Attitudinal and Behavioral Indicators of Discrimination
Thus far we have discussed experimental and nonexperimental approaches to drawing causal inferences about the presence and effects of racial discrimination in one or more domains. As discussed in the previous chapter, a primary challenge to researchers attempting to draw such inferences is the inability to identify and control for all the characteristics that might affect an outcome. In this chapter, we consider the challenges of collecting direct evidence of the incidence, causes, and consequences of discrimination from sample surveys, governmental administrative data, nongovernmental data, and in-depth interviews. We also review in some detail the scale measures of racial attitudes used in surveys.
Survey and administrative records data do not provide direct observations of actual discrimination, but they can measure reported experiences, perceptions, and attitudes that involve discrimination. Used in addition to statistical methods for drawing causal inferences, these data can help researchers better understand the nature and likely occurrence of racial discrimination.
Several parts of this chapter are drawn from a paper commissioned by the panel by Thomas Smith, director of the General Social Survey (GSS), on various approaches to measuring racial discrimination and their strengths and weaknesses (Smith, 2002). The approaches he assesses include administrative counts of discriminatory incidents reported to governmental and nongovernmental authorities, in-depth interviews about past experiences with discrimination, and survey-based studies in which representative samples of randomly selected respondents report on their experiences with discrimination and prejudice. Some approaches (e.g., large-scale surveys)
assess aggregate knowledge about the incidence of discrimination, while other approaches (e.g., information obtained from in-depth interviews) assess knowledge of such occurrences in a particular instance.
THE CHALLENGE OF DIRECT MEASUREMENT OF DISCRIMINATION
Prior to 1964, many forms of racial discrimination were legal (see also Chapter 3). Although a few states had passed civil rights laws, there was no national prohibition on discriminatory treatment. Many Americans were regularly denied access to jobs, housing, accommodations, and services on the basis of their race or ethnicity. Under these circumstances, the incidence, causes, and consequences of discrimination could be assessed by surveying employers, landlords, and business establishments about their racial policies or by surveying minority group members about the frequency and circumstances under which they were denied access to jobs, housing, or accommodations because of their race.1
The civil rights legislation passed in 1964 made open discrimination on the basis of race or ethnicity illegal, and perpetrators could be prosecuted under both criminal and civil law. In succeeding years, overt discrimination was prohibited progressively in various markets in the United States—first in markets for labor, goods, and services (in 1964); later in housing markets (in 1968); and finally in mortgage lending (in 1974, 1975, and 1977). Overtly denying nonwhites a job or apartment because of their race would invite prosecution and, if admitted in court, would virtually guarantee conviction.
Although readily observable acts of discrimination declined, the persistence of high levels of residential segregation along racial lines and large racial gaps with respect to income, wealth, and other societal outcomes indicate the continued existence of racial discrimination, at least at some level. Rather than being open and readily observable, however, discrimination was more often subtle in nature, assuming new forms that are not as easily identified but may be damaging nonetheless.
Under these circumstances, assessing the extent of racial discrimination directly from observational studies, such as asking black Americans whether they have been denied housing or employment because of their race, can be complicated. Those surveyed may suspect that they have been victims of discrimination, but ambiguity in such a situation can make proving discrimination—or sometimes even recognizing it—difficult. On the one hand, the true incidence of discrimination may be underestimated because a per-
son may not realize that it has occurred or be uncertain about whether it was actually present. On the other hand, the true incidence of discrimination may be overstated because, in an ambiguous situation, respondents may falsely attribute the denial of work or housing to discrimination that is in fact due to some other reason, such as qualifications, timing, or even chance. Likewise, asking white Americans whether they intend to discriminate or whether they support discriminatory policies is unlikely to provide a good indication of the prevalence of racial discrimination in American society (e.g., see Bonilla-Silva and Forman, 2000). Respondents to social surveys are understandably reluctant to admit to socially sanctioned behavior, particularly if it is illegal (Sudman and Bradburn, 1982). Thus, whites will tend to underreport discriminatory attitudes and behaviors and be unwilling to admit to perceptions or feelings that might appear to be prejudiced (a phenomenon known as social desirability bias).
Much of the information on racial attitudes and perceptions of discrimination comes from sample surveys and other observational studies (as distinct from experimental studies). It is therefore important to be cognizant of what such data sources can and cannot do. Survey measures capture reported experiences, perceptions, or attitudes about discrimination. These measures may provide some indication that discrimination is occurring, as well as what its causes and consequences may be. Nonetheless, evidence from surveys and other observational studies is not statistically valid in terms of causality. In such studies, it is not discrimination per se that is being directly measured but reports of experiences of discrimination or discriminatory attitudes.
In Chapter 7, we discussed the difficulty of using observational data to assess the extent to which racial differences in outcomes are the result of discrimination, as opposed to some other observed or unobserved variable that is correlated with race. Unknown characteristics or process-related differences in treatments (e.g., processes occurring before a reported incident) can bias observed effects. Because only a small set of characteristics is often available in any data set, other unobserved characteristics that may affect the outcome of interest are omitted, biasing the observed effect of racial discrimination. Thus, we again see how lack of experimental control can hamper the ability to draw causal inferences from observational data.
