Testing for Discrimination in Home Insurance

FOREWORD

Residential segregation continues to be one of the most important domestic challenges facing America. Fighting discrimination against those who seek to exercise choice in where to live is a central and long-standing responsibility of the United States Department of Housing and Urban Development (HUD). It is also a key support to the National Homeownership Strategy, which seeks to dramatically raise homeownership rates for all Americans, regardless of race, color, religion, sex, familial status, disability, or national origin, by the year 2000.

Paired testing methods, wherein carefully matched individuals or properties are subject to observed and recorded treatment in the private marketplace, have emerged as powerful tools for detecting and combating housing discrimination. By measuring differences in treatment, such tests support enforcement actions. HUD has successfully used such testing methods for more than two decades. The landmark 1977 Housing Market Practices Survey and the 1991 Housing Discrimination Study both demonstrated the extent to which white, African-American, and Hispanic testers seeking to rent or buy housing - individuals comparable for ethnic background - were treated differently by housing rental and sales agents. Similar methods have been employed successfully to generate evidence of discrimination in employers' hiring practices.

Discrimination in the provision of housing insurance may have important consequences for the vitality of America's neighborhoods. Recent studies by interest groups, independent researchers, and government agencies indicate that homeowners in heavily minority neighborhoods are less likely to have private home insurance, are more likely to have policies that provide more limited coverage in case of a loss, and are likely to pay more for comparable policies. These differences in outcomes might be caused in part by discrimination in the policies or practices of the home insurance industry.

Testing for Discrimination in Home Insurance represents a useful, exploratory effort to apply testing methods to the study of such policies and practices. HUD contracted with the Urban Institute to conduct an exploratory study of neighborhood-based discrimination in the provision of quotes for home insurance. The study looked at moderate-income minority and non-Hispanic white neighborhoods in Phoenix and New York. Testers telephoned insurance agents to obtain over-the-phone or written quotes for homes that were carefully matched. Further, these properties were located in neighborhoods that were matched on a range of traits but differed in terms of racial/ethnic make-up.

This study has several key limitations which readers should bear in mind. First, it is limited to first-time home buyers in moderate-income neighborhoods in only two cities, Phoenix and New York. For these reasons, and because industry practices vary widely city-to-city, no generalizations about all homeowners or all cities should be made. Second, the researchers examined only the first stage in securing home insurance - that of obtaining a price quote for policies - and not the other stages and domains that might be subject to racial/ethnic discrimination. These include additional underwriting policies or practices that might precede policy issuance, claims adjustments (after a policy is written and a loss occurs), cancellation of policies, and even agency location.

Most of the analyses in this study did not reveal reliable evidence of discrimination by insurance agents based on the racial/ethnic mix of the study neighborhoods. However, several modest, statistically significant differences in insurance policy options and service requirements favored mostly white neighborhoods. The study also finds higher insurance prices for comparable properties in mostly Hispanic, as opposed to mostly white, neighborhoods in Phoenix, but this was in keeping with the state-approved rating territories. It is possible that the drawing of the rating territory boundaries was itself discriminatory, but that important policy issue could not be addressed in this study. Indeed, future research should focus on aspects of home insurance beyond the provision of quotes by agents, including the designation of rating territories, underwriting, and claims adjustments. In general, readers should treat this as the first rigorous foray into a domain about which we have much to learn and promising techniques to refine. This study by The Urban Institute moves us in that direction.

Finally, this study is part of a larger, long-run effort by researchers, advocates, industry professionals, and government to improve and promote discrimination testing in many areas of economic life, from employment and home buying to business development and consumer transactions. It is a small part, then, of the large effort to build one America.

Paul A. Leonard
Deputy Assistant Secretary
  for Policy Development

Executive Summary

Recent studies have shown that, compared to homeowners in predominantly white-occupied neighborhoods, homeowners in minority neighborhoods are less likely to have private home insurance, more likely to have policies that provide less coverage in case of a loss, and are likely to pay more for similar policies. This study explores one possible source of these differences by testing for discrimination on the part of home insurance agents against homebuyers in minority neighborhoods who seek insurance quotes.

