Housing Discrimination Among Available Housing Units in 2012: Do Paired-Testing Studies Understate Housing Discrimination?

Rob Pitingolo
Urban Institute

Stephen L. Ross
The University of Connecticut

The 2000 Housing Discrimination Study (HDS2000) documented substantial declines in discrimination between HDS1989 and HDS2000, and the most recent study (HDS2012) tends to mirror HDS2000 in its findings. The results of HDS2000 led to considerable debate about whether paired-testing studies of the type conducted in HDS2000 understate the extent of housing discrimination. Using data from HDS2012 and earlier evidence, this article considers three of the significant concerns raised regarding paired-testing studies of housing discrimination: (1) exclusion of minority homeseekers during the process of setting up appointments, (2) the net measure of adverse treatment understating discrimination because some housing units are systematically not shown to White testers, and (3) the use of metropolitanwide advertisements that may systematically underrepresent neighborhoods where discrimination is higher. HDS2012 directly addresses the first concern, finding at most very low levels of discrimination in obtaining an appointment over the phone. The evidence for the second concern is mixed. Steering persisted against both Black and Asian homeseekers in owner-occupied housing. On the other hand, the levels of equal treatment in HDS2012 in terms of basic access were quite high, leaving little room for the systematic exclusion of White homeseekers from specific housing units. Further, three-person tests in HDS2000 involving same-race pairs did not suggest that the net measure was biased. To partially address the third concern, this article conducts a new empirical analysis in which we measure the availability of rental and owner-occupied housing in each broad neighborhood represented in HDS2012 and reweight the tests to represent the spatial availability of housing across each metropolitan site. Although the reweighting substantially changed the weights on individual tests, the average attributes of the neighborhoods represented by those tests experienced only modest changes from reweighting, and the estimated measures of adverse treatment were unchanged.