To the extent that members of disadvantaged racial groups report being discriminated against and whites admit to racist attitudes and discriminatory behaviors, these data likely represent lower-bound estimates of the actual occurrence of discrimination in society, although evidence on this issue is far from clear. The only “validation” of reported discrimination comes from credible descriptions of discriminatory incidents and from instances in which self-reports, matched-pair studies, and counts of complaints yield similar results, particularly similar intergroup differentials (Smith,
2002). Indeed, direct measures of experiences and perceptions of discrimination are probably best used to support valid findings from other kinds of studies to estimate the contribution of discrimination to observed disparities in outcomes among racial groups. Longitudinal data in particular can provide useful information about process-related differences across racial groups, such as changes in perceptions and experiences (as well as outcomes) over time.2 Thus, despite our caveats, we conclude that it is informative to consider directly reports of levels and changes in discrimination and prejudice.
SOURCES OF OBSERVATIONAL DATA
Probably the most extensive data on experiences and perceptions of racial discrimination come from survey research. We therefore begin this review of sources of observational data on discrimination by discussing in detail the content, uses, advantages, and limitations of surveys of interpersonal relations and racial discrimination. We then examine more briefly three additional data sources: governmental administrative data, nongovernmental data, and in-depth interviews.
Surveys of Interpersonal Relations and Racial Discrimination
Design and Strengths
Surveys of interpersonal relations use interviews or questionnaires to collect detailed information about respondents’ perceptions of and experiences with discrimination. Researchers use this information to better understand the causes and consequences of various forms of discrimination and to investigate relationships among racial attitudes and beliefs, on the one hand, and perceptions and experiences of discrimination on the other. The surveys are typically administered on a regular basis, either cross-sectionally or longitudinally, so that researchers can observe changes over time in attitudes and perceptions and in the relationships between them.
In his review of survey research methods, Fowler (1993) describes three key methodological components—sampling, interviewing, and question design—that are necessary for designing quality surveys and collecting credible data. First, survey researchers use probability methods to randomly
draw large representative samples from specified populations in order to ensure statistical generalizabilty. Second, they rely on multiple interviewers rather than a single investigator to reduce bias, and they use structured interviews or standardized questionnaires to ensure consistency of data collection and measurement. Third, they design survey questions to gather objective information or to measure subjective phenomena, often by using scale measures.3 It is essential that the survey questions be reliable and consistent with what the researchers are trying to measure.
When developing survey content, researchers consider the general data collection approach (e.g., personal interviews, computer-assisted telephone interviewing, or mail questionnaires), as well as such factors as question order, content, wording, and type (e.g., open-ended versus fixed responses). As demonstrated by Bobo and Suh (2000; see also Suh, 2000) and Smith (2001), the use of open-ended questions in surveys is particularly valuable for obtaining detailed descriptions of incidents of discrimination. Such questions combine some of the best strengths of surveys (e.g., generalizability and standardized data collection) and in-depth interviews (e.g., narrative richness and detail). Open-ended questions yield specific details about the nature and circumstances of discriminatory behavior that can often validate (or indicate inaccuracies in) direct, self-reported accounts, thereby providing useful information for understanding discrimination.
Key Examples of Surveys
Over the years, researchers have asked thousands of questions and conducted numerous studies on the state and nature of intergroup relations in America (Bobo, 1997; Bobo and Kluegel, 1997; Jackman, 1994; Kinder and Sanders, 1996; Newport et al., 2001; Schuman et al., 1997; Sears and Jessor, 1996; Sears et al., 1997; Smith, 1998, 2000, 2001; Sniderman and Piazza, 1993; Steeh and Krysan, 1996). Common sources of survey data include national polls (Henry J. Kaiser Family Foundation, 1999; Newport, 1999; Newport et al., 2001; Sigelman and Welch, 1991; Smith, 2000; Sniderman and Carmines, 1997; Weitzer and Tuch, 1999); community-based surveys (Bobo and Suh, 2000; Brown, 2001; Gary, 1995; Smith, 1993; Suh, 2000); and specialized studies of employees, employers, professionals, and other target populations (Antecol and Kuhn, 2000; Braddock and McPartland, 1987; Preston, 1998; Supphellen et al., 1997; Yen et al., 1999).
Some widely used surveys of race relations include the Roper polls (Roper Center for Public Opinion Research, 1982), the Multi-City Study of Urban Inequality (Bobo, 1997; Bobo and Kluegel, 1997), the GSS (discussed
further below; Davis et al., 2001), and public opinion polls from the Gallup Organization and Princeton Survey Research Associates (Smith, 2001). The Multi-City Study of Urban Inequality, fielded around 1993 and administered in four urban areas—Atlanta, Boston, Detroit, and Los Angeles—contains explicit measures of racial stereotypes and outgroup perceptions of whites, blacks, Hispanics, and Asians (Bobo and Massagli, 2001; Kluegel and Bobo, 2001; O’Conner et al., 2001). A compilation of holdings of the Roper Center archives in 1982 listed some 4,850 questions dealing with race relations (Roper Center for Public Opinion Research, 1982). Since 1995, the database of national polls maintained by the Roper Center has listed 2,453 questions on the topic of blacks or minorities, at least 359 of which deal with a wide range of items on discrimination, such as the level of discrimination experienced in general public policies dealing with discrimination, personal experiences of discrimination, and causes of discrimination. Surveys of discriminators are generally restricted to employers or other decision makers rather than the general public (e.g., Kirschenman and Neckerman, 1991; Neckerman and Kirschenman, 1991; Supphellen et al., 1997).
