An Estimate of Housing Discrimin QºmQȃ6Qȃ6 BOOKMOBIÿÿÿÑ 8 * : J Z j z Š š ª º Ê Ú ê ú * : J Z j z Š š ª º Ê Ú ê ú * : J Z j z Š š ª º Ê Ú ê ú * : J Z j z Š ~ú 7 MOBI ýé ¨\%
U.S. Department of Housing and Urban Development | Office of Policy Development and Research AN ESTIMATE OF HOUSING DISCRIMINATION AGAINST SAME-SEX COUPLES
Visit PD&R’s website www.huduser.org to find this report and others sponsored by HUD’s Office of Policy Development and Research (PD&R). Other services of HUD USER, PD&R’s research information service, include listservs, special interest reports, bimonthly publications (best practices, significant studies from other sources), access to public use databases, and a hotline (800–245–2691) for help accessing the information you need.
Prepared for U.S. Department of Housing and Urban Development Office of Policy Development and Research Prepared by M. Davis and Company, Inc. Philadelphia, Pennsylvania Samantha Friedman (University at Albany, SUNY) Angela Reynolds Susan Scovill Florence R. Brassier Ron Campbell McKenzie Ballou June 2013 AN ESTIMATE OF HOUSING DISCRIMINATION AGAINST SAME-SEX COUPLES
Acknowledgments Disclaimer Many individuals contributed to this study. The work could not have been completed without the commitment and hard work of the primary research team members and various administra- tive and support staff. We thank the following members of the project team: Gregory M. Herek; Madeline Hoffman; Thomas Roth and David Paisley of Community Marketing; Robert Schwemm; Thomas Sexton; Gregory D. Squires; Bryan Tomlin; and Donald Kahl and John Baker of the Equal Rights Center, who also made contributions to the design of the project. A panel of expert advisers made valuable contributions to the understanding of the housing issues faced by the lesbian, gay, bisexual, transgender community and of the e-mail testing approach. We express special thanks to Gregory M. Herek, Jim McCarthy, Robert Schwemm, and Gregory D. Squires; as well as Lee Badgett, Roger Doughty, Kate Kendall, Jerry Malitz, Justin Nelson, Terry Stone, and Laura Young. We also thank Jennifer A. Stoloff, the project’s Government Technical Representative from the U.S. Department of Housing and Urban Development, Office of Policy Development and Research, for providing significant guidance and support at all stages. We offer special thanks to Louisa Elfman of M. Davis and Company, Inc., for managing the packaging of the report. The contents of this report are the views of the contractor and do not necessarily reflect the views or policies of the U.S. Department of Housing and Urban Development or the U.S. government.
An Estimate of Housing Discrimination Against Same-Sex Couples iv Contents Acknowledgments .................................................................. ii Foreword ............................................................................... iii Executive Summary .................................................................v Introduction ........................................................................... 1 Fair Housing Laws .......................................................... 1 Paired Testing and Previous Housing Discrimination Studies ..................................................................... 2 Previous Research on Discrimination Against LGBT People ...................................................................... 3 Research Design ..................................................................... 5 Hypotheses ..................................................................... 5 E-mail Approach ............................................................. 5 Methodology ............................................... ........................... 6 Selecting Markets ............................................................ 6 Random Sample of Markets ............................................. 6 Sampling Strategy ........................................................... 7 Conducting the E-mail Testing ........................................ 9 Data Analysis ........................................................................ 12 Gross and Net Measures of Adverse Treatment .............. 12 Findings ............................................................................... 14 National-Level Estimates of Discrimination ................... 14 Estimates of Discrimination by Legislative Jurisdiction .. 16 Estimates of Discrimination by Metropolitan Market Size ........................................................................ 18 Conclusions and Implications .............................................. 20 Recommendations for Further Study ............................. 21 Technical Appendixes ........................................................... 22 Appendix A. Distribution of Population and Sample Characteristics Across Proposed Population Stratum .................................................................. 22 Appendix B. Weighted Results and Standard Errors ...... 23 Appendix C. List of Tester Names ................................. 24 Appendix D. E-mail Text and Subject Lines .................. 24 Appendix E. Data Dictionary ......................................... 27 References ............................................................................ 69 List of Tables Table E-1. Tests Favoring Heterosexual Couples (gross measures) ............................................................vii Table E-2. Tests Favoring Heterosexual Couples (net measures) ............................................................vii Table 1. Number of Paired Tests in Each Stratum ............... 7 Table 2. Outcome Variables.............................................. 14 Table 3. National-Level Gay Male-Heterosexual Tests ....... 14 Table 4. National-Level Lesbian-Heterosexual Tests ............ 15 Table 5. Gay Male-Heterosexual Tests by State Protection ... 16 Table 6. Lesbian-Heterosexual Tests by State Protection ...... 17 Table 7. Gay Male-Heterosexual Net Measures by Metropolitan Market Size ....................................... 18 Table 8. Lesbian-Heterosexual Net Measures by Metropolitan Market Size ....................................... 19 List of Figures Figure E-1. National-Level Adverse Treatment Against Gay Male and Lesbian Couples, 2011 ................viii
An Estimate of Housing Discrimination Against Same-Sex Couples v Executive Summary This is the first large-scale, paired-testing study to assess hous- ing discrimination against same-sex couples in metropolitan rental markets via advertisements on the Internet. The research is based on 6,833 e-mail correspondence tests conducted in 50 metropolitan markets across the United States from June through October 2011. For each correspondence test, two e-mails were sent to the housing provider, each inquiring about the availability of the unit advertised on the Internet. The only difference between the two e-mails was the sexual orientation of the couple making the inquiry. Two sets of correspondence tests were conducted, one assessing the treatment of gay male couples relative to heterosexual couples and one assessing the treatment of lesbian couples relative to heterosexual couples. This methodology provides the first direct evidence of discrimi- natory treatment of same-sex couples compared with the treat - ment of heterosexual couples when searching for rental housing advertised on the Internet in the United States. The study finds that same-sex couples experience less favor- able treatment than heterosexual couples in the online rental housing market. The primary form of adverse treatment is that same-sex couples receive significantly fewer re sponses to e-mail inquiries about advertised units than heterosexual couples. Study results in jurisdictions with state-level protections against housing discrimination on the basis of sexual orientation un- expectedly show slightly more adverse treatment of same-sex couples than results in jurisdictions without such protections. This study provides an important initial observation of dis- crimination based on sexual orientation at the threshold stage of the rental transaction and is a point of departure for future research on housing discrimination against same-sex couples. Background Federal fair housing laws, seeking to ensure equal access to housing, prohibit housing discrimination based on race, color, religion, national origin, sex, familial status, and disability. The Fair Housing Act, however, does not include sexual orientation or gender identity as protected classes. Although individual states and localities increasingly include sexual orientation and gender identity as protected classes, the fair housing laws of most states do not provide legal protections for the lesbian, gay, bisexual, and transgender (LGBT) community. Although various studies have gathered information on preju- dice and stigma against the LGBT community, little empirical research has examined housing discrimination based on sexual orientation. In community-based surveys conducted during the 1980s and 1990s with nonprobability samples, many lesbians and gay men reported that they had experienced some form of housing discrimination. In a statewide survey of lesbians and gay men by the Philadelphia Lesbian and Gay Task Force, between 9 and 16 percent of males (depending on race) and between 5 and 11 percent of females reported housing discrimination (Gross and Aurand, 1996). In a 2000 Kaiser Family Foundation survey, 11 percent of the lesbian, gay male, and bisexual respon - dents said they had personally experienced discrimination in renting an apartment or buying a home. Another 35 percent said they had not personally experienced such discrimination but knew someone who had (Kaiser Family Foundation, 2001). Using data from a nationally representative sample of lesbian, gay male, and bisexual adults, Herek (2009a) found that 3.8 percent of this population reports experiencing discrimination in the housing market at least once, with gay men experiencing the highest rate of housing discrimination (6.5 percent). Data on perceptions can miss discriminatory actions that are unknown to prospective renters, however. Recently, three cor- respondence test studies examined potential adverse treatment of lesbian and gay male couples, relative to heterosexual couples: two in Sweden (Ahmed, Andersson, and Hammarstedt, 2008; Ahmed and Hammarstedt, 2009) and one in Canada (Lauster and Easterbrook, 2011). Ahmed and colleagues did not find evidence of adverse treatment against lesbians, but they did find significant differences between gay male couples and heterosexual couples, with gay male couples receiving fewer responses and invitations to contact the provider and inspect the unit. Regarding a “gross” measure of adverse treatment, Ahmed and Hammarstedt (2009) found that in 12.3 percent of the correspondence tests (not matched pairs), heterosexual couples were favored over gay male couples in getting an e-mail response; the net measure was 11.4 percent and was the only dimension of adverse treatment that was statistically significant. No disparities emerged in invitations to contact the provider or to a showing of the unit. Lauster and Easterbrook (2011) also found no disparity between lesbian and heterosexual couples but found that gay male couples are less likely than hetero- sexual couples to receive positive responses from housing providers. No correspondence tests, however, were previously conducted for these groups in the United States.
