Overview of the 2000 Housing Discrimination Study
The HDS, announced by HUD Secretary Andrew Cuomo in November 1998, is intended to be a 3-year study of discrimination in the U.S. housing market. The results will extend and expand HUD's 20-year history of measuring discrimination in housing through the methodology of paired testing or audits.
The principal objective of the HDS is to develop a national estimate of discrimination in housing, as well as metropolitan-level estimates that can be compared across time. As noted earlier, report cards will be developed at both the national and community levels to provide a benchmark against which to measure progress toward eliminating discrimination. The study will explore the statistical concept of racial discrimination, which may or may not be the same as the legal concept. The law determines discrimination in individual cases of real people; the HDS audit seeks to create a measure of discrimination by which the incidences of discrimination can be counted.
While past HDS audits have measured some effects of discrimination for particular cities, the current audit will expand that effort and obtain more detail on where discrimination occurs within a city. Further, the study will assess what kinds of housing markets may have an effect on racially disparate treatment. The HDS audit has clear implications for
enforcing fair housing laws. Because the study is an audit, however, no enforcement activity will occur as a direct result of the testing outcomes.
As noted in Chapter 1, the 2000 HDS audit involves 60 sites and is a multiphase study: Phase I, including 20 sites, began in 2000; Phase II began in 2001; and Phase III will begin in 2002. In Phase I, an attempt is being made to obtain national estimates of disparate treatment in home seeking among African American and Hispanic groups; these estimates will be used to measure changes in discrimination over time since the most recent HDS audit (1989), as well as for future studies. To assess the appropriateness of the current testing methodology for other ethnic groups, the Phase I study also includes pilot testing for Asian American and American Indian groups. Phases II and III, which were still in the development stages when the workshop was convened, are intended to expand upon Phase I by producing more-precise estimates of the incidence of discrimination at the national level. Based on the results of pilot testing in Phase I, Phase II will extend the analysis of minority groups from African Americans and Hispanics to Asian Americans and American Indians. The study will include site-specific estimates and pilot sites for three Asian American groups— Chinese, Korean, and Southeast Asian—and for American Indian groups in rural areas.
Estimation of discrimination for these populations represents a new area for the HDS. Measurement issues arise with each new group being studied. For example, measuring discrimination in the American Indian community requires changes to the original sampling design. Rather than focusing in metropolitan areas, the sampling frame will include less-populated areas in an attempt to depict more accurately the housing market these groups face.
PHASE I STRATIFIED RANDOM SAMPLING DESIGN
The Phase I design uses two-stage cluster sampling. The sampling frame is a collection of 105 metropolitan statistical areas satisfying the following criteria: (1) according to 1980 census data, the population exceeded 100,000 residents; and (2) the concentration of African American households exceeded a nominal threshold. Metropolitan areas were stratified into (1) those sites for which both African American and Hispanic
testing would occur because the population and concentration thresholds were exceeded for both ethnic groups; and (2) those sites for which only African American testing would occur because the thresholds were met only for that population.
The first stage of the sampling frame, area probability sampling, involves selecting sites for the African American and Hispanic samples, with selection probabilities proportional to population size (using census data). The second stage entails selecting advertisements for both rental and sales housing from the Sunday newspapers within those metropolitan areas. Prior to sampling, analysts identify the major metropolitan newspaper for every site on the basis of circulation and geographic coverage. In the case of sites with more than one major newspaper circulating to different communities within the metropolitan area, the newspapers are rotated from week to week. This rotation is employed in an attempt to capture potential differences in patterns of advertisement across the different communities. If such differences exist, the sampling frame results in better coverage of the entire geographic area.
As noted in Chapter 1, the 2000 HDS employs the paired-testing methodology used in prior HDS audits. The study includes approximately 5,000 tests from a sample of about 20,000 newspaper advertisements of available renter- and owner-occupied housing units.
Paired testing has been used extensively in studies on employment, homeowner's insurance, mortgage lending, and automobile sales. Its most extensive use, however, is in the area of housing, both renter- and owneroccupied, in which there have been three national studies (including the current HDS). Paired testing has also been used in multiple small regional studies and in a great deal of enforcement testing, where it has led to numerous fair housing cases and settlements.
