Traditionalky white men dating black women

Data and Analysis

We use 2000 U.S. census confidential data to perform our analyses. Although publicly released data offer information about the location of mixed-race couples down to the scale of the PUMA (Public Use Microdata Sample Area, an area of about 100,000 people), confidential census data provide information about the location of such couples by census tract. This level of geographic detail requires that research be carried out in secure facilities, and our results were screened by Census Bureau employees to maintain confidentiality. We examine residential patterns of our sampled couples in 12 large metropolitan areas: Atlanta, Chicago, Dallas, Detroit, Houston, Los Angeles, Miami, New York, Philadelphia, San Diego, San Francisco, and Washington, DC.² This group comprises 11 of the 12 most populous metropolitan areas in the United States.³ These places range in racial composition and rates of mixing, capturing variations in racial population diversity in different parts of the country. The concentrations of mixed-race couples in these locations, combined with their large populations, provide samples big enough to sustain the analysis. Table 1 shows that these metropolitan areas were home to more than one-third of the nation’s mixed-race partnered couples in 2000 and that, on average, 8.8 % of couples were mixed, compared with the national average of almost 7 %. The averages mask variations within the sample, with the three West Coast metropolitan areas having double the share of mixed-race than same-race couples. In contrast, Atlanta, Detroit, and Philadelphia have fewer racially mixed-race than same-race couples.

Our investigation features the three most common heterosexual mixed-race household types: namely, those headed by black-white couples, Asian-white couples, and Latino/Latina–white couples (Passel et al. 2010). Overall, heterosexual couples head more than two-thirds of all mixed-race households. The remaining one-third of mixed-race households include same-sex couples, unrelated housemates, and households where children, many adopted, are reported as having a race different from their parent(s). We predict that the nonheterosexual share of mixed-race households will increase over time and warrant targeted and extensive analysis. Still, nonheterosexual mixed-race households are a broadly heterogeneous group whose locational decisions reflect a wide array of processes. Moreover, our theoretical focus on the possibility of a gender-by-race interaction effect precludes the inclusion of mixed-race households that do not have a male–female couple.

Table 2 shows the degree of variation in the gender configuration of the three types of mixed-race couples studied. The patterns exhibit no clear geography at this scale. For example, among black-white couples, Los Angeles comes closest to gender parity; among black-white couples in San Francisco, however, almost 79 % involve a black man partnered with a white female. Asian-white partnerships exhibit the common pattern of being dominated by a white male/Asian female configuration. Latino-white couples, by metropolitan area, cluster closely to the mean of 46 % white female/Latino male.

We next explore the typical neighborhoods of the three classes of mixed-race couples, contingent on the gender of the white person in the pairing. We do this by first comparing the typical neighborhoods of mixed couples with those of single-race black, Asian, and Latino couples.⁴ We then compare the racial diversity of the typical neighborhood of households headed by mixed-race couples with those headed by single-race pairs. This analysis relies on two variants of the exposure index. Conventionally, P^* represents neighborhood exposure:

where j indexes census tracts, w and x index racial groups, and t is the total population of all racial groups. W is the total population of group w across all tracts; and w_j, x_j, and t_j are tract counts of the respective groups. characterizes group x’s population share in group w’s typical tract: that is, the residential exposure of group w to group x. As we aim to assess the exposure of certain mixed-race households to whites (and, depending on the mix in the household, to blacks, Asians, and Latinos), we modify such that w represents counts of households, and x represents individuals (cf. Holloway et al. 2005). We can further modify by specifying the race of the (fe)male partner and thus describe the exposure, say, of households headed by a black man and a white woman to blacks (or whites): that is, the average tract percentage black (white) of the typical household headed by a black man and white woman.⁵

Figure 1 illustrates the patterns of exposure of the three different classes of couples to (1) whites and (2) the minority population associated with the nonwhite partner, summarized for all 12 metropolitan areas. Ignoring the race of the (fe)male partner for a moment, the values in Fig. 1 depicting exposure to white neighbors are considerably greater than those associated with nonwhite neighbors. Mixed-race households with one white partner are far more likely to encounter whites in their neighborhoods of residence than individuals who are the race of the nonwhite partner. Figure 1 also shows that an increased neighborhood exposure to whites occurs when the male in the partnership is white, regardless of the race of the female partner. The differences are small relative to the differences revealed between exposure to whites versus nonwhites, but are nevertheless consistent across the groups. The raw data in Fig. 1 support the ideas that households headed by mixed-race couples tend to reside in white neighborhoods, and there appears to be a small gender effect.

Echoing the asymmetry of the exposure index itself, we next ask whether households headed by mixed-race couples live in neighborhoods associated with the nonwhite partner. Figure 1 shows that gender has a relatively small effect for black-white and Latino-white couples. If the male partner is black or Latino, the chances of being exposed to black and Latino neighbors is slightly higher than if the female partner is black or Latina. There is no gender difference in the neighborhood exposure to Asians for household headed by Asian-white partners.

Figure 2 provides additional perspective, describing the likelihood of exposure to neighborhood racial compositions by households headed by mixed-race couples compared with their relevant single-race household referent groups. (As before, these data are averages of the 12 metropolitan area values.) Accordingly, Fig. 2, panel A, shows not only the exposure of mixed-race households headed by black-white couples with whites, blacks, Asians, and Latinos, but also the neighborhood exposure of (1) black and (2) white single-race households to those same groups.⁶

Figure 2, panel A, reveals that blacks (in this case, black couples) are far less likely to have white neighbors than are either Latinos or Asians. White couples are also far more likely to have white neighbors than any other racial group. Households headed by racially mixed couples occupy a median position, if you will, in their exposure to these four racial groups when compared with their same-race referents. Furthermore, Fig. 2 expands on the subtle gender differences depicted in Fig. 1. For example, black male/white female household types are not only more exposed to black neighbors than are black female/white male couples, but they are also comparatively more exposed to Latino neighbors. White-Latino couples replicate a similar pattern regarding exposure to blacks (panel C). Both Figs. 1 and ​and22 suggest that at least in the case of white-black and white-Latino household heads, a minority male partner increases the likelihood of having both black and Latino neighbors.