An Estimate of Housing Discrimination Against Same-Sex Couples vi Objectives of the Research The objective of this study is to develop the first national estimate of the level of housing discrimination based on sexual orientation; that is, discrimination against same-sex couples— men partnering with men and women partnering with women—at the initial stage of the rental housing transaction in the electronically advertised rental market. The study looks only at the issue of the sexual orientation of same-sex couples and not at other issues, such as gender identity. The study has two unique features. First, it examines the ex- perience and treatment of same-sex couples in their search for rental housing, a subject not previously observed on a national scale. Second, recognizing the increasing use of the Internet to search for housing, the study uses Internet advertising and matched-pair e-mails—the very threshold of the housing transaction—as the point of contact between the tester and the housing provider. The project also provides a novel, yet increasingly relevant, approach to a first look at barriers in the rental housing market for same-sex couples. Hypotheses Consistent with the findings of previous research outside the United States (for example, Ahmed, Andersson, and Ham- marstedt, 2008; Ahmed and Hammarstedt, 2009; Lauster and Easterbrook, 2011), we expect that a disparity would exist in the response of housing providers to inquiries expressing inter- est in electronically advertised rental housing by heterosexual and same-sex couples. The main hypotheses for the study are that (1) same-sex couples will experience more adverse treatment than heterosexual couples, (2) gay male couples will experience a greater degree of adverse treatment than lesbian couples, and (3) same-sex couples will experience lower levels of adverse treatment in places with state-level housing discrimi- nation laws inclusive of sexual orientation than in jurisdictions without such protections. Methodology The research adapts the well-established matched-pair testing methodology, which has been a hallmark of previous U.S. Department of Housing and Urban Development (HUD) housing discrimination studies (HDSs), for use in examining the electronically advertised rental housing market. A total of 6,833 matched-pair correspondence tests were completed via e-mail across 50 markets. Tests were divided between those examining discrimination against gay men (3,424 tests) and those examining discrimination against lesbians (3,409 tests), both relative to the treatment of heterosexual couples. The primary objectives of the study were to obtain data that would produce (1) nationally representative estimates of various measures of housing discrimination against same-sex couples; (2) to the extent possible, estimates of these measures by whether a state had legislative protections against housing discrimination based on sexual orientation; and (3) to the extent possible, estimates of these measures by market size. A total of 50 markets were randomly selected, proportional to population size (PPS), from among the 331 metropolitan statistical areas or primary metropolitan statistical areas, based on 2000 census definitions. The sampling elements were one-bedroom rental unit listings advertised on a national Internet listing site. This site was chosen as the universe from which to sample the electronic advertisements because the first contact between the prospective renter and the housing provider could always occur via e-mail; unlike other electronic search engines, the site does not require prospective renters to complete an online registration form asking for their phone numbers and current addresses; and because the format of the advertisements on the selected site and the nature of the contact between providers and prospective tenants is uniform throughout the country. For a market to be included in the sample, it had to have complete coverage on the selected listing site throughout the metropolitan area being sampled. For example, for the Wash- ington, D.C. market to be included in our sample, the range of advertised units had to be spread among different areas throughout the metropolitan area (for example, Fairfax County, Virginia; Prince George’s County, Maryland; the District of Co- lumbia; and so on). If the listing site did not completely cover a selected market, it was not included in the sampling frame, and another market was randomly selected using a PPS sampling approach. This procedure ensured a final sample of 50 markets with complete coverage. Each correspondence test involved sending two e-mails to the housing provider, each inquiring about the availability of the electronically advertised unit. The only difference between the two e-mails was whether the couple was same sex or heterosex- ual. Unfavorable treatment was measured based on the housing provider’s response to the e-mail, with the central focus being on whether each tester (1) received a response, (2) received more than one response, (3) was told the unit was available, (4) was told to contact the provider, and (5) was invited to inspect the unit. Each correspondence test resulted in one of three potential outcomes: (1) the heterosexual couple is favored over the same-sex couple, (2) the same-sex couple is favored over the
An Estimate of Housing Discrimination Against Same-Sex Couples vii heterosexual couple, or (3) both couples receive equivalent treatment (equally favored or disfavored). The most straight- forward measure, the gross measure, is the percentage of tests in which the heterosexual couple is favored over the same-sex couple. Gross measures are considered upper bound estimates of discrimination. Differential treatment might occur for ran- dom reasons, as well as reasons that have nothing to do with actual discrimination. For example, the housing provider might simply have forgotten to reply to the same-sex couple, or per- haps the unit was truly already rented by the time the same-sex couple inquired about it. To produce lower bound estimates of discrimination, net measures are calculated, borrowing from the methodology of the 2000 Housing Discrimination Study (HDS2000). Net measures subtract the percentage of same-sex couples favored on a given outcome from the percentage of heterosexual couples favored. The true estimate of adverse treatment against same-sex couples probably lies between the upper and lower bound estimates. This report presents results for the five key dimensions of treatment discussed previously and combines these dimensions to create a composite measure of treatment. In particular, the consistency index, adopted from HDS2000, reflects the extent to which one tester is consistently favored over the other in the treatment received from housing providers based on their inquiry e-mails. Tests are classified as “heterosexual favored” if the heterosexual couple received favorable treatment on at least one of the five dimensions and the same-sex couple (gay male or lesbian) received no favorable treatment. Tests are classified as “gay male or lesbian favored” if the same-sex couple received favorable treatment on at least one of the five dimensions and the heterosexual couple received no favorable treatment. Findings Same-sex couples are significantly less likely than heterosexual couples to get favorable responses to e-mail inquiries about electronically advertised rental housing. Comparing our gross measures of discrimination, heterosexual couples were favored over gay male couples in 15.9 percent of tests and over lesbian couples in 15.6 percent of tests (Table E-1). The net measures indicate that heterosexual couples are sig nif - icantly more likely than their gay male and lesbian counterparts to receive an initial e-mail response (Table E-2). At this prelim- inary stage of the rental housing transaction, barriers i ndicate a rejection of the tester based solely on the sexual orientation information provided in the e-mail rather than on any charac- teristics related to qualification for the housing, thus preventing basic access to rental units. Key Findings • Same-sex couples experience discrimination in the online rental housing market, relative to heterosexual couples (Figure E-1). • Adverse treatment is found primarily in the form of same-sex couples receiving fewer responses to the e-mail inquiry than heterosexual couples. • Overall, results in states with legislative protections show slightly more adverse treatment for gay men and lesbians than results in states without protections. • Adverse treatment of same-sex couples is present in all metropolitan areas, but no clearcut pattern exists in the magnitude of adverse treatment by metropolitan market size. • Lower bound measures of discrimination (net measures) re- veal similar results, although the magnitude of the difference in treatment between heterosexual and same-sex couples is less (that is, 2.2 percent for the gay male-heterosexual tests; 1.3 percent for the lesbian-heterosexual tests) than for the gross measures and is only statistically significant in the gay male-heterosexual tests. Table E-1. Tests Favoring Heterosexual Couples (gross measures) Gross Measures Tests Favoring Heterosexual Couples Versus ... Gay Male Couples (%) Lesbian Couples (%) Consistency index 15.9 15.6 Initial response provided 11.6 11.2 Table E-2. Tests Favoring Heterosexual Couples (net measures) Net Measures Tests Favoring Heterosexual Couples Versus ... Gay Male Couples (%) Lesbian Couples (%) Consistency index 2.2* 1.3 Initial response provided 3.1** 2.3** * p ≤ .05. ** p ≤ .01.
An Estimate of Housing Discrimination Against Same-Sex Couples viii Figure E-1. National-Level Adverse Treatment Against Gay Male and Lesbian Couples, 2011 15.9% 13.7% 2.2* 15.6% 14.3% 1.3% 11.6% 8.5% 3.1** 11.2% 8.9% 2.3** Heterosexual favored Gay male favored Gay male-heterosexual tests Net measure Heterosexual favored Lesbian favored Lesbian-heterosexual tests Net measure Consistency index Reponse provided * Significant at the p ≤ .05 level. **Significant at the p ≤ .01 level. Effect of Legislative Protections In states with legislative protections against housing discrimina- tion based on sexual orientation, heterosexual couples were consistently favored over gay male couples in 16.0 percent of tests and were favored over lesbian couples in 15.9 percent of tests. In states without such protections, however, heterosexual couples were favored over gay male and lesbian couples at rates that were 0.6 percentage points less than those in protected states (that is, 15.4 and 15.3 percent, respectively). Moreover, the net measure for gay male couples relative to heterosexual couples (3.1 percent) was statistically significant only in juris- dictions with state-level protections. Taken together, those results are surprising in that states with legislative protections prohibiting housing discrimination on the basis of sexual ori- entation do not show lower levels of adverse treatment. Several factors could account for this unexpected finding, including potentially low levels of enforcement, housing provider un- familiarity with state-level protections, or the possibility that protections exist in states with the greatest need for them. Effect of Metropolitan Market Size When disaggregated by the size of metropolitan areas, the results of the paired tests continue to reveal that heterosexual couples were consistently favored over gay male and lesbian couples, although variation exists in the magnitude of the differences in treatment. These differences exhibit no clearcut pattern by met- ropolitan market size, however. The net measure is statistically significant only in gauging the treatment of gay male couples relative to heterosexual couples in the largest metropolitan areas. Conclusions and Implications The findings presented in this report provide evi dence that discrimination exists against same-sex couples in the initial stages of the search for electronically advertised rental housing in metropolitan America. The study measured the response of housing providers regarding the sexual orientation of couples and did not examine other characteristics, such as gender iden- tity. The adverse treatment of same-sex couples stems largely from the fact that housing providers are less likely to respond to same-sex couples than to heterosexual couples. This study employed paired e-mail correspondence tests rather than in-person, paired tests, and it allowed for only one e-mail interaction with each housing provider. Because the observations are at the very threshold of the rental transaction, the estimates of discrimination presented here likely underestimate the extent to which heterosexual couples are favored over same-sex couples in the rental housing market. Nevertheless, the incidence of consistently favored treatment of heterosexual couples relative to gay male and lesbian couples (that is, 15.9 and 15.6 per- cent, respectively) is similar in magnitude to the incidence of
An Estimate of Housing Discrimination Against Same-Sex Couples ix consistently favored treatment of White homeseekers relative to Black and Hispanic homeseekers (that is, 21.6 and 25.7 per- cent, respectively) found using in-person audits in HDS2000. This study serves as the initial step toward future research on same-sex housing discrimination. Although its use of paired testing and its national scope are strengths, the study design is limited to e-mail tests of rental housing in metropolitan markets advertised by one source on the Internet. Moreover, the study captures the treatment of test e-mail inquiries by housing providers in response to only one e-mail sent by each tester, and it therefore does not consider what could happen to testers through additional contact (via additional followup e-mails, phone, or in-person communication). Thus, the testing conducted in this study is representative of the initial stage of the housing search by prospective renters in the metropolitan rental housing market. The contribution of this study is to demonstrate that same-sex couples are less likely than heterosexual couples to gain access to the targeted rental unit. When same-sex couples do receive a response, however, the treatment by housing providers is, for the most part, equal—at least for a single e-mail interaction. This type of discrimination diverges somewhat from that which has been found between Whites and non-White minorities (Friedman, Squires, and Galvan, 2010). In large part, the dis- parity between Whites and non-White minorities in the initial access to housing units (that is, getting a response from provid- ers) is less than the disparity observed in additional contact with providers (for example, getting more than one response or the potential to inspect the unit). This first set of national findings on the discrimination against same-sex couples in the metropolitan rental housing market should serve as a point of departure for future research on same-sex housing discrimination. In-person testing would provide valuable, additional information on the experiences of same-sex couples in the rental market, and it would replicate the approach of other HDS research and track the real-life sequence of a rental housing search. A broader sample of ad- vertised rental units, including other electronic media and print advertisements, could be used for in-person testing. Future studies could employ e-mail or in-person audits to further examine differences in treatment between same-sex and heterosexual couples in states with and without legislative pro- tections for sexual orientation or gender identity. Local jurisdic- tions’ protections were not within the scope of this study, and examining the effect of such local protections on differential treatment could be very useful. In addition, this study looked at treatment based only on the sexual orientation of same-sex couples and not on gender identity or gender expression. This study does not explicitly capture treatment of transgender people or people who do not conform to stereotypical gender norms because it used the e-mail testing approach. Finally, future research could seek to obtain metropolitan-specific estimates of discrimination against same-sex couples. Perhaps this approach could shed light on the mixed findings of rental housing discrimination revealed across metropolitan areas in this study.
An Estimate of Housing Discrimination Against Same-Sex Couples 1 Introduction This report presents the findings of the first large-scale, paired- testing study to measure treatment of same-sex couples in the electronically advertised rental housing market. The study, sponsored by the U.S. Department of Housing and Urban Development (HUD) and conducted by M. Davis and Company (MDAC), Inc., was developed to obtain a baseline national estimate of housing discrimination against same-sex couples at the initial stage of the search for rental housing. It builds on the well-established matched-pair testing method used in previous HUD housing discrimination studies (HDSs) examining racial and ethnic discrimination in the housing market. The in-person method was adapted to reflect the increased use of the Internet in the housing search. The results are based on 6,833 paired e-mail tests conducted in 50 metropolitan areas from June through October 2011. This study examines the treatment of same-sex couples seeking rental housing, a subject not previously observed on a national scale. Although the federal Fair Housing Act does not include sexual orientation or gender identity among its protected classes, evidence suggests that discrimination on the basis of actual or perceived sexual orientation limits housing opportunities for gay men and lesbians. Studies of public perception and attitudes toward sexual minorities show prejudice and stigma against this community (Herek, 2009a, 2009b). Smaller scale testing projects and other studies, discussed in the following section, indicate that same-sex couples experience prejudice based on their sexual orientation and encounter discrimination in their search for housing. State and local jurisdictions are increasingly enact- ing legislation prohibiting discrimination on the basis of sexual orientation or gender identity. Based on the existing evidence of discrimination experienced by lesbian, gay, bisexual, and transgender (LGBT) people, HUD recently issued a final rule to ensure that its core programs are open to all eligible individuals and families, regardless of sexual orientation or gender identity. 1 In addition, recognizing the increasing use of the Internet to search for housing, this study uses Internet advertising and matched-pair e-mails—the threshold of the housing transac- tion—as the point of contact between the tester and the hous- ing provider. In each paired test, the housing provider received two e-mails inquiring into an advertised rental unit: one from a self-identified same-sex couple and one from a self-identified heterosexual couple. Fair Housing Laws The federal Fair Housing Act (Title VIII of the Civil Rights Act of 1968) prohibits discrimination in the sale, rental, and financ- ing of housing or in other housing-related transactions on the basis of seven protected classes: race, color, religion, national origin, sex, familial status (the presence of children less than 18 years old, seeking custody of such people, or being pregnant), and disability. For example, housing providers are prohibited from considering these protected characteristics as the basis for rejecting or refusing to negotiate with individuals seek ing housing or housing-related services, from misrepresenting or limiting housing opportunities based on these protected char- acteristics, or in setting different terms or conditions because of these protected characteristics. Neither sexual orientation nor gender identity is a protected class under the federal Fair Housing Act, although some forms of such discrimination against LGBT people might be prohibited under currently protected classes: sex discrimination (for example, nonconformity with gender stereotypes), sexual harassment, or disability (such as a provider’s perception that a gay applicant might have HIV/AIDS). An increasing number of states and local jurisdictions have enacted legislative protections against housing discrimination based on sexual orientation or gender identity. As of early 2012, 20 states and the District of Colum- bia have enacted legislation prohibiting housing discrimination based on sexual orientation or gender identity: 2 California Nevada Colorado New Hampshire Connecticut (sexual orientation only) Delaware New Jersey (sexual orientation only) New Mexico Hawaii New York Illinois (sexual orientation only) Iowa Oregon Maine Rhode Island Maryland Vermont (sexual orientation only) Washington Massachusetts Washington, D.C. Minnesota Wisconsin (sexual orientation only) 1 “Equal Access to Housing in HUD Programs Regardless of Sexual Orientation or Gender Identity.” Final rule. Federal Register 77 (5662) February 3, 2012. Also available at http://portal.hud.gov/hudportal/documents/huddoc?id=12lgbtfinalrule.pdf. 2 See HUD’s Fair Housing Act LGBT web page at http://portal.hud.gov/hudportal/HUD?src=/program_offices/fair_housing_equal_opp/LGBT_Housing_Discrimination.