Currently, the best example of a large-scale survey of intergroup relations is the GSS, a biennial social indicator survey conducted by the National Opinion Research Center since 1972 (Davis et al., 2001). The GSS receives its core funding from the National Science Foundation and supplemental funding from a wide variety of organizations. The survey uses a national probability sample of 3,000 noninstitutionalized adults to collect information on trends in attitudes and behaviors using a set of complex questionnaires. Each questionnaire includes a standard core of demographic and attitudinal variables, as well as topics of special interest, such as Internet use, multiculturalism, national security, health status, and religion.
Methodological experiments are incorporated into the GSS data collection each year. Such field experiments are also conducted within the other surveys referenced above to assess how the framing of questions affects responses relevant to discrimination. Recent developments in survey research, principally involving computer-assisted telephone interviewing, have made it possible for survey respondents to be assigned easily to random sets or orders of questions (see, e.g., Emerson et al., 2001; Gilens, 1996). Box 8-1 provides an example of such an experiment.
Questions in these surveys are generally used to obtain data on various topics, including stereotypes, social distance, intergroup contact, and discrimination. Among those questions on discrimination in the GSS, for example, are items on personal experiences of discrimination, assessments of discrimination experienced by a particular group, or, less commonly, acts of discrimination committed by the respondent or people he or she knows. However, few of the questions dealing with discrimination provide vali-
The use of a “split ballot” experiment embedded in a large-scale national survey enables researchers to draw inferences about the general population of responders to surveys with samples that are larger and more diverse than the student population in a typical laboratory experiment. Gilens (1996) used such an experiment within a sample survey to examine whites’ racial attitudes and views about welfare using data from the 1991 National Race and Politics Study. The survey respondents were randomly assigned to one of two treatment groups: Half were asked about their beliefs regarding black welfare mothers and half about their beliefs regarding white welfare mothers. Under the assumption that the two groups are interchangeable—guaranteed (on average) by randomization—a difference in the evaluation of welfare provides an estimate of the degree to which race coding may be implicated in white opposition to welfare. The results showed that whites viewed black and white welfare mothers similarly but had more negative views of black welfare mothers when considering policies on welfare. As in laboratory experiments, there may be a difference between responding to questions in a survey and acting in a discriminatory fashion in settings that affect others. Yet as Gilens demonstrates, determining how such attitudinal differences relate to behavioral differences (such as differences in political choices) is important.
dated measures of the occurrence of unequal treatment (see Smith, 2002). The most valuable of these items measures the level of discrimination experienced by individuals or the level of discrimination experienced by particular groups or venues. Questions about the overall level of discrimination without regard to personal experiences, groups, or venues are probably too general to be of much use.
Imperfect Relationship of Attitudes to Behaviors
Smith (2002:14) notes that “many of the questions on intergroup relations in the holdings of the Roper Center or on the GSS provide important information on the state of intergroup relations and help one to understand the context and causes of discrimination, public support for policies to combat discrimination, and related matters.” However, relatively few surveys attempt to measure the incidence of discrimination directly at either the individual or the collective level. Smith continues: “Given that only a mod-
erate correlation exists between intergroup beliefs and attitudes (e.g., stereotypes and prejudice) and discriminatory actions (Dovidio, 1993; Dovidio et al., 1996; Fiske, 2000), studying the former is not the same as measuring the latter.”
In most studies of attitudes and behaviors, the correlation is typically moderate (Eagly and Chaiken, 1993), and the intergroup domain is no exception. The methodological solutions in the racial discrimination area are the same as in other areas:
Investigators must create a careful match between the attitude and the behavior. For example, some intergroup experiments (described in Chapter 6) indicate that subtle attitudes best predict subtle (but powerful) behaviors, whereas overt attitudes predict more overt behaviors (Dovidio et al., 2002). As another example, there is evidence that whites’ general attitudes toward blacks predict patterns of discrimination, not necessarily any one randomly chosen behavior; by the same token, there is evidence that whites’ attitudes toward a specific racial policy or practice (such as busing in a particular school district, housing in a particular neighborhood, or affirmative action in a particular employment or education context) predict their specific behavior in that regard. This elementary point is often missed (Ajzen, 2001).
The racial attitude–behavior match will depend on moderator variables relating to the type of person involved. Some people act on their attitudes more reliably than do others (Snyder and Swann, 1976).
The racial attitude–behavior match will also depend on the situation in which the attitude and the discrimination are measured. Social norms strongly affect both what is reported and what is enacted (e.g., Ajzen, 1991), so they can override the ability of attitudes to predict behavior.
The strength of the attitude itself affects related behavior (Petty and Krosnick, 1995).
Smith (2002) describes a number of limitations of surveys. These limitations encompass both methodological factors and reporting biases.
