Interpretation of Parameter Estimates
The 2010 intercepts and 1980 to 2010 slopes shown in Figure 1 represent averages for the group-specific samples of MAs. However, there is considerable variation across MAs around these average intercepts and slopes. Tables 2 and 3 examine the extent to which this variation is related to housing supply, affordable housing, suburban employment, and ecological context measures. Because all covariates have been grand mean centered, the intercepts can be interpreted as covariate-adjusted averages for all MAs. The coefficients in Table 2 are interpreted as variation in 2010 levels of poverty suburbanization associated with one-unit changes in the independent variables (measured in 2000). The log of MA population coefficient is interpreted as the expected change in the dependent variable associated with a one percent change in MA population.
Table 2. Coefficients From HLM Regression of 2006/2010 Levels of Poverty Suburbanization on Measures of Housing Supply, Affordable Housing, Suburban Employment, and Ecological Context
|Housing and employment variables|| || || || || || || || || || || || || || || || |
|Suburban:central city housing stock||7.70||0.85*||7.79||0.94*||5.05||0.84*||6.07||0.83*||1.59||0.44*||2.88||0.79*||0.65||0.71||1.27||0.70|
|Suburban affordable owner-occupied||—|| ||—|| ||—|| ||—|| ||0.21||0.11||−0.26||0.18||0.53||0.14*||0.49||0.16*|
|Suburban affordable rental||—|| ||—|| ||—|| ||—|| ||0.65||0.11*||0.40||0.20||0.64||0.15*||0.74||0.17*|
|Suburban total employment||—|| ||—|| ||—|| ||—|| ||0.47||0.05*||0.68||0.11*||0.17||0.08*||0.29||0.09*|
|Control variables|| || || || || || || || || || || || || || || || |
|% own group||—|| ||—|| ||—|| ||—|| ||−0.06||0.05||0.07||0.14||0.34||0.08*||0.01||0.06|
|Log MA population||—|| ||—|| ||—|| ||—|| ||−0.76||0.76||0.97||1.58||1.36||0.92||0.15||1.24|
|Northeast||—|| ||—|| ||—|| ||—|| ||0.49||2.13||−10.20||4.15*||−12.17||3.13*||−22.66||3.68*|
|Midwest||—|| ||—|| ||—|| ||—|| ||4.67||2.04*||−3.35||4.30||−6.87||3.06*||−1.80||3.13|
|South||—|| ||—|| ||—|| ||—|| ||13.20||1.76*||7.76||3.38*||−2.96||3.32||6.58||2.76*|
|1900 to 1939||—|| ||—|| ||—|| ||—|| ||−1.43||1.79||−4.41||3.74||1.29||1.90||−2.60||2.76|
|1940 to 1969||—|| ||—|| ||—|| ||—|| ||−1.67||2.29||2.59||4.56||7.49||2.69*||2.96||3.94|
|1970 or later||—|| ||—|| ||—|| ||—|| ||−3.35||2.40||−1.09||5.43||9.44||3.34*||1.21||4.28|
|Level 2 unconditional variance||442.7|| ||579.9|| ||386.4|| ||568.2|| ||442.7|| ||579.9|| ||386.4|| ||568.2|| |
|Model residual variance||230.8|| ||310.9|| ||261.3|| ||391.1|| ||106.9|| ||222.5|| ||119.2|| ||184.4|| |
|% of level 2 variance explained||47.9|| ||46.4|| ||32.4|| ||31.2|| ||75.8|| ||61.6|| ||69.2|| ||67.5|| |
Table 3. Coefficients From HLM Regression of 1980 to 2006/2010 Change in Levels of Poverty Suburbanization on Measures of Housing Supply, Affordable Housing, Suburban Employment, and Ecological Context
|Housing and employment variables|| || || || || || || || || || || || || || || || |
|Suburban:central city housing stock||0.14||0.01*||0.07||0.07||0.12||0.03*||0.19||0.03*||0.15||0.01*||0.07||0.07||0.10||0.03*||0.13||0.04*|
|Suburban affordable owner-occupied||—|| ||—|| ||—|| ||—|| ||0.00||0.01||−0.04||0.02||0.01||0.02||0.00||0.01|
|Suburban affordable rental||—|| ||—|| ||—|| ||—|| ||0.00||0.00||−0.04||0.04||0.01||0.01||0.03||0.02*|
|Suburban total employment||—|| ||—|| ||—|| ||—|| ||−0.01||0.01||0.04||0.07||0.01||0.02||0.00||0.04|
|Control variables|| || || || || || || || || || || || || || || || |
|% own group||—|| ||—|| ||—|| ||—|| ||−0.03||0.03||0.00||0.04||0.00||0.03||−0.02||0.02|
|MA population||—|| ||—|| ||—|| ||—|| ||−0.02||0.01||0.06||0.07||0.07||0.04||0.02||0.04|
|Northeast||—|| ||—|| ||—|| ||—|| ||0.56||0.42||−1.33||2.08||0.54||1.60||0.49||1.15|
|Midwest||—|| ||—|| ||—|| ||—|| ||0.56||0.42||0.46||1.83||0.49||1.16||1.50||1.19|
|South||—|| ||—|| ||—|| ||—|| ||0.61||0.36||1.02||1.69||−1.00||0.89||1.72||0.89|
|1900 to 1939||—|| ||—|| ||—|| ||—|| ||−0.46||0.33||−3.61||1.49*||−1.44||0.73||−1.50||0.91|
|1940 to 1969||—|| ||—|| ||—|| ||—|| ||−0.78||0.37*||−3.37||1.71||−2.14||0.85*||−2.39||1.29|
|1970 or later||—|| ||—|| ||—|| ||—|| ||−0.94||0.37*||−4.51||1.90*||−1.95||1.20||−3.00||1.18*|
|Level 2 unconditional variance||5.2|| ||40.8|| ||18.6|| ||23.3|| ||5.2|| ||40.8|| ||18.6|| ||23.3|| |
|Model residual variance||2.8|| ||39.2|| ||16.7|| ||18.4|| ||2.7|| ||39.5|| ||15.9|| ||17.4|| |
|% of level 2 variance explained||45.5|| ||4.0|| ||10.6|| ||21.3|| ||48.7|| ||3.0|| ||14.7|| ||25.6|| |