An Estimate of Housing Discrimination Against Same-Sex Couples 2 Several states and more than 240 local jurisdictions currently prohibit discrimination in employment and public accom- modations on the basis of sexual orientation. Many of these local jurisdictions also prohibit discrimination in housing on the basis of sexual orientation, gender identity, or gender expression. On February 3, 2012, HUD published its Final Rule, “Equal Access to Housing in HUD Programs Regardless of Sexual Orie ntation or Gender Identity,” which prohibits making a determination of eligibility for HUD-assisted or HUD-insured housing on the basis of sexual orientation or gender identity. The rule applies to all HUD programs, including public housing, HUD-assisted or HUD-financed housing, and FHA-insured mortgage financing. The rule has four main provisions: It (1) requires providers assisted by HUD or insured by FHA to make housing available without regard to actual or perceived sexual orientation, gender identity, or marital status; (2) clarifies that the definitions of family and household, which are integral to determining who is eligible for HUD’s core programs, includes people regardless of actual or perceived sexual orientation, gender identity, or marital status; (3) prohibits HUD-assisted and FHA-insured entities from inquiring about an applicant’s or occupant’s sexual orientation or gender identity for the purpose of determining eligibility or otherwise making housing available; and (4) prohibits FHA-approved lenders from basing eligibility determinations for FHA-insured loans on actual or perceived sexual orientation or gender identity. 3 Although less than one-half of all states provide state-level protection against housing discrimination for LGBT people, the effect of this lack of legislation remains unknown. Although most information is anecdotal, some previous formal research studies, described in the following section, illuminate the issue of housing discrimination against LGBT people. Paired Testing and Previous Housing Discrimination Studies Testing is an investigative technique used to observe the practices of housing providers. Testers, who pose as individuals seeking housing, contact housing providers in a variety of ways to in- quire about housing opportunities. The origins of paired testing as a method of studying housing discrimination and identifying differential treatment lie in fair housing enforcement, and test- ing was originally used to identify individual housing providers who were violating housing discrimination laws. HUD has used testing for more than 40 years to investigate discriminatory housing practices, because testing is a powerful tool for directly observing differences in the treatment that homeseekers experi- ence. When paired testing is applied to large, representative samples and implemented with rigorous controls, it provides reliable estimates of the differences in treatment among differ- ent populations. HUD funded three national studies using the methodology of matched-pair testing to measure the levels of housing discrimi- nation based primarily on race and ethnicity. In 1977, HUD’s Housing Market Practices Survey (HMPS1977) employed the auditor technique to observe differential treatment of Black and White homeseekers. The Black and White audi- tors were both male or both female and had approximately the same level of education, income, occupation, and family characteristics. Each member of the audit team responded separately to advertisements that were randomly selected from major newspapers, and then they recorded their treatment on standardized forms. In HMPS1977 1,609 rental and 1,655 sales audits were conducted in 40 metropolitan areas during May and June of 1977. The results revealed significant differences between the Black and White audits. The white auditor was favored 50.4 percent of the time in the sales market and 45.7 percent of the time in the rental market. The black auditor was favored only 19.7 and 17.9 percent of the time, respectively. HUD built on the HMPS1977 experience 10 years later by launching a second national audit study, the 1989 Housing Discrimination Study (HDS1989). This study involved 3,800 paired tests for discrimination against Black and Hispanic homeseekers. As in HMPS1977, both rental and sales markets were tested in a random sample of 25 major metropolitan areas. Black-White tests were conducted in 20 of these sites and Hispanic-non-Hispanic White tests were conducted in 13 sites. The HDS methodology also involved expanded sample sizes in 5 metropolitan areas, which supported in depth analysis of variations in patterns of discrimination within urban areas (Fix and Turner, 1998). In HDS1989, Black renters faced a 10.7-percent chance of being excluded altogether from housing made available to comparable White renters and a 23.5-percent chance of learning about fewer apartments. Real estate brokers were also much more likely to offer financial advice to White than to Black customers. In HUD’s third national study, Discrimination in Metropolitan Housing Markets (HDS2000), the results were based on 4,600 paired tests conducted in 23 metropolitan areas nationwide during the summer and fall of 2000. 3 http://portal.hud.gov/hudportal/documents/huddoc?id=12lgbtfinalrule.pdf.
An Estimate of Housing Discrimination Against Same-Sex Couples 3 HDS2000 Phase I was designed to provide updated national estimates of adverse treatment against Black and Hispanic home - seekers and to measure change in the incidence of differential treatment since 1989. In addition, Phase I provided estimates of adverse treatment against Black and Hispanic homeseekers in 20 individual metropolitan areas and exploratory estimates of adverse treatment against Asian (in 2 metropolitan areas) and Native American (in 1 metropolitan area) homeseekers. The basic testing protocols replicated those implemented in HDS1989. Random samples of advertised housing units were drawn weekly from major metropolitan newspapers, and testers visited the sampled office s to inquire about the availability of these advertised units. Both minority and White partners were assigned incomes, assets, and debt levels to make them equally qualified to buy or rent the advertised housing unit. Test part- ners were also assigned comparable family circumstances, job characteristics, education levels, and housing preferences. They visited sales or rental agents and systematically recorded the information and assistance they received about the advertised unit or similar units, including location, quality, condition, rent or sales price, and other terms and conditions. Test partners did not compare their experiences with one another or record any conclusions about differences in treatment; each simply reported the details of the treatment he or she experienced as an individual homeseeker. As reported in the Discrimination in Metropolitan Housing Mar- kets: National Results from Phase I HDS 2000 Final Report (HUD, 2002), researchers found that discrimination persisted in both rental and sales markets of large metropolitan areas nationwide but that its incidence had generally declined since 1989. Afri- can Americans still faced discrimination when they searched for rental housing in metropolitan markets nationwide. White renters were consistently favored over Black renters in 21.6 percent of the tests (the net measure, at 2.3 percent, was much lower). In particular, White renters were more likely to receive information about available housing units and had more op- portunities to inspect available units. Discrimination against Black renters declined between 1989 and 2000 but was not eliminated. The overall incidence of consistent White-favored treatment dropped by 4.8 percentage points, from 26.4 percent in 1989 to 21.6 percent in 2000. Hispanic renters nationwide also faced significant levels of discrimination. Non-Hispanic White renters were consistently favored in 25.7 percent of tests (the net measure was 6.1 per - cent). Specifically, non-Hispanic White renters were more likely to receive information about available housing and opportunities to inspect available units than were Hispanic renters. Discrimi- nation against Hispanic renters appeared to have remained essentially unchanged since 1989, and Hispanic renters ap- peared to face a greater incidence of discrimination than Black renters. Previous Research on Discrimination Against LGBT People None of the previous HUD housing discrimination studies in- cluded observation of the differential treatment of homeseekers on the basis of sexual orientation, and little empirical research has focused on housing discrimination against the LGBT com- munity. In community-based surveys conducted during the 1980s and 1990s with nonprobability samples, many lesbians and gay men reported that they had experienced some form of housing discrimination. For example, in a statewide survey of Pennsylvania lesbians and gay men conducted by the Philadel- phia Lesbian and Gay Task Force, housing discrimination was reported by between 9 and 16 percent of males (depending on race) and between 5 and 11 percent of females (Gross and Aurand, 1996). In a 2000 Kaiser Family Foundation survey, 11 percent of the lesbian, gay, and bisexual respondents said they had personally experienced discrimination in renting an apartment or buying a home. Another 35 percent said they had not personally experienced such discrimination but knew someone else who had (Kaiser Family Foundation, 2001). In a 2005 national survey of lesbian, gay male, and bisexual adults (Herek, 2009a), approximately 4 percent of respondents reported they had experienced some form of housing discrimi- nation because of their sexual orientation. Such discrimination was more common among gay men (reported by 6.5 percent) and lesbians (5.1 percent) than among bisexual men (nearly 2 percent) and bisexual women (1.3 percent). When interpreting these figures, it is important to remember that m any lesbian, gay male, and bisexual people refrain from revealing their sexual orientation in a variety of social situations as a way of avoiding stigma (Herek, 2009b). For this reason, evaluating the prevalence of housing discrimination would be facilitated by knowledge of the extent to which sexual minority adults have concealed their sexual orientation from potential landlords and real estate agents. A survey of more than 6,000 transgender people by the National Center for Transgender Equality and the National Gay and Lesbian Task Force Foundation (NCTE and NGLTFF, 2011) indicated significant levels of housing instability for
An Estimate of Housing Discrimination Against Same-Sex Couples 4 transgender people. Of the respondents, 26 percent reported having to find alternative places to sleep for short periods of time, 11 percent reported having been evicted, and 19 percent reported becoming homeless because of bias. Although stigma and prejudice based on sexual orientation are widespread, and employment discrimination against LGBT individuals has been well documented, little empirical research has examined housing discrimination against the LGBT com- munity in the United States. The only published studies on housing discrimination against gay men and lesbians were conducted in Sweden (Ahmed, Andersson, and Hammarstedt, 2008; Ahmed and Hammarstedt, 2009) and Canada (Lauster and Easterbrook, 2011). Ahmed and Hammarstedt (2009) e-mailed landlords who had advertised an available apartment on an Internet service comparable to the service used for this study. The e-mails were ostensibly sent by either a gay male couple or by an otherwise comparable heterosexual couple. The study found that the heterosexual couple was significantly more likely than the gay male couple to receive a response to the e-mail, to be asked to provide further information, and to be invited to an immediate showing of the apartment. Regarding a gross measure of adverse treatment, Ahmed and Hammarstedt (2009) found that heterosexual couples were favored over gay male couples in getting an e-mail response in 12.3 percent of the correspondence tests (not matched pairs); the net measure was 11.4 percentage points and was the only dimension of adverse treatment that was statistically significant. No disparities emerged in invitations to contact the provider or to a showing of the unit. Ahmed, Andersson, and Hammarstedt (2008) did not find comparable differences between the responses to a lesbian couple and a heterosexual couple. Lauster and Easterbrook (2011) also found no disparity between lesbian and heterosexual couples but found that gay male couples were less likely than heterosexual couples to receive positive responses from housing providers. The Fair Housing Centers of Michigan, which comprises four local fair housing organizations, conducted a testing audit of housing discrimination based on sexual orientation (FHC of Michigan, 2007) that found disparate treatment in 32 out of 120 (27 percent) fair housing tests it conducted. Testers posing as gay male or lesbian homeseekers received unfavorable treat- ment regarding whether housing was available, the amount of rent, application fees, and levels of encouragement compared with the treatment of testers posing as heterosexual homeseekers. The gay male and lesbian testers also were subjected to offensive comments.