Using pairs of testers posing as first-time buyers of homes in moderate-income minority and non-Hispanic white ("white" hereafter) neighborhoods, the study examines: whether a quote for insurance is provided; the type of coverage, options, and price of insurance quoted; and the quality of service provided (such as requirements for inspections and comprehensiveness of information provided). In this study we investigate agents' differential treatment (in the above categories) of insurance seekers buying their homes in moderate-income, predominantly black-or Hispanic-occupied neighborhoods in as carefully controlled a comparison as feasible. We emphasize at the start that we do not investigate possible discrimination in numerous other dimensions of the insurance business, rather we focus on the information, or "quote" for insurance, provided to those seeking insurance for a home they are in the process of buying.

Our pilot study uses "paired testing," a methodological advance for researching the issue of discrimination in the home insurance market. The approach measures the extent of differential treatment of matched pairs of testers posing as buyers of homes located in minority and white neighborhoods. In each test, two phone calls were made to the same insurance agency. One call was to obtain an insurance quote for a home in a minority neighborhood, the other for a home in a white neighborhood. The neighborhoods, homes, and insurance seekers were matched on a wide range of characteristics so that the primary difference within a paired test was whether the home was located in a minority or white neighborhood. The tested agents were chosen at random from the local metropolitan area Yellow Pages telephone directory. Comparing the outcomes of paired phone calls over a large sample of tests allows us to measure the extent of neighborhood-based differential treatment in the tested insurance market.

Our analyses are based on tests conducted in two metropolitan areas during May and June 1995, using ten pairs of matched homes and five pairs of matched testers in each site. In Phoenix, Arizona, we tested for differential treatment of Hispanic and non-Hispanic white neighborhoods. In three boroughs of New York City (Brooklyn, Queens, and the Bronx) we tested for differential treatment of black and non-Hispanic white neighborhoods. In Cleveland, Ohio, we initiated testing but terminated it after our project was discovered by the local insurance industry.

Research Methodology: Strengths and Limitations

We carefully matched testers, homes, and neighborhoods so that the primary difference between a pair of calls to an agency would be the racial or ethnic composition of the neighborhood where the homes were located. Pairs of testers were matched according to age, gender, and race/ethnicity (e.g., one pair consisted of two white female testers, another of two black male testers). Each pair of testers posed as first-time buyers of homes that also were carefully matched. Pairs of homes were matched on all major characteristics generally relevant for insurance underwriting and pricing, which were confirmed by two independent home insurance experts. The matched homes were located in census tracts with comparable socioeconomic and housing stock characteristics, subject to the approval of the State Insurance Commissioner, and, when possible, in the same insurance rating territory. (An insurance rating territory is the area in which an insurance company charges a common schedule of premiums based on losses experienced in that area.)

Several features of the research design helped ensure the comparability of matched calls to agents and the credibility of the information gathered. Matched testers phoned insurance agents about their matched homes within a three-week period, minimizing the likelihood of changes in the insurance market. All testers were trained to request basic insurance information and ask for a quote for the home they were "buying". Throughout the study the testers rotated the home pairs with which they conducted tests, so that each minority tester and each white tester inquired about insurance on homes located in both minority and white neighborhoods. Every aspect of this transaction was recorded and analyzed to determine whether there was systematic differential treatment of homes located in minority versus white neighborhoods.

The pairing of testers by race/ethnicity and sex (e.g., two white females), combined with testers of both racial or ethnic groups soliciting quotes on houses in both minority and majority neighborhoods, minimized the chances that any observed differences in coverage's and prices were the result of differences in individual tester characteristics. Our sample of more than 340 paired tests per site reduced the chances that the results were the product of chance.

The approach used in this study has important limitations, however, and the results must be interpreted within these limits. The findings are relevant to first-time homebuyers in moderate-income neighborhoods in two cities. Further research is needed to determine the extent to which the results are generalizable to other types of insurance seekers, to low-income neighborhoods, and to other U.S. cities. The study is designed to test for patterns of differential treatment in the market as a whole and does not identify individual instances of discrimination. This study also does not address the criteria used to draw rating territories or the fairness of the territories. Finally, we do not examine many other agent behaviors or insurance company practices that may be discriminatory against homeowners in minority neighborhoods, including the location of insurance offices, claims adjustments, underwriting of policies, and the cancellation of policies.