Methodological factors. As noted above, surveys cannot directly measure discrimination; they capture self-reported evidence on perceptions and experiences of discrimination that is not validated. Moreover, such factors as target group, data collection mode, interviewer race, venue, and time of occurrence, as well as question format (e.g., explicit or implicit, phrasing, order), mode (e.g., survey or in-depth interview), and context, can affect the accuracy and completeness of reported perceptions and experiences of dis-
crimination. Discrimination that is subtle or indirect, for instance, may not be readily detected using explicit items. Respondents may also use different meanings for discrimination from one reported account to another (e.g., individual, group, or structural discrimination).
Reporting biases. As the leading historical targets of discrimination in the United States, African Americans may be in the best position to assess the ongoing reality of race in public life. Some scholars, however, suggest that African Americans may have become more sensitive to discrimination because of their socialization since the passage of civil rights legislation, leading them to notice prejudicial actions now more than in the past and producing an upward reporting bias over time (Bobo and Suh, 2000; Brown, 2001; Gary, 1995; Gomez and Trierweiler, 2001; Sigelman and Welch, 1991; Suh, 2000). This reporting bias may also occur if respondents attempt to support a personal conviction that America is a racist society or to explain unfavorable situations and outcomes in their own lives (Harrell, 2000; Lucas, 1994).
Given that blacks are likely to have been well attuned to racial discrimination both before and after the civil rights legislation, however, such upward reporting biases are not likely to be severe. Coleman et al. (2002) actually find that blacks sharply underreport exposure to discrimination when such reports are compared with separate statistical estimates of wage discrimination. In general, researchers have found direct self-reports of discrimination to be accurate and reliable when cross-validated against other data sources (Bobo and Suh, 2000; Essed, 1991; Landrine and Klonoff, 1996). For example, in one recent national survey and one local survey, items on racial discrimination were followed by open-ended questions asking people to describe the mistreatment they supposedly had experienced (Smith, 2000). Across all racial and ethnic groups, some 95 percent of respondents were able to describe specific, appropriate, and credible incidents of discrimination.
Whatever biases may stem from changes in sensitivities among blacks or other racial or ethnic minorities, they are likely to be small in comparison with shifts in white behavior from fairly overt forms to more subtle forms of discrimination. As noted above, because most overtly discriminatory behaviors were declared illegal by the mid-1970s, continued differential treatment on the basis of race has had to become subtler and often may not be apparent even to its targets. Other things being equal, the shift from more overt to more subtle forms of discrimination by whites may actually increase underreporting by blacks because they may be less likely to perceive subtle forms of discrimination. To the extent that real shifts have occurred in white racial attitudes and behaviors, reported declines in discrimination by blacks will be accurate, but part of the decline may reflect
the fact that discrimination has been declared illegal by the federal government with the consequence that public expressions of racial prejudice and overt forms of discrimination are widely considered to be socially unacceptable. Annex 8-1 presents survey findings on perceptions of discrimination among both blacks and whites over the past few decades.
Means of Improving Survey Measures
Methodological improvements. Perhaps the most important area for methodological research to improve survey measures of discriminatory experiences and perceptions is research to help understand what aspects of survey and question design may affect levels of reported discrimination, such as the degree to which reports are influenced by race priming.4 To address this issue, Smith (2002) suggests investigating different approaches to the wording and placement of questions experimentally through random assignment on existing social surveys. If levels of reported discrimination vary by question wording and order, follow-up work will be needed to explain these variations and establish which wordings and placements yield the most accurate results. Another area for research is to use bounded and aided recall techniques in panel surveys to learn how cross-sectional reports may be distorted by errors of memory and telescoping. In addition, cognitive research is needed to understand what respondents specifically include and exclude when they hear such terms as “discrimination” and “unfair treatment.”
One fruitful avenue for improvement might be greater use of the factorial vignette method, in which stories are presented to respondents about people being questioned by the police, applying for a job, running for political office, and the like, with the race of the people in the vignette, along with other relevant factors (e.g., age, gender, education, work experience, criminal history), being systematically varied by random assignment. This approach enables researchers to see experimentally how race influences respondents’ reported perceptions and evaluations (for a recent example, see Emerson et al., 2001).
Finally, more and better validation studies are needed to discover which approaches yield the most accurate reports of behavior. One possible way of validating self-reports is to collect and compare information from representative samples on reported behaviors and practices that may be consid-
ered discriminatory (potential discriminators) and on reported experiences with discrimination (targets of discrimination). The former approach may not prove practical, however, if discriminators do not perceive their actions as unfair or biased (Dovidio, 1993; Essed 1991) or if self-recognized acts of discrimination are underreported because of social desirability bias (discussed above). Nonetheless, the approach merits investigation, as incidents of reported discrimination necessarily have both perpetrators and targets. Studies of employers also show that it is possible to collect self-reports of prejudiced attitudes and biased actions against members of disadvantaged racial groups (see Kirschenman and Neckerman, 1991; Supphellen et al., 1997).
Perhaps the most basic form of cross-validation is the use of multiple methods in measuring discrimination. For example, employment discrimination at a company might be studied by examining grievances filed, carrying out surveys of employees and bosses, conducting in-depth interviews, and analyzing employee records.