An Estimate of Housing Discrimination Against Same-Sex Couples 5 Research Design The objective of this study is to develop the first national estimate of the level of housing discrimination against same- sex couples—that is, men partnering with men and women partnering with women—in the electronically advertised rental market. Hypotheses Given the findings of previous research (for example, Ahmed, Andersson, and Hammarstedt, 2008; Ahmed and Ham- marstedt, 2009; Lauster and Easterbrook, 2011), we expect that a disparity will exist in the responses of housing providers to inquiries from heterosexual and from same-sex couples expressing interest in electronically advertised rental housing. The findings of Ahmed and his colleagues and of Lauster and Easterbrook suggest that disparities will be more prevalent be- tween gay male and heterosexual couples than between lesbian and heterosexual couples. We expect that same-sex couples, and in particular gay male couples, will be less likely than heterosexual couples to receive responses from housing provid- ers, invitations to contact providers, and invitations to inspect the advertised rental unit. We expect that states with legislative protections will have lower levels of discrimination against same-sex couples than those without protections because of the enforcement mechanism in place. E-mail Approach The way people search for housing has changed dramatically in the 21st century, with much of the housing search now taking place via the Internet. It is estimated that between 43 and 90 percent of renters consult the Internet to search for housing (Frank N. Magid Associates, Inc., 2005; Horrigan, 2008; Wagner, 2008). Housing in this segment of the market has not been scrutinized as carefully as housing advertised through newspapers or other printed media, however. Although fair housing advocates have brought attention to the discriminatory nature of the content of advertisements on electronic media, little research has brought a systematic focus to how housing providers actually treat prospective tenants at the initial stages of a housing search through e-mail inquiries. A small but growing literature has begun to examine the treat- ment of protected groups in the online housing market via cor- respondence testing (for example, Ewens, Tomlin, and Wang, forthcoming; Friedman, Squires, and Galvan, 2010; Hogan and Berry, 2011). E-mails that are matched on all characteristics except those indicating the nature of the protected group (for example, race or ethnicity) are sent to housing providers in- quiring about the availability of the unit. The only characteristic of the e-mail that varies is the text used to convey the nature of the protected class (for example, names traditionally associated with a particular racial or ethnic group). The results of an online survey conducted by Community Marketing, Inc. (CMI) and MDAC in March 2010 demonstrated that, among a sample of 297 LGBT renters who had conducted a search for new housing in the previous 12 months, 63 percent used the Internet as a primary source in their search, whereas 18 percent used newspaper ads and articles (CMI and MDAC, 2010). The CMI/MDAC pulse study further indicated that 77 percent of the respondents who inquired about rental housing through the Internet used the listing site selected for this study. The other two most frequently cited websites had only about one-third as much usage, 27 and 25 percent.
An Estimate of Housing Discrimination Against Same-Sex Couples 6 Methodology This study uses an e-mail, or correspondence, test methodology to gauge the level of discrimination against same-sex couples in the rental housing market, which has not previously been ex- amined on a national scale in the United States. The researchers completed 6,833 tests via e-mail, divided between gay male couples (3,424 tests) and lesbian couples (3,409 tests). In this study, the matched pairs were the e-mails sent to hous - ing providers. Each correspondence test involved sending two e-mails to the housing provider, both inquiring about the availability of the advertised unit. Each inquiry was from a person in a committed relationship asking about an advertised one-bedroom apartment. E-mails from heterosexuals referred to a relationship partner as “husband” or “wife”; e-mails from gay men and lesbians referred to a relationship partner as “partner.” Focus groups of gay men and lesbians in dicated that the term “partner” was most frequently used to indicate a relationship. Moreover, in many states, same-sex marriage is not legally recognized; thus, the terms “husband” or “wife” for same-sex partners might be used unevenly across the country. The text of the matched-pair e-mails contained similar language asking about the housing unit for rent. The only difference between the two e-mails was the sexual orientation of the tester making the inquiry. In the case of gay male couples, male names were used by the individual signing the e-mail and for the partner of that individual. In the case of lesbian couples, female names were used. The Internet listing site used in this project was chosen as the universe from which to sample the electronic advertisements because the first contact between the prospective renter and the housing provider can always occur via e-mail. Several other electronic search engines require prospective renters to complete an online registration form asking for their phone numbers and current addresses, which would permit housing providers to contact prospective renters by phone rather than limit the contact to e-mail responses. The chosen site is also ideal because the format of the advertisements and the nature of the contact between providers and prospective tenants are uniform throughout the country. Selecting Markets The primary objective of the sampling plan was to obtain data that would (1) produce nationally representative metropolitan- market estimates of various measures of housing discrimination against same-sex couples; (2) to the extent possible, produce estimates of these measures by whether a state had legislative protections prohibiting housing discrimination on the basis of sexual orientation; and (3) to the extent possible, produce esti- mates of these measures for individual markets, by market size. Random Sample of Markets The sampling frame used a proportional-to-population-size (PPS) approach to randomly select 50 markets from among the 331 metropolitan statistical areas (MSAs) or primary met- ropolitan statistical areas (PMSAs), based on the 2000 census definitions. The selection of these areas permitted inclusion of larger and smaller markets for the sample, facilitating a more nationally representative sample. Using information available in the 2005–2009 American Com- munity Survey (ACS) 5-year estimates, we estimated the total population for each market by applying the county allocation to MSAs and PMSAs available in MABLE/Geocorr. The markets selected for inclusion in the sampling frame are those with a population of at least 100,000, which is sufficient to (1) ensure adequate coverage on the Internet listing service so that enough unique housing providers could be sampled and (2) support a systematic testing effort on same-sex couples. Selecting markets with populations of 100,000 or more substantially increases the likelihood of choosing an area with a sufficient number and percentage of same-sex couples to support the proposed num- ber of tests. 4 Of the 331 MSAs, 5 94 percent (312) had sufficient total populations to be eligible for inclusion in the sampling population. 6 The sampling process then selected markets with PPS sampling. 4 The estimated number of same-sex households is highly correlated with estimated total population size (r = 0.97). The likelihood within each stratum of selecting a market with at least 250 same-sex households (based on 2005–2009 ACS 5-year estimates) is as follows: less than 100,000, 21 percent; 100,000 to 250,000, 69 percent; 250,000 or more, 100 percent. 5 The analysis uses 2000 census definitions for MSAs and PMSAs. We estimate population and same-sex households using the 2007–2009 ACS 3-year estimates for county-level data aggregated to the MSA level. 6 The analysis eliminates six MSAs and PMSAs in Puerto Rico from the sampling frame.
An Estimate of Housing Discrimination Against Same-Sex Couples 7 Sampling Strategy Allocation of Tests Across Markets In the first stage of the sampling procedure, we classified se- lected markets by stratum, which distinguishes markets by the size of the total population. We defined various strata based on population size: insufficient population to support a systematic testing effort (less than 100,000), small (100,000 to 249,999), medium (250,000 to 399,999), medium-to-large (400,000 to 749,999), large (750,000 to 1,499,999), very large (1,500,000 to 4,999,999), and largest (5,000,000 or more). The PPS sample yields a higher representation of larger markets, thus reducing concerns about sampling adequate numbers of unique housing providers and permitting within-market exploration of housing market discrimination against gay male and lesbian couples. Achieving the targeted number of tests in the small stratum was not possible in the relatively short study period (14 weeks) without duplicating providers and running the risk of detection. Given this difficulty, we collapsed the small and medium strata to create a small-to-medium stratum and renamed the medium-to-large stratum medium. The result was five strata for market size analysis. In the second stage of the procedure, we sampled advertised rental units from online listings. After randomly selecting a market using the PPS approach, we identified whether the market was covered by the selected listing service. For a market to be included in the sample, it had to have complete coverage throughout the metropolitan area being sampled. For example, for the Washington, D.C. market to be included in our sample, the range of advertised units had to be in areas throughout the metropolitan area (for example, Fairfax County, Virginia; Prince George’s County, Maryland; the District of Columbia; and so on) and not in only one of those jurisdictions. If the listing service did not cover all of a selected market, it was not included in the sampling frame, and another market was randomly selected using PPS sampling. This process ensured a final sample of 50 markets with complete coverage. Table 1 shows the distribution of the sampled listings repre- senting unique housing providers (landlords) across population size strata. The distribution of listings sampled generally con - formed proportionally to population across strata. The goals for the size and composition of the sample were 9,100 tests across the 50 sampled markets, with the number of tests within a market divided between gay male couple-heterosexual couple matched-pair tests and lesbian couple-heterosexual couple matched-pair tests. Appendix A provides a table illustrating the distribution of the originally proposed 9,100 tests. Table 1 reflects the distribution of the actual sample of 6,833 tests. We originally chose a target of 9,100 tests to ensure a reason- able margin of error—within 1 percentage point, at 95 percent confidence—for the national gross measure of discrimination against same-sex couples. Because we could not sample dupli- cate providers and the timeframe for conducting the testing was relatively short (14 weeks), we achieved 6,833 tests across the 50 sampled markets. We estimate the overall consistency meas- ure for gay male-heterosexual tests and lesbian-heterosexual tests within 1.7 percentage points. Therefore, despite the differ- ence between the originally proposed number of tests and the number of achieved tests, our estimates of differential treatment remain within a reasonable range. One of the strategies of the data collection was to sample housing providers only once within each metropolitan area to minimize the chances that the provider would detect the testing in the study. The listing service used for this study contains multiple advertisements from the same housing providers. Some are easy to identify and others are not. For example, in housing mark ets where real estate agencies list available rental units, such agencies can post multiple ads by multiple agents within the same company. If those agents see similar e-mail messages for different available rental units, they might detect the testing, which could bias the results of the study. To minimize the chances of detection, an automated scraping tool was used to scrape, or select, the ads from each of the 50 Table 1. Number of Paired Tests in Each Stratum a Based on the population residing in micropolitan or metropolitan markets with populations of at least 100,000. Less than 1 percent of all people reside in markets with populations of 100,000 or less. Population Size Stratum Percent of Population a (%) Matched-Pair Tests Gay Male (N) Lesbian (N) Total (N) Percent of Total Tests (%) 100,000 to 400,000 (small-to-medium) 15 124 122 246 4 400,000 to 750,000 (medium) 14 250 253 503 7 750,000 to 1,500,000 (large) 14 387 373 760 11 1,500,000 to 5,000,000 (very large) 39 1,763 1,761 3,524 52 5,000,000 or more (largest) 18 900 900 1,800 26 Total 100 3,424 3,409 6,833 100
An Estimate of Housing Discrimination Against Same-Sex Couples 8 markets. The scraping tool is an existing computer software program, modified to meet the needs of this study. On a given day, the tool created a partially complete entry for each ad by extracting some data from its page. The tool then filtered the information so that the sampling frame included only one-bedroom apartments for rent. The tool also used filters to exclude ads that did not provide the address, cross street, rent, or reply-to e-mail for the unit. These scrape files were loaded daily into a Microsoft Access database. Within a market, the Access database would then screen out any new advertisement for which the housing provider was already in the database (that is, contained a duplicate phone number, address, agent or owner name, e-mail address, and so on). Files that remained in the database were considered to be unique housing providers. These files were provided at random to test administrators, who then reviewed the partial data entry (correcting if neces- sary) and completed it with the information that the scraper was not designed to extract (for example, the kind of unit, presence of discriminatory statements, and so on). The test administrator would then submit the posting to the e-mail scheduling program. The Access database would run a second check for duplicate information on the newly completed post- ing entry and, if the entry was not from a duplicate landlord, the e-mail address associated with the posting entry would have a paired e-mail test scheduled. This selection process was continued until the targeted number of listings within each market for each stratum was reached. Listings within each stratum were divided equally between gay male-heterosexual tests and lesbian-heterosexual tests. Post Sampling: Identify LGBT Protections For each listing in the final sample, we determined whether the location for the listing is subject to state-level legislative protec- tions against housing discrimination on the basis of sexual orientation. We classified listings for which the location of the housing unit is covered by such protections as “Legislatively Protected,” while we classified other listings as “Not Legisla- tively Protected.” Only state-level protections were determined; local jurisdiction protections were not identified. Although the purpose of the study was to obtain statistically valid national estimates, sample sizes within subgroupings permitted statistically valid gross or net measures of housing discrimination against gay men and lesbians for three primary factors: (1) legislative protections (covered by protection for sexual orientation or not), (2) MSA population size stratum, and (3) individual markets for metropolitan areas with popula- tions exceeding 5 million. Selection of H ousing Providers After the 50 study areas were selected, the selection of housing providers or landlords commenced. This sampling was com- pleted in two stages. The first stage used an automated scraping tool or gather ads within each of the 50 study areas for one-bedroom apartments for rent. The tool took all the ads from a given regional web- site, screened them based on criteria provided, and stored the information from the ads in a searchable database. The software was programmed to navigate through the metropolitan areas listed in the sampling frame and save each newly added rental posting (advertising one-bedroom units) to a local server. This group of listings was then scraped to extract textual informa- tion contained within each posting. This scraping process involved two different kinds of extraction: 1. Structured extraction entails identifying textual information that is entered into categorized forms at the time of the posting’s creation or generated by the website. Structured items are associated with tags the listing service creates upon generation and are therefore readily identified with simple programming techniques. Structured information includes items such as the address or cross streets of a listing, “Post- ing ID,” and the date and time of the original posting. 2. Unstructured extraction entails identifying information that is not associated with predetermined tags but is likely to be in a posting. Phone numbers and e-mail addresses are often present in the body of the text generated by the poster. The program is cued to this information by the presence of “-” (dashes), “@” (at signs), and other patterns. These cues are not 100 percent accurate but, when used in conjunction with other known language patterns (that is, the number of digits associated with a telephone number), can yield high accuracy in the extraction of data from free-form text. In the second stage, newly saved postings and the correspond- ing data that were scraped from these postings were provided daily. The data provided were from a query that was filtered to retain only one-bedroom units that had address or cross- street information. These queried datasets were loaded to the Audit-Level Database in Microsoft Access. These data did not constitute complete audit-level entries, but they were screened by the scraper tool for a minimum level of data to constitute a useable listing. Test administrators reviewed the scraped data against the original listing and completed the coding of the list- ing. These data were screened twice by the Access database for duplicate landlords (once before the test administrator’s review and coding and once after).