In Table 3, the coefficients on the time-varying variables are interpreted as effects of one percent per decade changes in the independent variables on average decadal change in poverty suburbanization for each group. So, for example, a “suburb:central city housing stock” coefficient of 0.20 would indicate that for every one percent per decade increase in the ratio of suburban to central city housing supply, a group's poverty suburbanization increased an average of two-tenths of a percent per decade across all MAs. The region and age dummies are interpreted as increments or decrements to the intercepts for the included region and age categories, relative to the omitted categories (the West and older than 1900, respectively). For example, if the “South” coefficient were 1.0, this would mean that poverty suburbanization increased one percent per decade faster in the South than in the West.
Interpretation of Standard Error Estimates
The robust standard error estimates provided by the HLM software assume some kind of probability sample, though not necessarily a simple random sample. In this article, however, we analyzed repeated measures from all MAs that met the criterion of at least 1,000 members of a racial or ethnic group in 1990. Hence, the standard errors should be interpreted cautiously, and more as “estimates of parameter dispersion contaminated by measurement error” (Grodsky & Pager, 2001, p. 552) rather than sampling variability per se. In other words, smaller standard errors indicate more consistent correlations of the independent variables on minority suburbanization. We include the standard error estimates in Tables 2 and 3, and mark coefficients with an asterisk to indicate that they are at least twice the size of their associated standard errors.
Table 2 presents coefficients from HLM regressions of MA-level variation in poverty suburbanization in 2010 on the independent variables described above, measured in 2000. Model 1 estimates the relationship of housing supply only, allowing for a baseline estimate of the extent to which higher levels of suburban housing stock were associated with the suburbanization of the poor of each racial/ethnic group. We find that suburban housing supply is associated with the suburbanization of the poor for all four groups in 2000, although the relationship varies in magnitude. For Whites and Asians, on average, MAs that are one unit higher on the ratio of suburban to central city housing stock (about half of a standard deviation in 2000—see Table 1) have 7.7 and 7.8 point higher shares of the White and Asian poor in the suburbs, respectively. For Blacks and Latinos this association is somewhat lower, at 5.1 and 6.1, respectively.8 This finding suggests that, on average, the White and Asian poor have been more successful at responding to higher levels of suburban housing stock than have the Black and Latino poor. Whether this indicates less discrimination against the White and Asian poor, or more effective social networks enabling these groups to find out about suburban housing opportunities, or some other explanation, we cannot say with our data.
Model 2 in Table 2 introduces a host of controls for the association between suburban housing supply and the suburbanization of the poor. First, we include measures of housing affordability to assess the degree to which affordable housing in particular is associated with the suburbanization of the poor. Second, we control for the supply of jobs in the suburbs to enable us to assess the independent relationship of housing supply to poverty suburbanization, net of the correlation between housing supply and suburban labor demand.
In terms of housing affordability, we find that the supply of affordable owner-occupied and rental housing is positively and significantly associated with the suburbanization of the Black and Latino poor, with coefficient magnitudes ranging from about 0.49 to about 0.74. For instance, we find that the Latino poverty suburbanization rate varies with respect to affordable rental housing at a rate of 0.74% per one percent change in the supply of affordable rental housing. For Whites, only the rental housing coefficient is significant at conventional levels, and for Asians neither coefficient is significant.9 Hence, our findings suggest that, net of total housing supply, the Black and Latino poor are more sensitive to the supply of affordable housing than are the White and Asian poor.