Other improvements. Smith (2002) offers suggestions for improvements that do not involve additional methodological research. First, surveys should include more target groups. In most surveys, statistically reliable results are available only for whites and blacks, yet Hispanics and Asians are rapidly increasing their shares of the U.S. population, and Arabs and Muslims have recently become prominent as potential targets of prejudice.
Second, surveys should elaborate and extend their measures of discrimination. Some surveys have used only a single measure. Questions need to be refined substantively as well as methodologically to capture subtle and not just explicit discrimination (Dovidio, 1993; Pettigrew and Meertens, 1995; Sears et al., 1997). In this regard, Supphellen et al. (1997) find that projective measures of employment discrimination (e.g., rating the attitudes or opinions of others) are more valid than direct self-reports.
Third, surveys should inquire about discrimination within specific venues and not rely on global questions that leave time and place unspecified. Typical venues included in surveys to date include work, restaurants, stores, interactions with police, schools, housing, public transportation, banks, and government agencies (Brown, 2001; Collins et al., 2000; Gary, 1995; Newport et al., 2001; Smith, 2000). An “other” category might be included to capture incidents occurring outside these venues (Smith, 2000). Specific time periods should also be specified. Doing so would enable estimation not only of the incidence of discrimination but also of its frequency. Lifetime “ever” questions are less than ideal because they demand recall over extended periods of time (which is cognitively difficult and leads to underreporting); they confound cross-sectional monitoring and time-series analysis (because some people report on events that occurred only in the distant past, and others do
not); and they underestimate the frequency (as opposed to the incidence) of discrimination (Suh, 2000).
Fourth, not only is it important to measure discrimination, it is also important to assess racial preferment. In 1991, for example, 38 percent of black respondents to the GSS reported that their promotion opportunities were worse because of their race, but 18 percent indicated that their chances were better, yielding a net disadvantage of –20 points. In contrast, white respondents gave responses that yielded a net advantage of +19 points (Smith, 1998).
Governmental Administrative Data
After the passage of the various civil rights acts prohibiting racial discrimination, bureaucratic agencies were established and charged to investigate complaints, monitor compliance, and work to eliminate bias.5 For example, the National Fair Housing Alliance (NFHA) produces an annual report on the state of housing discrimination in the United States that is based on information from local fair housing groups, the U.S. Department of Housing and Urban Development, the U.S. Department of Justice, and numerous state and local government agencies. Discrimination is tracked in four different market sectors: rental markets, mortgage lending, home sales, and homeowner insurance. Complaints filed with NFHA and other agencies provide one source of data on the underlying trends in the frequency and incidence of housing discrimination. The NFHA data compiled for 2001 show that race continues to be the most commonly reported basis for discrimination, accounting for 32 percent of the 23,557 cases filed (National Fair Housing Alliance, 2002). Other potential sources of administrative data include antidiscrimination suits filed in state and federal courts (Garrett, 2001; National Research Council, 1989; Romero, 2000; Shivley, 2001) and registries of hate crimes maintained by state and local human rights commissions and race relations boards (Evans, 2001; Strom, 2001).
As noted by Smith (2002), administrative data have several advantages: They represent socially significant events, they are publicly accessible, and they are generally inexpensive to use because the costs of collection are borne by the enforcing agency. Using industry data from the Bureau of Labor Statistics, for example, Petersen and Morgan (1995) were easily able to document the skewed placement of men and women within occupational
categories within firms. The data showed that women were far more likely to be in poorly paid jobs. This finding suggests that if one were to look for possible gender discrimination, it would be important to look not just at wage differences within jobs but also at the mechanisms that allocate men and women to occupations within the firm.
As Smith points out, however, relying on government reports also has serious limitations. First, the data are collected to meet legal requirements, not to facilitate social science research. In other words, coverage is defined by law, and the information is collected largely for administrative rather than scientific purposes. Second, willingness to report discriminatory treatment may depend on the ease of reporting and the vigor with which an agency deals with complaints (Lucas, 1994). As these conditions vary over time, trends may be biased (Shivley, 2001). Finally, not all discriminatory practices are illegal (and thus covered by government enforcement mandates), and many acts of discrimination that are now illegal were not so several decades ago (Romero, 2000). The Anti-Defamation League (2001), for example, includes acts of anti-Semitic speech in its annual report on anti-Semitism, even though such speech is legal. Because of problems with the quality and completeness of administrative data, the usefulness of these data sets for research purposes is limited.6
In addition to government agencies, many private organizations maintain a record of discriminatory complaints. However, complaints in such cases are made to internal grievance boards governed by formal procedures within an organization; an example is complaints filed with an employer or labor union. Internal data from for-profit organizations have many of the same limitations as those from governmental agencies. Companies establish procedures to serve legal and business purposes, not to further a scientific research agenda. In addition, company reports are typically not made public, and their nature and specifics vary considerably across organizations.
Racial grievances may also be lodged with independent nonprofit organizations whose mission is to promote racial or ethnic equality, such as the Anti-Defamation League or the Mexican-American Legal Defense and Educational Fund. Reports to nongovernmental third parties have the advantage that they are publicly accessible, typically cover many different types of discrimination in various venues, and draw from across the nation. However, the resulting data are often compiled by highly self-interested organizations that are not scientifically motivated. Moreover, as with government
and other organizational reports, these sources rely on targets of discrimination knowing where and how to report mistreatment and being motivated and able to do so.