An Estimate of Housing Discrimination Against Same-Sex Couples 9 For each study area, a new scrape file was loaded to the Access database from the Internet listing service each day (for 6 days of a given week). The sampling of landlords took place during a 14-week period to allow for a comprehensive representation of landlords and available one-bedroom rental units within each metropolitan area. Manual processing of Internet listings in Access took place 6 days a week, although scraping took place 7 days a week (two scrape files were loaded each Monday). Conducting the E-mail Testing Overview After the markets to be sampled were selected and advertise- ments were scraped, postings from each of those markets were loaded to the database and matched-pair e-mail testing com- menced. During the 14-week data collection period, from June through October 2011, postings processed in the audit-level database were randomly assigned to matched pairs of e-mails from gay male and heterosexual couples or from lesbian and heterosexual couples. The procedures in the following list were used in executing the tests and in collecting the resulting data to minimize the risk of detection by housing providers involved in the corresponde nce tests. Procedures for Executing the E-mail Tests 1. Names. First names and e-mail accounts were created for the e-mails from prospective gay male, lesbian, and hetero- sexual renters. In addition, names were created for partners of the gay male and lesbian renters and for spouses of the heterosexual renters. Eight lists of names were created: a. Heterosexual renters inquiring about the unit who are female. b. Heterosexual renters inquiring about the unit who are male. c. Heterosexual renters’ husbands. d. Heterosexual renters’ wives. e. Lesbian renters inquiring about the unit. f. Lesbian renters’ partners. g. Gay male renters inquiring about the unit. h. Gay male renters’ partners. The names appearing on these lists came from the Social Secu- rity Administration’s website of popular baby girls’ and boys’ names. 7 The top 20 girls’ names and the top 20 boys’ names in the United States from 1970 through 1985 (between ages 25 and 40 at the time of the study; that is, individuals who were likely to be in the market for rental housing) were retrieved from this site. These boys’ and girls’ names were then consolidated into two master lists. The names on each of these master lists were then filtered to eliminate duplicates, gender-neutral names (for example, “Shannon”), and to include race-neutral names. After the master lists of male and female first names were cre- ated, within each study area the two sets of 20 names (40 total names) were randomly divided into the eight conditions of a 2 (Gender) X 2 (Sexual Orientation) X 2 (e-mail sender/sender’s partner) counterbalance. The result was eight lists of 5 names each, falling into the listed conditions, a through h, described in the preceding list. Appendix C provides the full list of male and female names. 2. E-mail Accounts. After the lists of names were completed, the e-mail addresses were developed. Each of the 20 names was randomly assigned to a Yahoo!, Hotmail, or Gmail e-mail address in each of the 50 study areas to create 1,000 unique e-mail addresses. Thus, in each study area, each name that was used to sign the e-mails from prospective renters had a unique e-mail account. We assigned an account that closely resembled the first name selected but made some modifications (for ex- ample, inserting numbers after the name) because of existing accounts. So, for example, the name “Jennifer” for heterosexual female renters could appear on the list for both the New York and Chicago metropolitan areas. Each of these Jennifers would be randomly assigned to a different e-mail account. For ex- ample, Jennifer in New York would have the e-mail account, jennifer312@gmail.com, but Jennifer in Chicago would be assigned jennifer65@yahoo.com. This one-time randomization of names to e-mail domains was critical to facilitating the track - ing of e-mails that were sent to and received from housing providers. 3. Randomization. After the names were assigned to the e-mail domains on the four lists, tables were developed in the Access database to randomize each of the following necessary elements of all the tests to be conducted: a. The names given to the testers in each test. b. The names given to the partner referred to in each e-mail text. (Note: this one-time randomization meant that the names of the renter and partner/spouse were always paired together. This procedure made the inquiries more realistic and lessened the chances of detection.) c. The e-mail subject line and text (including the greeting and closing). Appendix D reproduces the e-mail subject lines and text. d. Which e-mail was sent first to the provider. 7 http://www.ssa.gov/oact/babynames/state/index.html.
An Estimate of Housing Discrimination Against Same-Sex Couples 10 The e-mail text was developed to vary such that the wording realistically conveyed to the housing provider that the person inquiring about the unit came from a heterosexual, gay male, or lesbian couple. The remaining text in the e-mail inquiries and in the subject lines, however, was designed not to differ in any other significant ways. This methodology was employed as another way to reduce the potential for detection if a housing provider was accidentally sampled more than once. Random- izing the order in which the heterosexual and the same-sex couple e-mails were sent decreased the likelihood that the treatment of these couples was because of the order in which the e-mails were received. 4. Preparing the e-mails to be sent. Using a “mail merge” feature, all the randomized elements were combined into the messages sent to the housing providers. The text of the e-mails was constructed based on the components created in steps 2 and 3. 5. Sending the e-mails. Test administrator staff sent the e-mail inquiries to the landlord of the identified housing unit. The inquiries were made 1 day after the posting of the advertise- ment. In addition, the staff left about a 2-hour gap between the times when the e-mails were sent to the provider from each of the two parties in the matched-pair test. This procedure re- duced the likelihood that landlords or housing providers would suspect that they were being tested. Initially, the software utility package called “Letter Me Later” was used by staff to automate the process, but it was later discontinued, and an add-on for Access with the same function was used instead. This software enabled the staff to schedule the e-mails to be sent at particular times and ensured that the order in which landlords received inquiries was properly varied, with each e-mail having a 50-50 chance of being sent first. Procedures for Coding the Data The data coded for the project came from four sources: (1) the scraping tool, (2) items about the advertised unit or landlord that could not be gleaned from the scraping tool, (3) the process by which the correspondence test was conducted (the time of day the e-mail was sent, which couple sent their e-mail first, and so on), and (4) the response or nonresponse from the landlord. Also, the data gathered from sources 1 and 2 on the address of the unit were geocoded to the census tract level so that census data could be merged with the data collected in this study. Appendix E contains the complete data dictionary, and the following list presents each set of items in more detail. 1. The test administrator responsible for conducting the correspondence tests was also charged with coding the data for each test. The first set of data, on the housing unit and the landlord, came directly from the scraping tool. These data were included in an Access database for the geographic area in which the correspondence tests were conducted. Such data included the metropolitan area (area searched to find ad), subject line of the ad, posting ID, listing date, listing time, listing category (fee, no fee, by owner, “n/a”), monthly rent, size of largest image file (in number of pixels), site-specific e-mail (that is, reply-to e-mail), and information associated with the following Internet listing service tags (XXTAGS): “xstreet1” (that is, address or first cross street), “xstreet2” (that is, second cross street), “city,” “state,” “catsAreOK” (cats allowed), “dogsAreOK” (dogs allowed), “feedisclosure,” and “company name.” 2. The second set of data was coded based on other aspects of the advertisement or the landlord not directly available from the scraping tool: real estate or management company, agent name, name of owner, reference number, e-mail address (from body of text), phone numbers, street address, cross streets, city or town, ZIP code, state, additional unit information, broker fee, broker fee amount, application fee, security deposit, security dep osit amount, other fee, other fee amount, rent discount offered, description of rent discount, lease in ad, lease term, equal opportunity statement, protected class restriction, links to external URLs, full text of listing, and the presence or absence of the following key words and phrases: “equal opportunity housing,” “female,” “senior,” “section 8,” “vacation,” “per week,” “weekly,” “move in special,” “immediately,” and “before the first of the next month.” 3. The test administrator coded data in the Access database about the process by which the correspondence tests were conducted. These data included the name of the prospective renters used in the test, the names of their spouses or part- ners, the sexual orientation depicted in the e-mail, the specific e-mail text, the subject line, the e-mail domain, the order in which the e-mails were sent (that is, whose e-mail was sent to the provider first), and the time and date when the e-mail was sent. 4. The final set of data that each test administrator coded into the Access database included the response that each test e-mail received (or did not receive) from the housing provider. After test administrators conducted the correspondence test, they coded the data from the housing provider responses
An Estimate of Housing Discrimination Against Same-Sex Couples 11 as they came into the prospective renters’ e-mail inboxes during a 2-week period. The response was linked to the Access spreadsheet, and the content of the response was coded. The data collected on the responses included— a. Whether each prospective renter received a response. b. Whether they received more than one response. c. Whether they were told the advertised unit is available. d. Whether they were invited to inspect the unit. e. Whether they were advised to call the housing provider. f. Whether they were asked to provide additional informa- tion regarding their quality as an applicant (for example, their credit score or income). g. Whether they were reminded about qualifications they must possess to rent the unit. h. Whether they were given a reason for the unit not being available (if the unit was not available). i. Whether they were sent an ambiguous sign of availability (for example, “The unit is technically available, but an application has been filled out and we’re pretty sure it’s going to go through”). j. Whether they were encouraged to look at a different unit owned by the same landlord (for example, “This unit actually isn’t available, but I have another unit in the same building you might be interested in”).
