Housing Inequality in the United States: Explaining the White-Minority Disparities in Homeownership

Abstract

As the homeownership rate in the United States reached its highest ever level in 2004, the distribution of homeownership remained uneven along racial and ethnic lines. Using data from the 2005–2007 3-Year Sample of the American Community Survey (ACS), this paper employs a multivariate regression model and a decomposition technique to delineate the socio-economic and demographic characteristics as well as the immigration and spatial patterns that shape racial and ethnic inequality in homeownership. The findings reveal three distinct patterns; the Asian-white homeownership gap is explained entirely by differences in immigration and spatial patterns of residence, whereas the disadvantage of blacks and Puerto Ricans is attributable to demographic, socio-economic and unobserved factors. For Mexicans and other Hispanics, all four sources influence homeownership patterns, with socio-economic factors relatively important for Mexicans and spatial variables relatively important for other Hispanics.

Keywords:

• Homeownership
• race
• ethnicity
• immigration

Acknowledgement

The authors share equal responsibility for this paper; their names are listed alphabetically.

Notes

¹ For ease of exposition, the remainder of the paper refers to non-Hispanic whites, non-Hispanic blacks and non-Hispanic Asians as whites, blacks and Asians, respectively.

² According to the US Census Bureau estimates, by 2050 about one in two Americans will be non-Hispanic white, more than a quarter will be Hispanic, and the percentage of blacks and Asian and Pacific Islanders will be 15 per cent and 9 per cent, respectively (US Census Bureau, 2008).

³ PUMA is the smallest geographic unit identified in this dataset and it generally follows the boundaries of county groups, single counties or census-defined ‘places’. PUMAs consist of a minimum of 100, 000 residents and do not cross state lines. Note, however, that PUMAs cover a large geographical area particularly in small towns and rural areas and should not be interpreted as a neighborhood.

⁴ There is some concern about the endogeneity of the location choice of households (Myers, 2004; Saiz, 2006); the choice of PUMAs as the unit of location mitigates such concerns because the choice of macro-location is typically influenced by employment considerations rather than housing market conditions (Hilber & Liu, 2008).

⁵ Mexicans comprise approximately 64 per cent of the Latino population (US Census Bureau, 2009b) and 31 per cent of all foreign-born people in the US (US Census Bureau, 2009a).

⁶ The standard logit model assumes that the error terms of all observations are independent. When this assumption is violated, the Huber-White cluster method is able to obtain unbiased standard errors by allowing for correlation of error terms within clusters (e.g. PUMAs) (Rogers, 1993).

⁷ In the ACS sample survey design, all households do not have the same probability of selection. Specifically, minority households have a lower probability of selection than white households. To obtain coefficient estimates for a sample representative of the US population, i.e. avoid bias caused by the overrepresentation of whites, household sampling weights are used. The mean household weights are 31.17 for whites, 35.35 for Asians, 41.81 for blacks, 40.48 for Mexicans, 41.75 for Puerto Ricans and 39.44 for Mexicans.

⁸ The change in the magnitude of the racial-ethnic coefficients depends on the product of two factors that are assumed independent: (1) the partial correlation between X and homeownership; and (2) the partial correlation between X and race-ethnicity. The extent to which the coefficient of one racial-ethnic dummy variable changes relative to the coefficient of another racial-ethnic dummy variable when X variables are added to the model depends on the differences in X variables across racial and ethnic groups.

⁹ When we control for demographic and socio-economic variables the predicted homeownership gap decreases to 0.1966 for blacks, 0.1381 for Asians, 0.087 for Mexicans, 0.2183 for Puerto Ricans and 0.1548 for Other Hispanics.

¹⁰ For example, the predicted homeownership gap of Asians, as represented by the marginal effect of the Asian dummy variable, decreases from − 0.1381 to − 0.0405.

¹¹ For example, the marginal effect of the Mexican coefficient reduces to − 0.0049, and is no longer significantly different from zero.

¹² For example, the marginal effect of Puerto Ricans decreases from − 0.2183 to − 0.513 when immigration and spatial controls are added.

¹³ The race-ethnic specific regressions are equivalent to running a logit regression with a full set of interactions between race-ethnicity dummies and the covariates.

¹⁴ The difference between white and black coefficients is not statistically significant (p = 0.40), but these two groups have significantly smaller coefficients compared to the other groups (p < 0.05).

¹⁵ The homeownership rates for native-born blacks and Puerto Ricans are 43 per cent and 39 per cent, respectively, compared to 72 per cent for whites, 61 per cent for Asians, 54 per cent for Mexicans and 51 per cent for Other Hispanics.

¹⁶ The difference in the coefficients between these two groups and the other four is statistically significant (p < 0.05).

¹⁷ The markers represent the duration at which immigrants of each group reach the mean predicted homeownership rate of their native born counterparts. Duration is capped at age 40 because the number of observations is relatively low for older cohorts, and the interpretation of the duration effect is more difficult for the oldest cohorts because of lifecycle effects associated with downsizing.

¹⁸ This figure uses cross-section data from different immigration cohorts to construct a snapshot of the immigrant experience in homeownership. The patterns that emerge can be generalized to predict the trajectory of a particular cohort only if the average quality of immigrant cohorts, in terms of favorable socio-economic, demographic and spatial attributes, does not change over time. If the quality of immigrants has improved in recent cohorts, the trajectories constructed with cross-cohort data provide lower bounds of the expected trajectory of the recent cohorts.

¹⁹ It is possible to use minority coefficients or a pooled sample coefficient to construct the explained part. The white coefficients were chosen because it was possible to estimate how much the homeownership gap will narrow if minority groups had the similar marginal ‘returns’ in the housing market as the majority (white) group.

²⁰ In fact, these two factors collectively explain more than 100 per cent of the gap. This over-explanation is due to negative socio-economic and demographic components that reflect an Asian advantage relative to whites in these respects.

²¹ For Mexicans, the unexplained part is negative, i.e. if Mexicans achieve the mean level of whites in the observable variables, their homeownership rate will exceed that of whites.

²² The minority or pooled sample coefficients could also be used to construct the explained part. White coefficients were chosen because it allowed interpretation of the explained component as the remaining homeownership gap if minority groups had the similar marginal ‘returns’ in the housing market as the whites.