In conducting in-depth or qualitative interviews, a researcher engages one or more subjects in an extensive, semistructured conversation, which is often audio recorded (Bonilla-Silva and Forman, 2000; Essed, 1997; Feagin, 1991; Feagin and Sikes, 1994; St. Jean and Feagin, 1998, 1999; Zweigenhaft and Domhoff, 1991). The advantages of such interviews are that they ask individuals about their actual experiences of discrimination and often elicit information that is richly detailed. Essed (1991), for example, claims that using unstructured interviews rather than highly structured ones allowed her to draw out more detailed accounts of people’s experiences. She also notes that participants were able to express intuitive feelings regarding their experiences with prejudice and discrimination that might otherwise be difficult to articulate and to report events in ways that others might consider oversensitive.
In-depth interviews are generally based on small, usually unrepresentative samples that are often biased because participants are of higher status, more articulate, and more politically aware than most of the subject population (see Essed, 1991; Feagin, 1991; Zweigenhaft and Domhoff, 1991). The most frequently used method of selection is the chain referral method, also known as snowball sampling, in which respondents provide leads to other potential interviewees within their social networks. Because social networks tend toward homogeneity, snowball sampling can yield biased samples and findings that are difficult to generalize (Goodman, 1961). Moreover, the accuracy of self-reports may be affected by ways in which respondents react to interviewer probes for reports of discrimination or other behaviors, or if respondents provide vague and poorly detailed accounts (Smith, 2002).
SCALE MEASURES USED IN SURVEYS
As discussed above, although survey-based self-reports have been found to be reliable, accurate, and useful ways of measuring experiences of discrimination, the shift from overt to subtle forms of discrimination has made it more difficult to assess the occurrence of discrimination or to capture people’s beliefs using survey questions. Many observers have noted an ongoing conflict between principle and practice regarding racial prejudice in the American psyche and have theorized the contradiction under a variety of conceptual rubrics (Fiske, 1998): modern racism (McConahay, 1986),
symbolic racism (Kinder and Sears, 1981), ambivalent racism (Katz et al., 1986), aversive racism (Dovidio and Gaertner, 1986), laissez-faire racism (Bobo et al., 1997; Bobo and Smith, 1998), and subtle racism (Pettigrew and Meertens, 1995) (see also Annex 8-1). Although individual survey questions may not capture such complex attitudes and ambivalence about race, scale measures, which combine information from multiple questions, can capture the sometimes contradictory mechanisms that lead to more subtle types of discrimination. We turn to these measures next.
Measures of Modern Racism
The psychological source of a conflict between racial equality in principle and practice results from people’s assimilation of egalitarian laws and norms, along with their internalization of continued cultural messages that blacks are inferior to whites. For example, many whites, regardless of explicit individual prejudices, make rapid associations between whites and positive ascriptions and between blacks and less positive ascriptions (e.g., Devine, 1989; Dovidio et al., 1986; Gaertner and McLaughlin, 1983; see Chapter 6). In part, these associations may reflect media coverage depicting blacks less favorably than whites in news reports and in entertainment. Many Americans resolve the conflict between these culturally influenced associations and the nation’s egalitarian values by maintaining those values simultaneously with subtle, automatic, or indirect forms of prejudice (see Chapters 4 and 6).
In survey settings, measures of modern racism examine reactions to black Americans as perceived threats to whites’ traditional values and economic status (Duckitt, 2001). The single most commonly used measure since the mid-1970s has been McConahay’s (1986) Modern Racism Scale—an outgrowth of work on symbolic politics (Kinder and Sanders, 1996; Kinder and Sears, 1981). Generally split into three parts, this scale typically encompasses modern (subtle) racism, old-fashioned (overt) racism, and filler items on irrelevant current events to disguise the measure. Items on modern racism include believing, for example, that the government and the media give blacks more respect than they deserve, that discrimination is no longer a problem, that blacks have gotten more economically than they deserve, and that blacks are too demanding. In contrast, items on old-fashioned racism ask about open opposition to fair housing laws, integration, intermarriage, and having a black neighbor.
Scores on the Modern Racism Scale predict a variety of variables related to discrimination to varying degrees (Kinder and Sanders, 1996). Examples include conscious endorsement of stereotypes about black Americans (Devine, 1989), anti-black feelings, simulated hiring decisions, voting for white over black candidates (even after controlling for ideology), and
opposing busing to achieve school desegregation (even after controlling for related variables). A variant of the scale (Pettigrew and Meertens, 1995) distinguishes subtle and explicit prejudice against immigrant minorities in Europe. Items on this scale include blaming minorities for being too pushy and intrusive but simultaneously not trying hard enough, as well as with-holding sympathy for their disadvantaged situation.