Key Findings Related to Discrimination

Our primary research goal was to test for certain insurance agent behaviors that discriminated on the basis of neighborhood racial or ethnic composition. In pursuing this goal, we focused on five critical areas: (1) whether homes in minority neighborhoods are offered a quote for insurance as often as homes in white neighborhoods; (2) whether they are offered the same types of policies (replacement cost, all-perils coverage); (3) whether they are offered the same policy options (replacement cost on contents, guaranteed replacement cost); (4) whether they are quoted comparable prices for similar policies; and (5) whether inspections are required and quality service is provided equally. Analyses of policy types, options, and premiums are based on tests in which both teammates received a quote. By using only these tests, we maintain the paired-testing design in which matched testers representing homes in majority and minority neighborhoods ask the same insurance agencies for service. In New York, incomplete information supplied by tested agents limits our ability to draw conclusions about differences in treatment in the provision of optional coverage's and to a lesser extent all-perils coverage. Key findings in these areas follow.

• Quotes. The vast majority of agents provided quotes to testers from both white and minority neighborhoods. In Phoenix, testers from white neighborhoods received quotes in 98 percent of the tests, while testers from Hispanic neighborhoods received quotes in 97 percent of the tests. In New York, testers from white and black neighborhoods received quotes in 82 percent and 78 percent of the tests. These differences of at most a few percentage points were not statistically significant.

• Policy Type. In general, insurance policy types are distinguished by two key dimensions: the method by which payment will be set in case of a loss and the sources of loss covered by the policy. We examined two features of policy types that reflect these dimensions:
  • Replacement cost coverage on the dwelling (which pays, up to a limit, the cost of rebuilding the home when it is damaged) was offered in 9 out of every 10 tests in which both teammates received a quote. Quotes that did not provide this coverage pay the market value of the home or the cost of repairing it. In New York there was a four-percentage point difference (not statistically significant) favoring white neighborhoods; in Phoenix there was no difference in rates of replacement cost quotes.

  • All-perils coverage (which covers damage to the dwelling under most circumstances) was also offered in most instances to testers from minority and majority neighborhoods. Quotes that do not provide all-perils coverage cover damage in a more limited set of circumstances. In each site, there was less than a percentage point differential (not statistically significant) in the rates at which testers from minority and majority neighborhoods were offered this coverage.

• Optional Coverage's. Out of the many options offered for additional coverage, we focused our analysis on two that are generally strongly recommended:
  • Replacement cost coverage on contents, (which pays the cost of replacing personal property in case of damage) was offered more frequently to testers from white neighborhoods. In Phoenix, this coverage was offered in 95 percent of the quoted policies for white neighborhoods and 92 percent for Hispanic neighborhoods-a statistically significant differential. In New York, the corresponding figures were 46 and 39 percent for tested white and black neighborhoods. The difference was not statistically significant. However, this may be due to the small number of tests in which we could learn whether the option was provided.

  • Guaranteed replacement cost coverage on the dwelling (which pays, with no limit, the cost of rebuilding the home when it is damaged) was offered to testers in more that 54 percent of the quoted policies. In Phoenix, testers from white neighborhoods were offered these coverage five percentage points more frequently (59 percent vs. 54 percent). In New York, testers from black neighborhoods were offered this coverage one percentage point more frequently (55 percent vs. 54 percent). Neither difference was statistically significant, although the New York findings is based on a small number of tests in which we could learn whether the option was provided.

• Premiums. In Phoenix, testers from Hispanic neighborhoods were quoted premiums that were an average of 12 percent higher than those quoted for a comparable policy in white neighborhoods. The difference was statistically significant. In New York there was no difference in the premiums charged homes in tested black and white neighborhoods.

  The interpretation of the Phoenix finding is complicated by the fact that we could not match most of our test homes within the same rating territory. For each of the seven home pairs located within the Phoenix city limits, the home in the Hispanic neighborhoods was in a higher priced rating territory that its matched home in the white neighborhoods. Our analysis shows the entire 12 percent difference in premiums reported above is accounted for by the difference in premium rates filed with the Arizona State Insurance Commission. The three Phoenix home pairs located outside the city limits were in the same rating territory and showed no premium differences

  We cannot determine whether Phoenix city rating territories were drawn in a discriminatory manner or whether the higher premiums are justified by higher losses in Hispanic neighborhoods. Nevertheless, the fact that significant differences are observed even after matching the homes and neighborhoods on a range of characteristics likely to predict loss rates suggests that further research into this issue is needed.