To begin to examine the question of whether gender asymmetries of couples heading mixed-race households is related to neighborhood racial diversity, we deploy a second variant of neighborhood exposure: the Neighborhood Diversity Exposure index (NDE) (Holloway et al. 2005). The NDE indexes the amount of racial diversity (measured using scaled entropy) in the typical neighborhood of a particular group. It captures a group’s exposure to racial diversity in their typical residential neighborhood. The standardized entropy diversity measure for each tract is

where k indexes the racial groups. The maximum value of E_j is obtained when tract j’s population is evenly divided between the k racial groups; the constant s (1 / ln(k)) ensures that E_j ranges between 0 and 1. The NDE captures the racial diversity for group w’s typical tract in the following way:

If group w—say, a household headed by an Asian woman and a white man—disproportionately concentrates in neighborhoods with considerable racial diversity, the NDE takes on a relatively large positive index value. Conversely, if such a household disproportionately concentrates in tracts with little racial diversity, the NDE takes on a relatively small positive value.

Figure 3 portrays the results of this analysis. The three sets of pairs record roughly similar exposures to neighborhood racial diversity; the scores range from a low of .44 (white male/Latina couples) to a high of .51 (black male/white female couples). Figure 3 reveals that mixed-raced couples with a nonwhite male partner encounter elevated levels of neighborhood racial diversity in their place of residence relative to those encountered by mixed-race couples with a white male partner. This is consistent across the three sets of pairs of partners under investigation and also adds a gender dimension to the conjecture that white-nonwhite couples gravitate to racially diverse neighborhoods. This elevated likelihood of exposure to neighborhood diversity is attenuated if the male partner is white.

Figure 3 also illustrates that white-Asian and white-Latino couples (but not white-black couples) encounter higher levels of neighborhood racial diversity than their white same-race counterparts but lower levels than their nonwhite same-race reference groups. White-black couples, however, encounter more neighborhood racial diversity than either white or black same-race partners.⁷ Again, in registering the small gender effect, Fig. 3 reveals that mixed-race households headed by a white male partner trend consistently toward the patterns of single-race white households.

A set of models assesses whether the differences detected in the descriptive phase of the research are statistically significant, taking account of an extensive set of control variables. With tracts serving as proxies for neighborhoods, we estimate three sets of logistic regression models, one set for each mixed-race couple classification, with the following form:

where γ_ij is a community-level measure for tract j and thus is assumed to be constant for all household types i in the same tract j. We estimate this model using three measures of community racial composition as dependent variables: tract racial diversity, measured by scaled entropy, which ranges between 0 and 1; the proportion of whites among tract residents; and, depending on the mixed-race couple being analyzed, the proportion of blacks or Asians or Latinos in tract j. The specifications of the dependent variables match with the descriptive analysis that focused on P^* and neighborhood racial diversity. P^* measures the neighborhood proportion white (or black, Asian, and Latino) in which the average white or minority person (in our case, mixed-race couple) lives. Scaled entropy measures racial diversity in a census tract.

Given that our dependent variables have ranges restricted to fall between 0 and 1, all models are specified as generalized linear models with a binomial variance function and a logit link function. Parameters appear in log odds form and are estimated using maximum likelihood; robust standard errors account for the clustering of observations within tracts. The estimation and significance of the variables of principal interest are very stable throughout this process.

To test the effect of gender on the tract location of households headed by mixed-race couples, we create a simple dummy variable for the male partner being white. (We can equivalently call this dummy variable “female partner not white.”) The controls are other individual- and household-level variables that predict residence in a community with a given level of the trait measured by the dependent variables. These controls are of the following general types: (1) a pair of dummy variables that account for the racial ancestry of the partners in the household; (2) for Asian and Latino/ Latina partners, a set of national-origin ancestry dummy variables; (3) a set of controls that account for mobility, migration, and immigration history, such as location of previous residence and places of birth; (4) standard socioeconomic variables, many of which are used in both SA- and PS-type models, such as household income, education (the education variables are specified as a polychotomous suite of dummy variables that reflect both overall educational attainment and the homogamy of attainment between the partners), and age; (5) two variables that account for military service; and (6) a set of metropolitan area fixed-effect dummy variables.

The decision to aggregate the 12 metropolitan area samples into a single pool represents a trade-off between examining three outcomes for a set of metropolitan areas and examining one outcome for each metropolitan area separately. Our core research hypotheses require that we examine tract percentage white, percentage nonwhite, and racial diversity. Attempting this analysis separately for a dozen metropolitan areas would have produced an unwieldy amount of output. Accordingly, we maintain our attention on these three variables of interest and, following convention, use metropolitan fixed-effect dummy variables to control for unobserved locational heterogeneity across the sample. Aggregating the 12 metropolitan areas and leveraging the 1-in-6 sample, these data produce large samples of each household type. Our models are based on samples of 15,700 households headed by black-white married and partnered couples, 32,338 Asian-white households, and 92,644 white-Latino households. The analysis thus boils down to an analysis of the residential geography of households headed by mixed-race couples composed of a white male partner or a nonwhite male partner from a sample pooled from 12 large U.S. metropolitan areas.

We begin with the models in which neighborhood percentage white serves as the dependent variable (see Table 3). To reduce clutter we exclude the estimations, where applicable, of dummy fixed-effects controls for ancestry.