The protocol for a paired test in housing studies is designed to establish a point of entry into the housing market that is realistic and consistent for all testers. In paired testing, two people (auditors) pose as equally qualified customers inquiring about an advertised housing unit. The only apparent difference between the two auditors is their race or ethnicity; they are similar in age, gender, dress, and other observable characteristics that can be controlled a priori. The testing coordinator is responsible for sending audit pairs to a test site, determining their assigned characteristics, and
ensuring that they match on all observable characteristics except race. The auditors are trained to follow a common protocol when they engage in the housing market transaction to minimize the effect of naturally occurring personal differences. Since characteristics other than race that might distinguish the testers and possibly influence the housing agent have been controlled for, differences in auditor reports are assumed to be attributable to racial differences in treatment.
There are a number of advantages to using paired testing instead of market data. One advantage is that the testing provides a structured point of entry that yields an endogenous sample and protocol. Once a point of entry has been established and a sample drawn, that sample represents an exogenous entry into the market or is an exogenous sample of entries into the market.
A second advantage of paired testing relates to mitigating behavior of individuals. Economic theory suggests that people act to maximize their welfare. A limitation of market data is that they cannot measure whether individuals anticipate that they will be discriminated against and take mitigating actions to limit the effect of that discrimination. For example, an individual may pursue a higher-cost lender because he expects to face discrimination in the prime mortgage market. Mortgage lending data (market data) will show that this person applied for a mortgage and was approved for the loan. However, the data will not reflect racial differences in underwriting or the higher premium the individual paid for the loan. Housing segregation is another area that may reflect mitigating behavior. A family may visit a particular real estate agency because they expect to be treated fairly. If, however, that agency focuses on housing that happens to be located in minority communities, the agency's sales will result in racial residential segregation. When analyzing market data, one often cannot control for the cost of such mitigating behavior. Paired testing is therefore preferred because it does not count this potentially costly behavior.
By exploiting the benefits of paired testing, the 2000 HDS can provide estimates of housing discrimination that are not obtainable using housing market data. The methodology employed in the HDS audit uses a common protocol. By assigning characteristics and controlling for the behavior of the auditors, the researchers attempt to ensure the objectivity of the study and limit the influence of mitigating factors. If a minority tester's expectation about being discriminated against affected his or her behavior in the test, the analysts would observe differences in behavior
across the minority testers. Therefore, one measure of mitigating behavior is deviation from the protocol, observed through heterogeneous outcomes of tester pairs.
Housing market data and paired testing are ideally complementary, with testing capturing the level and frequency of discrimination while ignoring the potential effect of mitigating behavior. Market data will capture mitigating behavior, but may miss some of the substantial costs associated with that behavior. Unable to measure whether observed outcomes result from perceived discrimination or racial preferences, market data may understate the effects of discrimination. Conversely, testing data may overstate the effects of discrimination in a particular housing market. Thus, for example, African American auditors may be discriminated against more often than African American home seekers in the housing market because testing does not allow for mitigating behavior.
During his comments, Stephen Ross, Department of Economics, University of Connecticut, suggested that paired testing can clearly distinguish between disparate treatment and disparate impact discrimination and other possible biases that may exist in market data. Market data may also capture variation across practices of housing agents resulting in racial differences in outcomes. Testing avoids this variation by sending auditors to the same housing agent and thus provides a clean test for one type of discrimination. Paired testing also avoids omitted variable bias, endogeneity, and sample selection bias (Ross, 2000). It provides a controlled environment that assigns or controls for many characteristics of the auditor and collects data on unassigned attributes of the auditor and housing unit. Extensive training and protocols diminish the effect of these potentially biasing components.
Testing has already proven to be effective in the enforcement arena, possibly resulting in more-comprehensive settlements. Some social science researchers have also used testing to identify the underlying dynamics of discrimination in ways previously unavailable. Although there are practical and theoretical limitations to audit tests, it is important to note the usefulness of the tool and the advantages to further developing audit methodology. Improvements to the current audit methodology discussed during the workshop included performing more tests within a site, making the audit process more standardized, and using actual home seekers instead of trained auditors.
PAIRED TESTING AND THE 2000 HDS
The use of paired testing in the 2000 HDS begins with the researchers sending the selected newspaper advertisements to the local test coordinator, who makes advance calls to verify the availability and eligibility of the advertised unit. Workshop participants asked about the criteria used to determine which audit pair will visit a particular advertised unit. Urban Institute researchers responded that advertisements from each metropolitan newspaper are randomized and assigned a control number from 1 to N. The local test coordinator receives a faxed copy of the advertisement and begins with control no. 1, proceeding in sequential order until the required number of tests for that week has been performed. After verifying the unit's availability, the test coordinator completes a test assignment form that standardizes the profiles of the two auditors. The form also randomizes the order in which the auditors visit the housing unit and provides a complete financial and household profile to ensure the credibility of the auditor as an applicant for housing.