• Requirements to Obtain Quotes. Agents in Phoenix were three percentage points more likely to tell testers with homes in Hispanic neighborhoods that a quote "was not guaranteed without an inspection," a statistically significant difference. This was not true in New York. In neither site was there a statistically significant difference between majority and minority neighborhoods in the rate at which testers were told that an exterior inspection of the home was required in order to obtain a quote.

• Service. In New York, testers with homes in white neighborhoods were six percentage points more likely to receive both written and verbal quotes rather than verbal quotes alone, a statistically significant difference. In none of the other service measures in either site did we observe any statistically significant differences in treatment for the written or verbal format of communicating the quote, the propensity to be referred to another agent, or the explicit notation of premium discounts on the quote. For the frequency of information not mentioned in the quotes, there was 1 significant difference out of 10 measures across the two sites.

A secondary research goal was to examine discrimination on the basis of the race/ethnicity of the tester-differences in treatment of minority and white home insurance seekers as distinct from differences in treatment of neighborhoods. Because our testers had distinctive voices that typically would permit the agent to ascertain their race/ethnicity, and Hispanic testers all used identifiably Hispanic names, such an examination proved feasible. Even though tester pairs were matched on race/ethnicity, we were able to assess this form of discrimination by comparing the outcomes of tests conducted by unmatched testers of different races and ethnicity's. Our analysis revealed the following:

• White testers were not favored to a statistically significant degree over the minority testers in either test site. In Phoenix, offers of quotes and preferred policy types favored white testers by a statistically insignificant one to four percentage points. In New York, offers of quotes and preferred policy type favored black testers by a statistically insignificant two to nine percentage points.

• When results for minority testers calling about homes in minority neighborhoods were compared with those for white testers calling about homes in white neighborhoods. one statistically significant result emerged in Phoenix. Hispanic testers who received quotes on homes in Hispanic neighborhoods were eight percentage points less likely to be offered replacement cost coverage on the personal property content of their home that white testers who received quotes on homes in white neighborhoods. These comparisons revealed no other statistically significant differences in either site.

Key Findings Related to Home Insurance Industry

Although the intent of this study was to compare how insurance agents treat homebuyers in white and minority neighborhoods, an incidental outcome of the study was to gain insights into the nature of the home insurance industry. We believe these insights are noteworthy for future research and for possible policy interventions.

First, the home insurance industry operated very differently in our two sites. Although in both Phoenix and New York the top 20 firms wrote approximately 85 percent of the policies, relative to Phoenix, the New York market was distinguished by (1) a greater predominance of insurance brokers (who serve as intermediaries between consumers and insurance companies); (2) larger insurance rating territories that did not distinguish different racial or ethnic areas within boroughs; (3) in general, less willingness to provide quotes on the telephone; and (4) when quotes were provided, they involved fewer written quotes, quotes with less information, and fewer offers of policies with all-perils coverage and replacement cost coverage on personal property.

Second, agents varied considerably in the type, clarity, and amount of information provided prospective customers, both within and across sites. Some agents provided extremely detailed written information on standard forms; others provided only sketch verbal information. In the latter cases, key elements of the policy were either missing or ambiguously specified. Even with many detailed, written quotes, much of the information was coded in such a technical way that it was difficult for all but insurance professionals to decipher. This evidence suggests that bona fide insurance seekers may often find it difficult to exercise informed consumer judgements in selecting home insurance.

Third, agents do not appear to use a common method to estimate the cost of replacing a home, which plays a large part in determining both the type of policy provided and the cost of the policy. Our examination of New York agent's replacement cost estimates (provided by them on some written quotes) revealed that, on average, our test homes received replacement cost estimates that varied for 31 percent below the median replacement cost estimate to 30 percent above. In Phoenix, the average estimated varied from 16 percent below to 21 percent above. This inconsistency was similar for all test homes in both sites, regardless of the home's age or construction type or the racial or ethnic composition of the neighborhoods. Although we found no apparent bias in the high variability of replacement cost estimates; this variation is in itself an area of potential consumer concern.