In contrast, our findings for the association of labor demand with poverty suburbanization suggest that the White and Asian poor tend to live in suburbs at a higher rate when there are more suburban jobs available. For example, for every one percent change in suburban employment the expected poor White and Asian suburbanization rates increase between 0.47% and 0.68%, respectively.10 It is difficult to know the precise mechanism for this association; however, it indicates the ease with which the White and Asian poor have responded to employment pull factors from the suburbs, relative to their poor Black and Latino counterparts.11 It is possible that job suburbanization has occurred in suburban areas that are either more preferred by Whites and Asians, or more discriminatory toward Blacks and Latinos. Indeed, some research on the spatial mismatch between areas of Black settlement and employment has argued that the cause of low rates of Black suburban employment is “race, not space” (Cohn & Fossett, 1996, 1998; Ellwood, 1986).12
The findings in model 2 also indicate that the association between housing supply and poverty suburbanization rates is heavily attenuated by the presence of controls. Indeed, the housing supply ratio coefficients were reduced from model 1 to model 2 by 60% to 80% for each group. This indicates that cities with large supplies of suburban housing also had large supplies of affordable housing and jobs. This is unsurprising, of course; however, the findings in model 2 indicate that a residual relationship between suburban housing supply and the presence of poor in the suburbs remains after controlling for these confounding factors, especially for Asians.
Model 2 also controls for several measures of ecological context, including MA size, percent of each racial/ethnic group, region, and the period in which the central city of each MA surpassed 50,000. For Blacks, higher percentages of Blacks are associated with higher percentages of poor Blacks in the suburbs. We did not observe a similar relationship for the other groups. MA size was not significantly correlated with poverty suburbanization, controlling for all other variables in the model. We did observe substantial regional variation by race and ethnicity. For Whites, Asians, and Latinos, the poor are more suburbanized in the South compared to the West, whereas for Blacks, suburbanization rates in the South and West are statistically indistinguishable. Poverty suburbanization for Blacks is somewhat lower in the Midwest than the West, and poverty suburbanization for all three minority groups is much lower in the Northeast. For example, the share of the Latino poor in the suburbs of the average northeastern MA is fully 22.66 percentage points lower than the statistically equivalent Western MA. This is likely due to the historical concentration of Puerto Rican and Dominican immigrants in northeastern central cities, whereas western MAs have historically been more evenly populated by Mexican and other Latin American immigrants. Finally, we observe higher rates of poverty suburbanization for Blacks in newer metropolitan areas relative to the oldest category, those whose central city reached 50,000 prior to 1900. This is likely due to newer MAs having less of an entrenched history of racialized residential patterns (Farley & Frey, 1994; Timberlake & Iceland, 2007).
Change in Poverty Suburbanization Over Time
Table 3 presents results from our analysis of change over time in the share of the poor who live in suburbs in U.S. metropolitan areas. The “intercept” row in these models indicates the average percent per decade increase for MAs that are average on all included covariates. As indicated above in our discussion of Figure 1, the rate of increase in poverty suburbanization was higher for Asians and Latinos than for Whites and Blacks. We interpret the relatively low rate of increase for Whites as reflecting the already-high rate of poverty suburbanization of the White poor. Because the Black rate was so low in 1980, it suggests that the Black poor have experienced disproportionate difficulty in attaining suburban residence relative to their Asian and Latino counterparts.
Model 1 of Table 3 only includes the measure of per-decade change in the suburban:central city housing stock ratio (see equations (1) and 2). We find that although the poor of all groups were able to convert increases in suburban housing stock into increases in suburban location, this conversion occurred more efficiently for Whites and Latinos than for Asians and Blacks. However, statistical tests indicate that we cannot reject the null hypothesis that the between-group differences are statistically indistinguishable from zero. Hence, although the point estimates for Whites and Latinos are more than double that for Asians, the relatively inconsistently measured association for Asians renders the differences not statistically significant.
In Model 2 we control for the relationship between changes in the supply of affordable housing and jobs in the suburbs. In contrast to Table 2, the presence of these controls and the ecological context variables did not attenuate the correlations of changes in housing supply. In addition, only changes in affordable rental housing were associated with changes in the suburbanization of the Latino poor. This indicates that changes in overall housing supply were independently related to changes in the rate at which the White, Black, and Latino poor suburbanized; that is, net of correlations between changes in overall housing supply, affordable housing supply, and suburban employment demand.
Model 2 also controls for changes in MA population size, percent racial/ethnic group, plus region and age. For the first two variables we observe no consistently measured associations; hence, it does not appear that changes in MA size or racial/ethnic distributions were significantly related to change in poverty suburbanization, net of all other variables in the models. Similarly, we found no significant differences by region in the rate at which the poor of these four racial/ethnic groups suburbanized. Finally, we observed relatively uniform associations of MA age, with the poor in older MAs suburbanizing more rapidly than the poor in newer MAs. For example, compared to the oldest MAs, those whose central city passed 50,000 prior to 1900, the Asian poor in the newest MAs were suburbanizing at a 4.51% per decade lower rate. The combination of the cross-sectional and temporal correlations suggests that older MAs had lower minority suburbanization rates to begin with, and therefore had more “room” to increase over time.