An Estimate of Housing Discrimination Against Same-Sex Couples 12 Data Analysis The study provides national-level estimates of gross and net adverse treatment separately for gay male and lesbian couples, by legislative status (whether the advertised unit is in an area covered by state-level protections for gay men and lesbians) and by the size of metropolitan areas. The results provide margin-of-error estimates for the gross and net measures. Each test consisted of an e-mail inquiry from the same-sex couple and the heterosexual control couple. Results from the e-mail tests were analyzed to determine whether the gay male or lesbian prospective renter was treated unfavorably relative to the control, heterosexual prospective renter. Previous studies of housing market discrimination showed that housing providers can respond in various ways, both favorable and unfavorable. The study looked at five response outcomes, starting with whether the test e-mails received any response at all from the housing provider and continuing through increasing levels of contact and encouragement. More specifically, the outcomes included whether the test renter (1) received a response, (2) received m ore than one response, (3) was told the unit was available, (4) was invited to inspect the unit, and (5) was told to contact the provider. (Although 10 potential response variables were contemplated, only these 5 had sufficient responses to be included in the analysis.) For each test, the response to the e-mail inquiry from each couple was categorized by whether the inquiry received a favorable response. We consider a favorable response to mean a response in which the tester received affirmative values on any of the five outcomes listed previously (for example, received a response and was told the unit was available). Gross and Net Measures of Adverse Treatment The gross measure of adverse treatment is the proportion of tests in which the heterosexual control couple’s inquiry receives favor - able treatment and the gay male or lesbian couple’s inquiry receives unfavorable treatment during the housing transac- tion. Let Y ij denote the audit outcome of favorable (Y = 1) or unfavorable (Y = 0) for test i for couple j (j = 0 if heterosexual [control] and j = 1 if gay male or lesbian for test i). Differential treatment that is unfavorable to the gay male or lesbian tester will occur when Y i0 = 1, Y i1 = 0. The gross measure of adverse treatment is expressed as: Gross Measure = P[Yi0 = 1, Yi1 = 0]. We estimate the gross measure of adverse treatment at the national level for: 1. Gay male and lesbian couples. 2. State-level protections against housing discrimination based on sexual orientation. a. Markets with protections. b. Markets without protections. 3. Market size. a. Small-to-medium (100,000 ≤ population < 400,000). b. Medium (400,000 ≤ population < 750,000). c. Large (750,000 ≤ population < 1,500,000). d. Very large (1,500,000 ≤ population < 5,000,000). e. Largest (population ≥ 5,000,000). We provide estimates of the gross measure and the associ- ated margin of error at a 95-percent level of confidence. The estimated standard errors associated with the measures vary because of the differing sample sizes within each group. The corresponding margin of error is computed as: Margin of Error Gross = z c x σ n σ n , where Margin of Error Gross = z c x σ n σ n is the standard error and z c = 1.96 at 95 percent confidence. The gross measure of discrimination only estimates the likeli- hood that the control couple is favored relative to the gay male or lesbian couple and could overstate incidences of discrimina- tion. In addition to the gross measure, we compute the net measure of discrimination, the difference in the proportion of tests in which the control couple’s (that is, the heterosexual couple’s) inquiry receives a favorable outcome relative to the same-sex couple’s inquiry and the proportion of tests in which the same-sex couple’s inquiry is favored over the control couple’s inquiry. The net measure of adverse treatment is expressed as: Net Measure = P[Y i0 = 1, Y i1 = 0] - P[Y i0 = 0, Y i1 = 1]. Whereas the gross measure represents a test of one proportion, the net measure represents a test of differences in proportions; thus, the standard error and resulting margin of error estimates differ.
An Estimate of Housing Discrimination Against Same-Sex Couples 13 The margin of error estimate for the net measure is represented as: Margin of Error Gross = z c x σ p 2 – p 2 σ p 2 – p 2 , where z c = 1.96 at 95 percent confidence and Margin of Error Gross = z c x σ p 2 – p 2 σ p 2 – p 2 represents the standard error of the difference in the proportions and is expressed as: + P[Y i0 = 1, Y i1 = 0] x 1 – P[Y i0 = 1, Y i1 = 0] n 1 P[Y i0 = 0, Y i1 = 1] x 1 – P[Y i0 = 0, Y i1 = 1] n 2 . consistency index if the heterosexual couple received favorable treatment on at least one of the five outcomes measures and the corresponding same-sex couple received no favorable treatment. Tests are classified as “gay male favored” or “lesbian favored” if the same-sex couple received favorable treatment on at least one of the five outcomes and the heterosexual couple received no favorable treatment. Each of these consistency measures represents the gross measures. The net measure of the consist- ency index is calculated by taking the difference between the heterosexual-favored index and the gay male-favored or lesbian- favored index. For a matched-pair test, the number of tests for heterosexual and same-sex couples are equal (that is, n 1 = n 2 = n). The overall objective of the study was to estimate gross measures of adverse treatment of same-sex couples. We also calculate composite measures of housing discrimination against gay male and lesbian prospective renters. Similar to previous HDS stud- ies, we estimate consistency measures, which reflect the extent to which one tester is consistently favored over the other in the treatment received from housing providers based on their inquiries. Tests are classified as “heterosexual favored” on the
An Estimate of Housing Discrimination Against Same-Sex Couples 14 Findings This section presents the findings from the data analysis, focus- ing on the gross and net measures of discrimination and on the consistency indexes. First, we present the national-level estimates of the adverse treatment of gay male and lesbian couples, relative to heterosexual couples. We then calculate estimates of adverse treatment by legislative status (whether the advertised unit is in an area covered by state-level protections against housing discrimination on the basis of sexual orientation) and the size of the metropolitan area in which the unit is advertised. The data were gathered from June through October 2011. For the five dependent variables of interest, each correspondence test could result in one of four outcomes: (1) both testers received equivalent responses, (2) neither tester is favored, (3) the hetero - sexual couple is favored, or (4) the gay male or lesbian couple is favored. As discussed in the previous section, the gross meas- ure of adverse treatment refers to when the correspondence test favors heterosexual couples. The net measure takes the differ- ence between the proportion of tests favoring the heterosexual couple and the proportion of tests favoring the gay male or lesbian couple. The consistency index is a summary measure of the treatment of heterosexual couples relative to gay male and lesbian couples on all five variables. Table 2 presents the variables of interest and their definitions. Table 2. Outcome Variables Variable Definition (1) Response provided Whether each prospective renter received a response. (2) More than one response Whether they received more than one response. (3) Available Whether they were told the advertised unit was available. (4) Inspect Whether they were invited to inspect the unit. (5) Contact Whether they were advised to call the housing provider. National-Level Estimates of Discrimination Treatment of Gay Male Couples as Compared With Heterosexual Couples Table 3 presents the outcomes of the e-mail correspondence tests between gay male and heterosexual couples at the national level. Table 3 shows that, in 49.3 percent of the e-mail correspondence tests between heterosexual and gay male couples (column 1), both couples received a response from the housing provider. In 30.6 percent of the tests (column 2), neither the heterosexual couple nor the gay male couple received a response from the housing provider. Thus, in 79.9 percent of the correspondence tests, heterosexual and gay male couples received equal treatment. On the gross and net measures of adverse treatment in terms of the “response provided” v ariable (columns 3, 4, and 5), hetero - sexual couples were significantly more likely than gay male couples to get a response from housing providers. In 11.6 per- cent of the correspondence tests, only the heterosexual couple received a response from housing providers compared with the 8.5 percent of tests in which only the gay male couple received a response. The net measure (3.1 percent) was statistically significant. On the other four variables (that is, whether the tester received more than one response, was told the unit was available, was invited to inspect the unit, and was told to contact the provider), the results in columns 3 and 4 of Table 3 demonstrate that, except for the “contact” variable, a slightly greater share of tests favored heterosexual couples over gay male couples. The net measures in column 5, however, show that none of the differ- ences on these four variables is statistically significant. Table 3. National-Level Gay Male-Heterosexual Tests NA = not applicable. * p ≤ .05. ** p ≤ .01. Note: The results are weighted to normalize the population levels. Unweighted results are substantially the same and are available upon request. Outcome Percent Tester Favored Both (%) None (%) Heterosexual (%) Gay Male (%) Net Measure (%) N (1) Response provided 49.3 30.6 11.6 8.5 3.1** 3,424 (2) More than one response 6.1 88.8 2.6 2.6 0.0 1,681 (3) Available 71.9 22.0 3.5 2.6 0.9 1,681 (4) Inspect 66.9 24.1 4.7 4.3 0.4 1,681 (5) Contact 49.4 39.0 5.7 6.0 – 0.3 1,681 (6) Consistency index NA NA 15.9 13.7 2.2* 3,424
An Estimate of Housing Discrimination Against Same-Sex Couples 15 The consistency index in the last row of Table 3 mirrors the findings for each of the five outcomes analyzed separately. In 15.9 percent of the correspondence tests, the heterosexual couple was favored on at least one of the five outcomes and the gay male couple was favored on none of the outcomes. In 13.7 percent of the tests, the gay male couple was favored on at least one of the five outcomes and the heterosexual couple was favored on none of the outcomes. The difference between the two indexes, or net measure (column 5), is statistically significant. In sum, the results of the correspondence tests between hetero- sexual and gay male couples demonstrate that the most consis- tent form of adverse treatment against gay male couples is the lack of an initial e-mail response from housing providers. Gay male couples were significantly less likely than heterosexual couples to receive a response from the housing provider, a result that suggests discriminatory barriers at the very threshold of the rental housing search. No significant difference emerged between gay male and heterosexual couples regarding the other four variables, which gauge different aspects of the interaction that take place deeper into the search for rental housing. Treatment of Lesbian Couples as Compared With Heterosexual Couples Table 4 presents the outcomes of our e-mail correspondence tests between lesbian and heterosexual couples at the national level. The pattern of results is similar to that of the correspon- dence tests between gay male and heterosexual couples. Table 4 shows that, in 49.4 percent of the tests (column 1), both couples received a response from the housing provider. In 30.5 percent of the tests (column 2), neither the heterosexual couple nor the lesbian couple received a response from the housing provider. Thus, in 79.9 percent of the correspondence tests, heterosexual and lesbian couples received equal treatment. On the gross and net measures of adverse treatment in terms of the “response provided” variable (columns 3, 4, and 5), heterosexual couples clearly were significantly more likely than lesbian couples to get a response from housing providers, although the magnitude of the difference was slightly less than in the tests between heterosexual and gay male couples. In 11.2 per cent of the correspondence tests, only the heterosexual couple received a response from housing providers compared with the 8.9 percent of tests in which only the lesbian couple received a response. The net measure (2.3 percent) was statisti- cally significant but of less magnitude than the net measure on the same variable in the gay male-heterosexual tests. On the other four variables of interest, the results in columns 3 and 4 of Table 4 show that, by contrast to the gay male- heterosexual tests, the lesbian couples were favored in a slightly greater share of the tests than the heterosexual couple. The net measures in column 5 show that the “contact” variable is significant in the negative direction, indicating that the lesbian couple receives favorable treatment compared with that of the heterosexual couple. Although this result is unexpected, the magnitude is quite small (1.6 percent). On the other hand, the result for the variable “more than one response” is in the expected direction, with heterosexual couples significantly more likely to be favored than lesbian couples. With respect to the consistency measures in the last row of Table 4, in 15.6 percent of the correspondence tests, the heterosexual couple was favored on at least one of the five outcomes and the lesbian couple was favored on none of the outcomes. In 14.3 percent of the tests, the lesbian couple was favored on at least one of the five outcomes and the heterosexual couple was favored on none of the outcomes. The difference between the two indexes (column 5) is not statistically significant. In sum, the results of the correspondence tests between heterosexual and lesbian couples reveal that the most prevalent form of adverse treatment against lesbian couples is not get- ting an e-mail response from housing providers as often as heterosexual couples. Lesbian couples were significantly less likely than heterosexual couples to receive a response from the housing provider, a result that, as for gay male couples, Table 4. National-Level Lesbian-Heterosexual Tests NA = not applicable. * p ≤ .05. ** p ≤ .01. Note: The results are weighted to normalize the population levels. Unweighted results are substantially the same and are available upon request. Outcome Percent Tester Favored Both (%) None (%) Heterosexual (%) Lesbian (%) Net Measure (%) N (1) Response provided 49.4 30.5 11.2 8.9 2.3** 3,409 (2) More than one response 7.3 86.2 4.1 2.4 1.7** 1,679 (3) Available 71.7 21.3 3.4 3.6 – 0.2 1,679 (4) Inspect 67.3 24.5 3.6 4.7 – 1.1 1,679 (5) Contact 49.0 41.3 4.1 5.7 – 1.6* 1,679 (6) Consistency index NA NA 15.6 14.3 1.3 3,409
