Defining Housing Discrimination
The goal of the HDS is to measure the incidence of disparate treatment discrimination by housing agents during their interaction with borrowers who approach them as a result of a newspaper advertisement. Adverse disparate treatment could result from racial prejudice, financial incentives of the real estate agent, or other factors. The goal of the study is not to determine the cause of racial differences in treatment.
Stephen Fienberg commented that studies of discrimination in labor markets (e.g., Heckman, 1998) address the notion of distinguishing between market discrimination and the discrimination encountered by a random person responding to a randomly selected advertisement. The methodology in the labor market context is similar to that employed in the HDS audit. Heckman's paper offers the following definition of discrimination: “an otherwise identical person is treated differently by virtue of that person's race or gender, and race and gender by themselves have no direct effect on productivity.” According to Heckman, discrimination is the effect of race that arises from a ceteris paribus hypothetical experiment in which race is allowed to vary while all other aspects of the individual and circumstances are held constant.
DISPARATE TREATMENT VERSUS DISPARATE IMPACT
During the discussion of methodological implications of the Phase II audit design, participants explored the differences between disparate treat
ment and disparate impact discrimination. Disparate treatment discrimination is defined as negative treatment of minority candidates due solely to the candidates' race. Disparate impact discrimination occurs when a system is put in place that is not discriminatory in intent, but negatively impacts a particular group of individuals. When housing providers deny or make housing unavailable to persons on the basis of characteristics not protected by the Fair Housing Act and when these characteristics are correlated with race, the result is disparate impact discrimination, not disparate treatment discrimination.
Disparate impact would occur, for example, if a lending institution did not finance older homes. In this case, the basis for denial would not be one of the classes protected by fair housing laws, and the policy would be universally applied. If racial minorities are much more likely to live in older homes than whites, however, the policy would exclude a higher proportion of racial minorities. Although whites and minorities would be treated similarly, the policy would adversely impact the protected group and thus constitute a form of discrimination. The housing provider would have to demonstrate that there was a business necessity for the policy and establish that there was no less discriminatory alternative that could serve the same business objective. Audit methodology is designed to measure only disparate treatment discrimination.
During his presentation, Gregory Squires, Department of Sociology, The George Washington University, suggested that, contrary to what some believe, paired testing can potentially uncover the existence of disparate impact discrimination in a given housing market. He asserted that, based on information provided to the minority and white auditors during the test, analysts can observe instances of disparate impact not recorded as disparate treatment. For example, the housing provider might share with the auditor information about the agency's policies and practices that may differ for minority and white home seekers. This information would not necessarily appear on the auditor's forms, but would be part of the narrative the auditor provided to the researchers. Though the policies highlighted would be applied to both minority and white auditors, they could differentially impact minority home seekers.
A related discussion addressed the ability to measure discrimination statistically given the legal definition. Some individual and household characteristics that are associated with disparate effects have a disparate impact because their distributions vary with race. As noted earlier, for enforcement audits, testing coordinators control for other factors to isolate the
effect of race on the treatment recorded. Participants questioned the experimental approach used in enforcement audits and employed in the HDS. While conceptually one can hold all factors but race constant, doing so may not be possible in the actual housing market. In contrast with an experiment where a patient is given a placebo or treatment, one cannot assign or change an individual's racial identity. In the absence of this ability to randomize, researchers typically use an “approximate” study design that cannot be manipulated experimentally.
Participants noted that the methodology used is dependent on the research question of interest. Some participants expressed the underlying question as: If African Americans were whites or whites were African Americans, how would they be treated in the housing market? This question suggests a baseline of no racial discrimination. Other participants argued for a different framing of the question: In the absence of racial discrimination, how would a minority or majority home seeker be treated during the housing transaction?
GROSS AND NET ADVERSE TREATMENT
As discussed by Ross in his workshop remarks and his paper “Paired Testing and the 2000 Housing Discrimination Study” (see Appendix A), the HDS data are used to generate two common alternative measures of differential treatment. The first, gross adverse treatment, measures the frequency of audits in which a white auditor was treated favorably and a minority auditor was treated unfavorably. The second measure of differential treatment, net adverse treatment, measures the frequency with which the white auditor was treated favorably, minus the frequency with which the minority tester was treated favorably. Differential treatment could result from discrimination by the housing agent or from legitimate, not discriminatory factors. The gross measure will count legitimate nondiscriminatory racial differences (e.g., the unit having actually been rented between the visits of the white and minority auditors) as instances of adverse treatment because researchers cannot observe the intent of the housing agent. While these instances will be counted in the net measure as well, the presumption is that if differential treatment is not due to race and the order of audit visits is randomized by race, rates of adverse treatment for whites and minorities will cancel each other out.
The gross measure of adverse treatment, then, overestimates discrimination by including nondiscriminatory disparate treatment resulting from
random, unobserved differences in auditors. Conversely, the net measure, although intended to capture differences in treatment that result from racial discrimination, underestimates discrimination. The hypothesis underlying the net measure is that the frequency with which the minority auditor is treated adversely because of factors unrelated to race can be proxied by the frequency with which the white auditor is treated adversely. Underestimates of discrimination result because instances in which the white auditor is treated less favorably are netted out, even though these differences may be attributable to unobserved adverse treatment of the minority auditor (Ondrich et al., 2000). The use of gross and net measures is discussed in greater detail in Chapter 5.