Audits for Discrimination
Audits for Discrimination
An audit is a survey technique that isolates the impact of a person’s group membership on the way he or she is treated in the marketplace. Audits, also called tests, first appeared in the 1950s. Since then they have been used to study racial, ethnic, or gender discrimination in car sales, hiring for entry-level jobs, home insurance quotes, preapplication treatment in the mortgage market, house sales, and apartment rentals. National audit studies of housing discrimination were conducted in the United States, for example, in 1977, 1989, and 2000, and audits have been used to help enforce civil-rights laws, particularly in the housing market.
In an audit study, people from two groups are selected, trained, and paired so that they are equally qualified in the market being studied. Audit teammates may be equally qualified to buy a house or a car, for example. One member of the team belongs to a legally protected minority group and the other belongs to the comparable majority group. A sampling frame, such as newspaper advertisements, is then selected; a sample is drawn; and an audit is conducted for each observation in the sample. Audit teammates paired for an observation successively visit the associated economic agent, usually in random order, to engage in the relevant market activity, such as applying for a job or inquiring about an available apartment. Each teammate then independently records how he or she was treated. Discrimination exists if minority auditors are systematically treated worse than their teammates. Most audit studies observe several types of agent behavior; discrimination may exist for some types and not for others.
Audits are an alternative to regression studies, which use statistical procedures to determine whether some economic outcome is less favorable for people in a minority group, after controlling for relevant individual characteristics. This approach leads to biased results if key characteristics are omitted from the regression. An audit study minimizes the possibility of omitted-variable bias by matching similar individuals; giving them the same training; assigning them similar or identical characteristics, such as income or education, for the purposes of the audit; and sending them to visit economic agents in response to the same advertisement within a short time of each other. These procedures ensure that audit teammates do not differ significantly in terms of any characteristic, other than group membership, that might influence their treatment in the marketplace.
Audit studies face many challenges of design and management. Researchers must decide, for example, whether audits should be “blind” in the sense that teammates are unaware of the purpose of the study or that they have a teammate. Some scholars believe that this step is needed to protect the integrity of audit results. Discriminatory treatment is sometimes so egregious, however, that it can upset auditors to the point of compromising their ability to fill out the audit survey forms. To preserve the accuracy of the audit information, therefore, it sometimes makes sense to tell auditors the purpose of the study and train them to fill in the forms as objectively as possible, no matter what happens during the audit.
Audit results must be interpreted with care. An audit study indicates the discrimination that occurs when members of a certain minority group with certain assigned traits visit a random sample of economic agents identified through a particular sampling frame. Because discrimination may not be the same under all circumstances, however, the discrimination experienced by the average member of a group may not be the same as the differential treatment of that group in an audit study. Moreover, audit studies do not provide comprehensive measures of discrimination, but instead measure discrimination only in certain types of behavior, and they may not be not feasible for complex market transactions.
Audit studies also face challenging statistical issues. Different statistical procedures are required to measure discrimination for different types of agent behavior, for example, and statistical tests need to recognize that some factors relevant to the treatment of auditors, such as the skill or mood of the agent being audited may be shared by teammates but not observed by the researcher. In addition, audit studies must recognize that people in a legally protected minority are sometimes favored, for either systematic or random reasons. This fact leads to two different measures of the incidence of discrimination, which serve as upper and lower bounds. The gross incidence of discrimination is the share of audits in which the nonminority auditor was favored. The net incidence is the gross incidence minus the share of audits in which the minority auditor was favored. The net incidence measure is accurate if all favorable treatment of minority auditors is due to random factors, so that subtracting this favorable treatment is a correction for randomness. Minority auditors might be favored for systematic reasons, however. Systematic favoring of minority auditors is troubling in its own right, of course, but it should not be subtracted from gross incidence to determine systematic unfavorable treatment of minorities; a real estate broker who refuses to show any houses to black customers is discriminating regardless of whether or not another broker fails to show white customers any houses in black neighborhoods.
Finally, audit studies raise ethical questions because they make demands on economic agents, including many agents who do not practice discrimination. In 1982 the U.S. Supreme Court decided in Havens Realty Corporation vs. Coleman that housing audits are a legitimate investigative technique. Nevertheless, scholars conducting audits have a responsibility to make certain that their studies are well managed and do not make unnecessary demands on the economics agents being audited.
Audit studies have uncovered racial, ethnic, and gender discrimination in several countries. Recent audits studies in the United States have found discrimination against blacks and women in car sales, against blacks and Hispanics in hiring for entry-level jobs, against women in hiring for servers at expensive restaurants, and against blacks and Hispanics in preapplication mortgage procedures. The 2000 national housing audit study found, among other things, that both blacks and Hispanics receive less information about available housing units than do non-Hispanic whites in both the sales and rental markets. In the sales market, for example, white auditors but not their black teammates were shown additional units similar to the advertised unit in 22.9 percent of the audits (gross incidence), whereas black auditors were favored over their white teammates on this measure of treatment in 16.0 percent of the audits. The difference between these measures, 6.9 percent, is the net incidence of discrimination.
Not all audit studies find discrimination, however, and some audit studies find that certain types of discrimination are declining. An audit study of the home insurance industry, for example, found no clear evidence of discrimination. Moreover, a comparison of the results of the 1989 and 2000 national housing studies reveals significant declines in many types of discrimination, including several associated with housing availability. This comparison also reveals increased discrimination against blacks and Hispanics in some other types of agent behavior, including the number of housing units shown and steering, which is defined as directing different customers to different types of neighborhoods based on their race or ethnicity. Real estate agents were more likely, for example, to steer black customers away from white neighborhoods in 2000 than in 1989.
Finally, audit studies provide some insight into the causes of discrimination. One study found, for example, that the more expensive a house, the more likely it is to be withheld from black customers. Because black and white teammates in this study had the same qualifications and made the same requests, this result suggests that some real estate brokers act on the basis of a stereotype that black households cannot afford expensive houses, rather than on the basis of a customer’s stated income and wealth.
SEE ALSO Correspondence Tests; Discrimination; Inequality, Gender; Inequality, Racial
Ayres, Ian, and Peter Siegelman. 1995. Race and Gender Discrimination in Bargaining for a New Car. American Economic Review 85 (June): 304-321.
Fix, Michael, and Raymond Struyk. 1993. Clear and Convincing Evidence: Testing for Discrimination in America. Washington, DC: Urban Institute.
Ondrich, Jan, Stephen Ross, and John Yinger. 2003. Now You See It, Now You Don’t: Why Do Real Estate Agents Withhold Houses from Black Customers? Review of Economics and Statistics 85 (4): 854-873.
Riach, Peter A., and Judith Rich. 2002. Field Experiments of Discrimination in the Market Place. Economic Journal 112 (November): F480–F518.
Ross, Stephen L., and Margery Austin Turner. 2005. Housing Discrimination in Metropolitan America: Explaining Changes between 1989 and 2000. Social Forces 52 (2): 150-180.
Wissoker, Douglas A., Wendy Zimmerman, and George Galster. 1997. Testing for Discrimination in Home Insurance. Washington, DC: Urban Institute. http://www.urban.org/publications/307555.html.
Yinger, John. 1996. Closed Doors, Opportunities Lost: The Continuing Costs of Housing Discrimination. New York: Russell Sage Foundation.