An Estimate of Housing Discrimination Against Same-Sex Couples 16 suggests discriminatory barriers at the threshold of the rental housing search. Although lesbian couples are also significantly less likely to have received more than one response from the housing provider, they are significantly more likely to be advised to contact the housing provider. This mixed pattern is worthy of further investigation. Estimates of Discrimination by Legislative Jurisdiction Effect of Legislative Protections for Gay Male Couples Table 5 presents the outcomes of our e-mail correspondence tests between gay male and heterosexual couples, disaggregated by whether the advertised unit was in a state with legislative protections against housing discrimination based on sexual orientation. The results in the table present the gross and net measures and the consistency indexes. The level of equal treat- ment is about the same as that shown in Table 3. On the gross and net measures of adverse treatment in terms of the “response provided” variable (the first row of Table 5), legislative protections appear to do little to change the overall pattern of results found in Table 3. In states both with and without protections for sexual orientation, heterosexual couples were sign ificantly more likely than gay male couples to get a response from housing providers. In protected states, in 11.6 percent of the correspondence tests, only the heterosexual couple received a response from housing providers compared with the 8.3 percent of tests in which only the gay male couple received a response. The net measure (3.2 percent) was statisti- cally significant. In unprotected states, in 11.5 percent of the tests, only the heterosexual couple received a response from housing providers, compared to 8.3 percent of tests in which only the gay male couple received a response. The net measure (3.2 percent) was statistically significant. On the other four variables of interest, the results in Table 5 indicate that, in states with protections, a slightly greater share of tests favored heterosexual couples over gay male couples. The net measures for Legislatively Protected, however, show that none of the differences on these four variables is statisti- cally significant. In unprotected states, a slightly greater share of tests favored heterosexual couples over gay male couples on the variables “available” and “inspect.” On the variables “more than one response” and “contact,” however, a slightly greater share of tests favored gay male couples over heterosexual couples. The net measures for Not Legislatively Protected show that none of these differences is statistically significant. The results for the consistency indexes in the last row of Table 5 illustrate a negative effect of legislative protections. In 16.0 per - cent of the Legislatively Protected correspondence tests, the heterosexual couple was favored on at least one of the five out comes and the gay male couple was favored on none of the outcomes. In 12.9 percent of the tests, the gay male couple was favored on at least one of the five outcomes and the heterosexual couple was favored on none of the outcomes. The difference between the two indexes, or net measure (column 4), is statisti- cally significant. The Not Legislatively Protected columns in Table 5, however, reveal that the net measure (1.6 percent) is not statistically significant in states with no state-level protections against housing discrimination based on sexual orientation. The slightly greater prevalence of gay male-favored treatment on the “more than one response” and “contact” variables likely accounts for the lack of significance of the net measure in unprotected states. The results of the correspondence tests between heterosexual and gay male couples, disaggregated by legislative protections, demonstrate that adverse treatment against gay male couples, regardless of legislative jurisdiction, consists of whether they received an initial e-mail response from housing providers. Legislative protections appear not to confer an advantage to gay male couples by protecting them from adverse treatment in this respect. Several factors could account for this unexpected Table 5. Gay Male-Heterosexual Tests by State Protection ** p ≤ .01. Outcome Percent Tester Favored Legislatively Protected Not Legislatively Protected Heterosexual (%) Gay Male (%) Net Measure (%) N Heterosexual (%) Gay Male (%) Net Measure (%) N (1) Response provided 11.6 8.3 3.2** 1,548 11.5 8.3 3.2** 1,876 (2) More than one response 2.7 2.6 0.1 780 2.8 2.9 – 0.1 901 (3) Available 3.6 2.6 1.0 780 3.4 2.6 0.9 901 (4) Inspect 4.1 3.6 0.5 780 5.4 4.4 1.0 901 (5) Contact 6.0 5.6 0.4 780 4.9 4.1 – 1.2 901 (6) Consistency index 16.0 12.9 3.1** 1,548 15.4 13.9 1.6 1,876
An Estimate of Housing Discrimination Against Same-Sex Couples 17 finding, including potentially low levels of enforcement, hous- ing provider unfamiliarity with state-level protections, or that protections exist in states with the greatest need for them. Effect of Legislative Protections for Lesbian Couples Table 6 presents the outcom es of our e-mail correspondence tests between lesbian and heterosexual couples, disaggregated by whether the advertised unit was in a state with legislative protections against housing discrimination based on sexual orientation. The results in the table present the gross and net measures and the consistency indexes. The level of equal treat- ment is about the same as that shown in Table 4. On the gross and net measures of adverse treatment in terms of the “response provided” variable (the first row of Table 6), legislative protections appear to affect the overall pattern of results found in Table 4. In protected states, the percentage of tests favoring heterosexual couples (10.8) and those favoring lesbian couples (9.2) exhibit no difference in terms of receiving a response from housing providers. The net measure (1.7 per - cent) is not statistically significant. In unprotected states, how- ever, heterosexual couples were significantly more likely than lesbian couples to get a response from housing providers. In unprotected states, in 11.3 percent of the correspondence tests, only the heterosexual couple received a response from housing providers compared with the 9.1 percent of tests in which only the lesbian couple received a response. The net measure (2.3 percent) was statistically significant. Another pattern in Table 6 that diverges from the pattern at the national level of analysis is that, in unprotected states, a slightly greater share of the correspondence tests favored heterosexual couples over lesbian couples in terms of receiving more than one response from housing providers. In 3.9 percent of the Not Legislatively Protected correspondence tests, only the hetero- sexual couple received more than one response from housing providers compared with the 1.8 percent of tests in which only the lesbian couple received more than one response. The net measure (2.1 percent) is statistically significant. In protected states, the net measure for “received more than one response” is not significant. On the other three variables of interest, the results in Table 6 show no statistically significant differences. In states with protec- tions, a slightly greater share of tests favor heterosexual couples over lesbian couples on the variables “available,” “inspect,” and “contact.” The net measures for Legislatively Protected, however, reveal that none of the differences on these three variables is statistically significant. In unprotected states, a slightly greater share of tests favor lesbian couples over heterosexual couples on the variables “available,” “inspect,” and “contact.” The net measures for Not Legislatively Protected show that none of these differences is statistically significant, however. By contrast to the results for “response provided” and “more than one response” examined individually, the results for the consistency indexes in the last row of Table 6 show no effect of legislative protections. In 15.9 percent of the Legislatively Protected correspondence tests, the heterosexual couple was favored on at least one of the five outcomes and the lesbian couple was favored on none of the outcomes. In 14.5 percent of the tests, the lesbian couple was favored on at least one of the five outcomes and the heterosexual couple was favored on none of the outcomes. The difference between the two indexes, or net measure, is not statistically significant in protected or unprotected states. In sum, the results of the correspondence tests between heterosexual and lesbian couples, disaggregated by legislative protections, reveal that adverse treatment against lesbian couples, in terms of getting an initial e-mail response from housing providers and getting more than one response, is more prevalent in states without legislative protections. Unlike the case with gay male couples, legislative pr otections do ap- pear to confer some advantages to lesbian couples in preventing them from experiencing adverse treatment, consistent with our hypothesis. Table 6. Lesbian-Heterosexual Tests by State Protection Outcome Percent Tester Favored Legislatively Protected Not Legislatively Protected Heterosexual (%) Lesbian (%) Net Measure (%) N Heterosexual (%) Lesbian (%) Net Measure (%) N * p ≤ .05. ** p ≤ .01. (1) Response provided 10.8 9.2 1.7 1,558 11.3 9.1 2.3* 1,851 (2) More than one response 3.6 3.5 0.1 796 3.9 1.8 2.1** 883 (3) Available 4.0 3.6 0.4 796 3.2 3.7 – 0.6 883 (4) Inspect 4.5 4.3 0.3 796 3.6 4.3 – 0.7 883 (5) Contact 5.3 4.4 0.9 796 4.2 6.2 – 2.0 883 (6) Consistency index 15.9 14.5 1.4 1,558 15.3 14.1 1.2 1,851
An Estimate of Housing Discrimination Against Same-Sex Couples 18 Estimates of Discrimination by Metropolitan Market Size Table 7 presents the outcomes of our e-mail correspondence tests between gay male and heterosexual couples, disaggregated by five categories of metropolitan area based on the total popu- lation: (1) small to medium (100,000 to 399,999), (2) medium (400,000 to 749,999), (3) large (750,000 to 1,499,999), (4) very large (1,500,000 to 4,999,999), and (5) largest (5,000,000 or more). For the sake of brevity, the results in the table focus on the net measures and consistency indexes. Table 7 presents no clear-cut pattern in the effect of metropoli- tan market size on the net measures of adverse treatment of gay male couples relative to heterosexual couples. The results for the “response provided” variable appear to indicate that a greater share of tests favored heterosexual couples over gay male couples. Only the net measure for the largest stratum (6.0 percent) is statistically significant, however (see column 5). In the very large stratum, a slightly greater share of tests favored heterosexual couples over gay male couples on the variables “available” and “inspect.” The net difference measures in column 4 (1.7 and 1.9 percent, respectively) are statistically significant. None of the net measures in the other categories of metropolitan market size for the other four variables is statistically significant. The results for the consistency indexes in the last row of Table 7 also do not reveal any clear-cut pattern of adverse treatment. Only in the case of the largest stratum is the net measure (4.9 percent) statistically significant. The greater prevalence of heterosexual-favored treatment on the “response provided” variable in that stratum likely accounts for the significance of the net measure for the consistency index. In sum, the results of the correspondence tests between het- erosexual and gay male couples, disaggregated by metropolitan market size, reveal that the effect of metropolitan market size on adverse treatment is not straightforward. In the largest stratum, consistent with the national-level results, the adverse treatment of gay male couples existed in the form of not getting an e-mail response (the results are insignificant for the other market sizes). In the very large stratum, heterosexual couples were significantly more likely than gay male couples to be told that the unit was available (that is, the difference is 1.7 percent- age points) and to be invited to inspect the unit (that is, the difference is 1.9 percentage points). Table 8 presents the outcomes of our e-mail correspondence tests between lesbian and heterosexual couples, disaggregated by categories of metropolitan market size. Like the results for the gay male-heterosexual correspondence tests, the results in Table 8 reveal no clear-cut pattern in the effect of metropolitan market size on the net measures of adverse treatment of lesbian couples, relative to heterosexual couples. The results for the “response provided” variable (the first row in Table 8) appear to indicate that a greater share of tests favored heterosexual couples over lesbian couples. Only the net measure for the small-to-medium stratum (8.2 percent) is statistically signifi- cant, however (see column 1). On the other four variables of interest, the results in Table 8 show that, in the large stratum, a slightly greater share of tests favored heterosexual couples over lesbian couples on the variable “more than one response.” The net measure for this variable reveals that the difference (4.5 percentage points) is statistically significant. Contrary to our expectations, within the same metropolitan market size category, the results show that a slightly greater share of tests favored lesbian couples over heterosexual couples on the variable “contact.” The net measure (-5.1 percent) for this variable is statistically significant. None of the net measures in the other categories of metropolitan mar- ket size for the other four variables is statistically significant. Consistent with the results disaggregated by legislative protec- tions, the results for the consistency indexes in the last row of Table 8 show that none of the net measures are statistically sig- nificant. Taken together, the results of the correspondence tests between heterosexual and lesbian couples, disaggregated by metropolitan market size, reveal that the effect of metropolitan market size on adverse treatment is not straightforward. Table 7. Gay Male-Heterosexual Net Measures by Metropolitan Market Size Outcome Net Measure (%) Small-to-Medium (100,000–400,000) Medium (400,000–750,000) Large (750,000–1.5 million) Very Large (1.5–5 million) Largest (5+ million) (1) Response provided 4.0 1.2 3.9 1.9 6.0** (2) More than one response 3.1 0.8 – 1.1 – 0.8 1.4 (3) Available – 3.1 0.8 1.6 1.7* – 0.2 (4) Inspect – 1.6 0.8 – 1.6 1.9* – 0.2 (5) Contact – 7.8 3.0 0.5 – 0.1 – 1.6 (6) Consistency index 0.0 2.8 2.3 1.0 4.9** Total audits/audits in which both testers received response 124/64 250/133 387/182 1,763/877 900/425 * p ≤ .05. ** p ≤ .01.
