefficient of this variable will indicate whether metropolitan area segregation indexes are generally higher in Great Britain or the United States (the omitted category). The second independent variable measures nativity, which is at the heart of research question 3 above. Specifically, for each metropolitan area, a segregation index is computed for the native-born of the panethnic group in question, as well as for the foreign-born of that group. Thus, each of the regression models actually contain two observations per metropolitan area (one representing the index of native-born and one representing the index of the foreign-born of the panethnic group being considered). For example, in the regressions that examine the effect of nativity on Asian-white segregation, we include two observations per metropolitan area: one for the segregation of foreign-born Asians from whites and the second representing the segregation of native-born Asians from whites. A dummy variable is then included to indicate if the segregation index represents that of foreign-born Asians (with the native-born Asian index as the omitted category). With this strategy, we can directly test whether segregation indexes are higher among foreign-born Asians than native-born ones in the metropolitan areas in our analyses. We follow the same strategy when looking at the segregation of blacks. Because in these regression models the same metropolitan areas are included up to two times in the models, we produce corrected standard errors by using Generalized Linear Regression models that account for the correlated error structure (i.e., because we are using repeated, clustered observations by metropolitan area) among the independent variables.8 This modeling strategy is similar to that employed in some other studies of ethnic and/or nativity variation in segregation (Iceland & Nelson, 2008; Massey & Denton, 1989).
Model 2 in the regressions answers our fourth research question, whether the effect of nativity varies by country (United States vs. Great Britain) and by panethnic group being considered. Specifically, this model adds an interaction term between country (Great Britain) and nativity, which tells us if the effect of nativity varies by country. The regressions with the H index have one set of models that examine the segregation of multiple groups from each other (whites, Asians, blacks, and “others,” as described above). However, in order to examine whether the effect of nativity varies by panethnic group (and country), we also run separate models for Asian-white and black-white segregation (using both Pxx and H). This will allow us to see if the interaction between country and nativity applies, for example, to Asian-white segregation but not to black-white segregation. Interpreting results from these separate models is easier than including three-way interaction terms between panethnic group, nativity, and host country. Model 2 in each set of regressions also includes two basic control variables—for metropolitan area population size and population size of the group in question in a given metropolitan area.
In terms of how this analytical strategy speaks to the theoretical perspectives (assimilation, segmented assimilation, and ethnic disadvantage), if the coefficient for foreign-born dummy variable is positive and significant, this indicates that segregation scores for the foreign-born are generally higher than those for the native-born. This finding would be consistent with the predictions of spatial assimilation, which predicts lower segregation across generations. If nativity matters for some groups (e.g., blacks) but not others (e.g., Asians), this provides some support for segmented assimilation—that groups are experiencing different kinds of assimilation trajectories. Finally, if nativity does not matter for any panethnic group in any country, this provides some support for the ethnic disadvantage approach. That is, over time and across generations, we may not see decreasing ethnic residential segregation. Finally, the country * nativity interaction describes the extent to which assimilation for a given group might apply more in one country than the other (i.e., Great Britain or the United States).