The auditors then have a face-to-face meeting with the housing agent. They inquire about the specific advertised unit, as well as other housing that may be available. In both sales and rental cases, the auditors inspect the housing unit. After departing from the site, each auditor immediately completes a series of standard forms that captures the testing experience. The test forms are used to record various objective factors, including waiting time, name of the agent, documents received, documentation required, amenities of the unit, and other factors related to unit cost. For sales testing, the forms include information on whether the tester was asked about loan prequalification or the process involved in securing a mortgage.
From the forms, researchers obtain characteristics of the homes that were inspected, along with addresses of additional units recommended by the agent. For a percentage of cases, a narrative of the entire visit is also provided and reviewed during debriefing. The test coordinator compiles all materials related to the test and sends them to the Urban Institute for data entry. The local test coordinator performs no treatment comparisons; only members of the HDS audit research team make an assessment of racially disparate treatment.
As noted, Phase I includes 20 sites; among them are African American/ white sites, Hispanic/white sites, Asian American/white sites, and Ameri
can Indian sites. Overall there are approximately 1,200 paired rental tests and 1,200 paired sales tests. The current Phase I design does not include non-white testers paired with other non-white testers (e.g., African American/Hispanic or African American/Asian American pairs); these pairings may be considered in Phase II. Participant Stephen Fienberg, Maurice Falk University Professor of Statistics and Social Science and Acting Director, Center for Automated Learning and Discovery, Carnegie Mellon University, noted that non-white/non-white testing could bolster what is learned from white/non-white comparisons.
Fienberg expressed concern about the fact that those conducting the HDS are involved in both measurement and enforcement activities. That linkage could have implications for the ongoing nature of the study and the validity of the data collected, especially if it is in the minds of the parties being tested. The HDS audits are not the only testing occurring in the test sites. HUD has funded testing for many years, and all of the agents included in the HDS results are operating in markets where there is also a fair-housing group performing enforcement testing. Housing providers are aware that testing is ongoing; therefore the confidentiality of the HDS audits is an issue. Perhaps advance word of the HDS could spread and distort the results. If real estate agents suspect they have been audited, they may contact other members of the agent community within a test site, thus invalidating the remaining audits. The researchers responded that they believe they would learn quickly if housing agents were aware of nonroutine testing efforts. 1
Of methodological concern is whether researchers discard an audit when a housing provider identifies a member of the audit pair as an auditor. The researchers responded that auditors are trained to address issues of detection and to continue with the test when possible. In some instances, the test coordinator may decide to invalidate a test. Detection occurred in another study in a manner that would have compromised subsequent audits at a site, and testing in that site was terminated. Researchers have not encountered this problem in the HDS to date.
1Researchers stated they believe they would be able to learn whether housing agents suspected a systemic audit because those performing field reconnaissance and local fair housing agencies would be made aware of this information.
CHANGES IN THE NATURE OF HOUSING DISCRIMINATION AND IMPLICATIONS FOR THE AUDIT DESIGN
One participant expressed concern that the paired-testing methodology generally focuses on racially differential treatment at the pre-application stage of housing market transactions. Yet testing and enforcement data from some fair housing agencies suggest that minority home seekers are also vulnerable to discrimination in the form of different terms and conditions after their application has been submitted for review. Consequently, audit results will yield an estimate of discrimination for a portion of housing market transactions, but may not provide the benchmark that HUD desires—changes in the nature of discrimination over time. The number of incidents of discrimination may decline not because discrimination has decreased, but because it has shifted to another stage of the housing transaction. If this is the case, the current audit protocol will underestimate racial discrimination.
Representatives from HUD and the Urban Institute acknowledged that the nature of housing discrimination may be changing. The question they face is how to measure post-application discrimination, since auditors do not submit applications. While auditors involved in sales testing make multiple visits in order to appear as serious buyers and to view multiple properties, they do not make offers on any of the units they see. The paired-testing methodology is probably not the solution for addressing this issue, and the researchers asked for suggestions for alternative methods that could be used to capture this phenomenon.
It was also noted that in many housing markets, real estate agents are now asking individuals to sign up with a buyer's broker before viewing any housing units. This practice poses an additional challenge for the audit structure. First, since auditors may be involved in multiple tests, the practice can increase the likelihood of their being detected. Also, researchers believe the influence of the buyer's broker is more important in some housing markets than in others. Through auditor reports, they collect information on whether minority auditors are required to sign such an agreement while white auditors are not. Racial differences in these requirements may represent discrimination.