Although widely used, the Modern Racism Scale has provoked two kinds of scholarly debate. First, while few dispute the empirical correlates of the scale, their meaning is open to interpretation, either as a perceived threat to the situation of one’s own group or as principled conservatism. For example, some argue that opposition to racial integration, school desegregation, affirmative action, and welfare programs reflects primarily opposition to government intervention (Sniderman, 1985); however, the role of such ideology appears to be empirically small (Kinder, 1998). Some argue that racial attitudes reflect narrow economic self-interest (Bobo and Kluegel, 1997); that is, people vote their own personal wallet. Yet perceptions of the economic standing of one’s racial group as a whole appear to be at least as strong a predictor of racial attitudes and voting behavior as perceptions of one’s own economic standing (Kinder and Kiewiet, 1981; Sears and Funk, 1991). This suggests that modern racism is based on the interests of one’s group, not one’s own self-interest. Moreover, consistent with the idea of modern racism, the importance that white Americans attach to equality of opportunity, relative to their other core values, correlates with their more specific racial attitudes and policy preferences (Kinder and Sanders, 1996). In other words, modern racism represents a syndrome of perceived own group interest, basic values, and specific attitudes toward policies—all of which disadvantage minorities.
Some common ground for this first type of debate—over the meaning of modern racism as group threat versus principled conservatism—appears in findings that modern, subtle forms of prejudice correlate with (a) subjectively perceived threats to group status (Bobo, 1983), (b) positive feelings toward one’s own racial group, and (c) negative feelings toward racial outgroups (Wood, 1994). Group prejudice may not be the only factor predicting racial attitudes and policy preferences, but as one commentator notes, “it is always present, and of all the ingredients that go into opinion, it is often the most powerful” (Kinder, 1998:805).
A second form of scholarly debate has emerged more recently. The original items on the Modern Racism Scale were considered relatively nonreactive; that is, respondents did not necessarily interpret them as measures of discrimination and so apparently did not monitor their responses for social desirability. Over the past decade, however, growing evidence suggests that the items have become reactive (Fazio et al., 1995). Modern racism scores now correlate with more explicit measures of deliberative
discrimination, rather than with automatic or implicit measures (e.g., Dovidio et al., 1997), although correlations with more implicit, spontaneous measures of discrimination still occur (Wittenbrink et al., 1997, 2001).
More generally, various dimensions of prejudice matter to understanding both prejudiced attitudes and discriminatory behavior. One dimension contrasts spontaneous, implicit, automatic, and subtle versus deliberative, explicit, and controlled reactions (see Chapter 6, Box 6-3); another important dimension contrasts emotional, evaluative, and affective processes versus belief, conceptual, and cognitive processes (see Chapter 6 and Talaska et al., 2003); and a third dimension contrasts ingroup preference versus outgroup derogation (see Chapter 4).
One key component of measuring subtle racism is that it depends heavily on context. As an example, to the extent that white attitudes are ambivalent—encompassing both sympathy and rejection—people may act on different aspects of their attitudes under different circumstances. Racial ambivalence (Katz and Hass, 1988; Katz et al., 1986) suggests the co-occurrence of blaming anti-black feelings (the perceived irresponsibility of black families, leaders, and values underlies the continuing disadvantage of black Americans) with paternalistic pro-black feelings (emphasizing obstacles, discrimination, and unequal opportunities). The work by Katz and colleagues demonstrated empirically that anti-black attitudes correlated with white perceptions that blacks violate values related to the Protestant work ethic and that pro-black attitudes correlated with humanitarian and egalitarian values. Whites can simultaneously possess both sets of attitudes. The implication for discriminatory behavior is that reactions toward a single black individual can be affected by a small push in either a positive or negative direction (e.g., slightly superior or slightly inferior credentials for a job applicant). Racially ambivalent whites then overreact, making excessively positive or excessively negative decisions as compared with their decisions about a comparable white individual. The Modern Racism Scale elicits a similarly exaggerated response, either overly positive or overly negative, in simulated hiring decisions.
The common thread of all the work on modern racism, symbolic racism, subtle prejudice, and ambivalent racism is that appearing racist has become aversive to many white Americans (Dovidio and Gaertner, 1986), creating more complex racial attitudes than in the past. Whatever their source, subtle, modern forms of racism predict avoidance of interactions with racial outgroups, and such passive rejection can result in discrimination (see Chapter 3).
Survey research would be improved by a more explicit model of what forces determine expressed attitudes: expectations, experiences, social pressure, and so on. The analysis of relationships between past experiences, expressed attitudes, and future behavior seems remarkably undeveloped.
Embedding attitude reports in a lifetime of behaviors would permit this area of research to develop in new ways. Furthermore, research on modern racism can inform statistical and experimental research by examining relevant individual characteristics (e.g., personal values) and features of encounters (e.g., interdependence, accountability). Some social science models, such as the theories of reasoned action and planned behavior, provide a framework for incorporating a variety of factors to predict behavior and evaluating the respective weights and values of own beliefs, perceived social norms, and perceived behavioral control, as well as own prior behavior (Ajzen, 2001).
Measures of Explicit Racism
Not all white Americans eschew overt racial bias. By some estimates, about 10 percent of white Americans openly embrace racial discrimination; if accurate, this figure would mean there is at least one prejudiced white person for every black person in the United States (Fischer et al., 1996). Hence, even this apparently low incidence of prejudice can be viewed as high. Moreover, measures of more explicit racially biased attitudes correlate with more explicit and potentially violent forms of discrimination. Not everyone holding these attitudes is violent, of course, but perpetrators of violence do hold these attitudes. Thus, explicit racism is a necessary but not sufficient predictor of the most serious forms of discrimination.