Finally, we found some evidence that agents share information with one another about the insurance quotes they provide to prospective customers. This networking seemed to be the product of clustering of agents' offices in close proximity, frequency of several relatives working in the industry, and common computerized databases, both within a company and across companies. This finding has implications for future testing studies in which insurance is requested for the same homes from many different agents.

Conclusion

In this pilot study, we used paired homes and testers to compare home insurance agents' responses to phone requests for insurance for homes in majority and minority neighborhoods. The testers represented themselves as first-time homebuyers purchasing homes in moderate-income neighborhoods. We examined the provision of quotes, offers of specific policy types and options, prices of that coverage, requests made of the homebuyers, and quality of service provided.

We summarize the results in the table below. The table reports, for each site where we conducted tests, the frequency of behaviors exhibited toward testers buying a home in white neighborhoods and in minority neighborhoods, as well as the percentage point difference between these frequencies. The difference between the frequencies provides our measures of differential treatment. Asterisks denote differences that are statistically significant.

On the primary measures of whether both testers in a pair were provided a quote and whether they received comparable types of policies, we found no differences in either Phoenix or New York City. However, testers in Hispanic neighborhoods of Phoenix were quoted insurance premiums that were, on average, 12 percent more expensive that those offered to testers in white neighborhoods. Our analysis shows that this difference was produced by agents quoting rates in line with those filed with the state insurance commissioner for the different rating territories encompassing the white and Hispanic areas in which our Phoenix test homes were located. Nonetheless, this finding raised questions about whether the rating territories in Phoenix were fairly constructed on the basis of differential loss histories, an issue that is not within the scope of this research but clearly warrants further study.

We found limited evidence of discrimination directed against moderate-income minority neighborhoods in providing options on policy quotes. In Phoenix, homeowners in Hispanic neighborhoods were four percentage points less likely to be offered the option of replacement cost coverage on personal property contents of the home that homeowners in white neighborhoods.

There was no statistically significant difference in offers of the guaranteed replacement cost option in Phoenix. In New York City, we found no statistically significant evidence of discrimination in the provision of policy options, but this conclusion is limited because of incomplete information.

In Phoenix, testers buying homes in Hispanic neighborhoods were three percentage points more likely to be told that the quote they received would not be guaranteed without a home inspection-a small but potentially important difference if the inspection would have kept them from receiving insurance. In New York, the only indicator of discrimination was that testers with homes in white areas received both written and verbal quotes and their teammates in black areas only received verbal quotes in 18 percent of the cases, whereas the converse occurred in only 12 percent of the cases, a six percentage point differential. However, this difference did not result in testers with homes in black neighborhoods receiving less information than those did with homes in white neighborhoods.

Overall, of the 20 indications (10 for each city) of potential differential treatment analyzed, one in New York City and three others in Phoenix indicated statistically significant differences that were consistent with the hypothesis of discrimination. In addition, out of 10 measures (across the two cities) of information not mentioned in the quotes, one indicated a statistically significant difference. Although these differences warrant further investigation they do not indicate systematic discrimination by agents directed against first-time homebuyers in moderate-income minority neighborhoods in the provision of home insurance quotes measured across a wide range of indicators.

Our study was designed to assess whether any significant differences in treatment were revealed by the market as a whole, not whether an agent discriminated in any individual instance. Thus, these findings do not mean that there are no instances of discrimination regarding these or other behaviors in the home insurance market.

The findings suggest, however, that discrimination on the part of agents as they respond to phone requests for quotes does little to explain the findings of earlier studies that homeowners in minority neighborhoods are less likely to have private insurance and more likely to have policies that provide less coverage in case of a loss. On the other hand, if the premium differences we observed in Phoenix were replicated in other cities and neighborhood types, they may help explain why homeowners in minority neighborhoods have been shown to pay more for similar policies.

In closing, we reiterate that our findings are most applicable to understanding the provision of quotes by insurance agents in Phoenix and New York to first-time homebuyers from moderate-income neighborhoods. Further testing is needed to understand the generalizability of these findings to other cities, to low-income neighborhoods, and under other insurance purchasing scenarios. Future analyses should expand upon the existing body of research, including the location of agencies; the role of race in underwriting, claims adjustment, and cancellation of policies; and the design of rating territories, as well as practices not yet investigated.

Order Testing for Discrimination in Home Insurance using HUD USER's online order service. Copies may also be obtained by calling HUD USER at 1-800-245-2691.