An Estimate of Housing Discrimination Against Same-Sex Couples 19 Table 8. Lesbian-Heterosexual Net Measures by Metropolitan Market Size Outcome Net Measure (%) Small-to-Medium (100,000–400,000) Medium (400,000–750,000) Large (750,000–1.5 million) Very Large (1.5–5 million) Largest (5+ million) (1) Response provided 8.2* 2.8 2.7 1.2 2.2 (2) More than one response 5.4 2.1 4.5* 0.7 – 0.2 (3) Available 1.8 1.4 – 2.3 0.0 – 0.2 (4) Inspect 3.6 – 5.7 – 2.8 – 0.5 2.6 (5) Contact – 5.4 – 3.6 – 5.1* 0.0 1.4 (6) Consistency index 8.2 0.8 0.3 0.8 2.0 Total audits/audits in which both testers received response 122/56 253/140 373/177 1,761/880 900/426 * p ≤ .05.
An Estimate of Housing Discrimination Against Same-Sex Couples 20 Conclusions and Implications This is the first national-scale, paired-testing study to assess rental housing discrimination against same-sex couples in metropolitan rental markets that selects properties via Internet advertising. In addition, the study calculates estimates of adverse treatment by state-level protections and by the size of metropolitan areas. From June through October 2011, same-sex couples experienced significant levels of adverse treatment relative to comparable heterosexual couples when they responded to electronically advertised rental housing in metropolitan rental housing markets nationwide. Our gross estimates of discrimi- nation, which reflect the extent to which heterosexual couples were consistently favored over gay male or lesbian couples, are 15.9 and 15.6 percent, respectively. These estimates from 2011, which are based on e-mail correspondence tests, are comparable to, but lower in magnitude than, the incidence of consistently White-favored treatm ent, relative to Black and Hispanic homeseekers, found through in-person audits in HDS2000 (that is, 21.6 and 25.7 percent, respectively). The net measures (or lower bound estimates of discrimination) reported in this study, as based on the consistency indexes, are 2.2 percent (gay male-heterosexual tests) and 1.3 percent (lesbian-heterosexual tests), and only the former is statistically significant. Although the magnitude of the level of discrimina- tion reflected by the net measure is considerably lower than that conveyed by the gross measure, it is nearly the same as the net measure reflecting the treatment of White prospective renters relative to Black prospective renters in HDS2000 (that is, 2.3 percent, not statistically significant). Discrimination against gay men and lesbians appears to take a relatively consistent form in the rental housing market. Adverse treatment of same-sex couples stems largely from the lower likelihood that housing providers will respond to their initial e-mail compared with the greater likelihood that heterosexual couples will receive a response. Given that this study employed the use of paired correspondence tests rather than in-person, paired tests, and given that it allowed for only one e-mail inter- action between each tester and the housing provider, the esti- mates of discrimination presented here could underestimate the extent to which heterosexual couples are favored over same-sex couples in the rental housing market. In addition, relatively few tests were conducted in smaller metropolitan areas, where some researchers hypothesize that more discrimination occurs. Nevertheless, it is helpful to know that the results of this study are reasonably consistent with the results from HDS2000, which examined racial and ethnic disparities in access to rental housing and conducted in-person tests rather than e-mail cor- respondence tests. Disaggregating the results by state-level legislative protec- tions reveals some findings that run contrary to our main expectation. The results of the correspondence tests between heterosexual and gay male couples show that adverse treat- ment against male couples in the form of getting an initial e-mail response from housing providers exists in states with and without legislative protections. Legislative protections for sexual orientation appear not to confer an immediate advantage to gay male couples in whether they experience adverse treat- ment. By contrast, for lesbian couples, the results imply that adverse treatment in the form of whether an e-mail response was received from housing providers and in whether more than one response was received is more apparent in states without legislative protections. Unlike what the results suggest for gay male couples, legislative protections do appear to confer a slight advantage to lesbian couples in preventing them from experiencing adverse treatment, a finding that is consistent with expectations. It could be that places with legislative protections have higher levels of discrimination against same-sex couples than those without protections. It is possible that a historically higher level of discrimination in such places necessitated the protections. Such protections take time to work, and it is pos- sible that disparities could be higher in places with protections because of this adjustment period, given the recent passage of many of the protections. The results of the correspondence tests between heterosexual and same-sex couples, disaggregated by metropolitan market size, are not straightforward for either gay male or lesbian cou- ples. Consistent with the national-level results, we find some adverse treatment of gay male and lesbian couples regarding their getting an initial e-mail response. For gay male couples, the adverse treatment exists only in the very large stratum of markets. For lesbians, it exists only in the small-to-medium stratum. It is not clear what accounts for this patt ern of results. Although its use of paired testing via correspondence tests and its national scope are strengths, the study design is limited to tests of rental housing that is in metropolitan markets and advertised by one source on the Internet. Moreover, it captures the treatment of testers by housing providers in response to
An Estimate of Housing Discrimination Against Same-Sex Couples 21 only one e-mail sent by each tester and, therefore, does not consider what could happen to testers through additional contacts with the housing provider (via additional followup e-mails, by phone, or in person). Thus, the testing conducted in this study is representative of only the initial stage of the housing search process. The contribution of this study is that it demonstrates that same-sex couples are significantly less likely than heterosexual couples to be able to access the targeted rental unit. When same-sex couples do receive a response, however, their treat- ment by housing providers relative to that of heterosexual couples is roughly equivalent. This type of discrimination is somewhat divergent from that which has been found between White and non-White renters. In large part, the disparity between White and non-White applicants is less in terms of receiving a response from providers or initial access to the unit than it is in terms of subsequent potential contact with provid- ers (for example, more than one response or the potential to inspect the unit). Recommendations for Further Study Because these data are the first national-scale findings about discrimination against same-sex couples in the metropolitan rental housing market, the study serves as a point of departure for future research on same-sex housing discrimination. Future research should conduct further testing, including e-mail and in-person testing, and look at a variety of specific aspects of disparate treatment that were beyond the scope of this project. This study provides an initial look at how same-sex couples are treated relative to heterosexual couples at the threshold of the rental housing search—the initial e-mail contact. It sampled only rental housing advertised on one Internet listing service. Because of the limited universe of available rental properties and the single initial contact, this study possibly underestimates the extent to which same-sex couples face discrimination in the rental housing market compared with the treatment of heterosexual couples. The study was conducted via e-mail correspondence tests, and the e-mails contained information to ensure that the housing provider knew if the couple seeking housing was of the same sex or heterosexual. In-person audits might, however, yield higher levels of discrimination against same-sex couples because providers would observe visual cues that would unequivocally establish the sexual orientation of the homeseekers and could result in discriminatory actions. Other areas for future inquiry include a deeper look at the next steps in the rental process, through follow-up e-mail commu- nication to confirm interest in the unit and, more importantly, to request an appointment; telephone contacts to confirm an appointment; and in-person visits to housing providers by testers. In-person testing would provide valuable, additional information on the experiences of same-sex couples in the rental market, replicate the approach of other HDS research, and track the real-life sequence of a rental housing search. A broader sample of advertised rental units, including other electronic media and print advertisements, could be used for in-person testing. In addition, future studies should employ e-mail or in-person audits to further examine differences in treatment of same-sex and heterosexual couples in states with and without legislative protections for sexual orientation or gender identity. Local jurisdiction protections were not within the scope of this study, a nd examining their effect on differential treatment could be very useful. In addition, this study looked only at treatment based on sexual orientation and not at treatment based on gen- der identity or gender expression, so it did not capture other forms of discrimination that LGBT people might experience. Finally, future research should seek to obtain metropolitan area-specific estimates of discrimination against same-sex couples. Perhaps this approach could shed light on the mixed findings observed across metropolitan areas in this study.
An Estimate of Housing Discrimination Against Same-Sex Couples 22 Technical Appendixes Appendix A. Distribution of Population and Sample Characteristics Across Proposed Population Stratum MSA = metropolitan statistical area. PMSA = primary metropolitan statistical area. Population Size Stratum Population Characteristics (%) Sample Characteristics (%) Percent of Total MSA/PMSA Percent of Population Percent of Same-Sex Households Market Within Stratum Percent of Sample Total Target Number of Paired Tests Within Stratum Percent of Total Target Number of Paired Tests Within Stratum Less than 100,000 5.7 0.7 0.6 0.0 — 0.0 100,000 to 250,000 37.8 8.6 7.4 5.9 180 2.0 250,000 to 400,000 14.8 6.4 5.9 11.8 720 7.9 400,000 to 750,000 18.7 13.6 12.2 21.6 1,364 15.0 750,000 to 1,500,000 9.7 14.1 13.4 13.7 1,400 15.4 1,500,000 to 5,000,000 11.5 38.6 41.6 35.3 3,636 40.0 5,000,000 or more 1.8 18.0 18.9 11.8 1,800 19.8 Total 100 100 100 100 9,100 100
An Estimate of Housing Discrimination Against Same-Sex Couples 23 Appendix B. Weighted Results and Standard Errors Weighted Computed S.E. Number of Paired Tests Percent Hetero- sexual Favored Percent Same-Sex Favored Net Measure N Gross Measure Net Measure Label in Word File Test Type S.E. MOE [95%] MOE [90%] S.E. MOE [95%] MOE [90%] p-Value (2-Tailed t Test) (1) Response provided Lesbian-heterosexual 3,409 11.2 8.9 2.3 6,818 0.004 0.007 0.006 0.007 0.014 0.012 0.002** Gay male-heterosexual 3,424 11.6 8.5 3.1 6,848 0.004 0.008 0.006 0.007 0.014 0.012 0.000** (2) More than one response Lesbian-heterosexual 1,679 4.1 2.4 1.7 3,358 0.003 0.007 0.006 0.006 0.012 0.010 0.005** Gay male-heterosexual 1,681 2.6 2.6 0.0 3,362 0.003 0.005 0.005 0.005 0.011 0.009 1.000 (3) Available Lesbian-heterosexual 1,679 3.4 3.6 – 0.2 3,358 0.003 0.006 0.005 0.006 0.012 0.010 0.753 Gay male-heterosexual 1,681 3.5 2.6 0.9 3,362 0.003 0.006 0.005 0.006 0.012 0.010 0.129 (4) Inspect Lesbian-heterosexual 1,679 3.6 4.7 – 1.1 3,358 0.003 0.006 0.005 0.007 0.013 0.011 0.110 Gay male-heterosexual 1,681 4.7 4.3 0.4 3,362 0.004 0.007 0.006 0.007 0.014 0.012 0.576 (5) Contact Lesbian-heterosexual 1,679 4.1 5.7 – 1.6 3,358 0.003 0.007 0.006 0.007 0.015 0.012 0.032* Gay male-heterosexual 1,681 5.7 6.0 – 0.3 3,362 0.004 0.008 0.007 0.008 0.016 0.013 0.711 (6) Consistency index Lesbian-heterosexual 3,409 15.6 14.3 1.3 6,818 0.004 0.009 0.007 0.009 0.017 0.014 0.132 Gay male-heterosexual 3,424 15.9 13.7 2.2 6,848 0.004 0.009 0.007 0.009 0.017 0.014 0.010* MOE = margin of error. S.E. = standard error. * p ≤ .05. ** p ≤ .01.