Three contemporary scales predict endorsement of overtly prejudiced attitudes. One such scale measures blatant or explicit prejudice (Pettigrew and Meertens, 1995), defined as resentment of racial and ethnic minority groups (their allegedly stealing ingroup jobs but also using welfare), as well as rejection of ties to minorities (having a mixed grandchild or an outgroup boss). High scores on this scale predict generalized ethnocentrism and overall rejection of outgroups not one’s own, as well as approval of racist political movements and hate crimes (Green et al., 1999). Explicit prejudice stems from perceived threat to the economic status of one’s own group.
A related scale measures social dominance orientation (Sidanius and Pratto, 1999); it also focuses on economic and status competition between societal groups. High scores on this scale indicate agreement by respondents that some groups are just more worthy than others, that group hierarchy is inevitable and good, and that dominance is necessary. These attitudes correlate with believing that the world is competitive and that force is sometimes necessary to keep inferior groups in their place. They also correlate with explicit racial prejudice.
In addition to perceived economic threat, perceived threat to values predicts discriminatory attitudes. A scale on right-wing authoritarianism (Altemeyer, 1988, 1996) measures belief in old-fashioned ways, censorship,
leadership based on superior power, and rejection of troublemakers and deviants. High scores on this scale correlate with social conservatism and predict approval of aggression against nonconformers. They also correlate with limited intergroup contact and limited education.
SUMMARY AND RECOMMENDATIONS
Following the enactment of civil rights legislation, overt racial discrimination became illegal and socially unacceptable, and measuring discrimination became increasingly difficult as a result. The more subtle forms of discrimination evident today complicate the way we assess the causes and consequences of discrimination. Surveys provide valuable evidence for understanding the extent of discrimination; however, they cannot directly measure its occurrence. Most survey items are intended to measure self-reported attitudes, perceptions, or experiences of discrimination, and these items can be unreliable for at least two reasons. First, if a discriminatory occurrence is ambiguous, a black or other minority respondent may under-or overreport its incidence. Some subtle forms of discrimination, for instance, may not be easily detected. Second, white respondents are often not willing to admit to practicing or supporting discrimination, which may lead to less-accurate reporting of their true attitudes or beliefs.
On the other hand, although survey measures cannot capture discrimination directly, results to date suggest that valid and reliable data on racially discriminatory attitudes and experiences can be gathered on social surveys. Conducting repeated cross-sectional surveys is very useful to provide time series; the GSS is the best example of a large-scale survey that collects data on changes over time in racial attitudes and experiences with discrimination through yearly interviews with samples of the population. Conducting longitudinal surveys to analyze the incidence, causes, and consequences of changes in attitudes about race and experiences of racial discrimination at the individual level is also very valuable, although none of the major longitudinal surveys to date has included attitudinal or perceptual variables (see further discussion in Chapter 11). More generally, research is needed to evaluate and improve the reliability and consistency of survey reports of discriminatory attitudes and behavior. In particular, as expressions of prejudice and discriminatory behavior change and become more subtle, survey questions on racial attitudes and experiences of discrimination may be necessary. Scale measures can be very useful to capture complex racial attitudes that can lead to more subtle types of discrimination. Open-ended questions on surveys can provide some of the advantages of in-depth interviews in regard to the detail of information obtained combined with the advantages of surveys of large, representative samples.
Reports of discrimination in administrative records systems, such as
those of government civil rights enforcement agencies, private organizations, and nonprofit groups, can also provide useful information for analysis that is available at low additional cost to the researcher. However, some administrative records may be difficult to obtain by researchers, the reporting of events may be biased in several ways (e.g., by changes in the vigor with which an agency pursues enforcement of antidiscrimination laws and policies), and data on covariates may be very limited, thereby restricting the use of the records for research as well as administrative purposes.
Recommendation 8.1. To understand changes in racial attitudes and reported perceptions of discrimination over time, public and private funding agencies should continue to support the collection of rich survey data:
The General Social Survey, which since 1972 has been the leading source of repeated cross-sectional data on trends in racial attitudes and perceptions of racial discrimination, merits continued support for measurement of important dimensions of discrimination over time and among population groups.
Major longitudinal surveys, such as the Panel Study of Income Dynamics, the National Longitudinal Survey of Youth, and others, merit support as data sources for studies of cumulative disadvantage across time, domains, generations, and population groups. To further enhance their usefulness, questions on perceived experiences of racial discrimination and racial attitudes should be added to these surveys.
Data collection sponsors should support research on question wording and survey design that can lead to improvements in survey-based measures relating to perceived experiences of racial discrimination.
Recommendation 8.2. Agencies that collect administrative record reports of racial discrimination should seek ways to allow researchers to use these data for analyzing discrimination where appropriate. They should also identify ways to improve the completeness, reliability, and usefulness of reports of particular types of discriminatory events for both administrative and research purposes.
ANNEX 8-1: BLACK AND WHITE AMERICANS’ PERCEPTIONS OF DISCRIMINATION AND AMBIVALENT ATTITUDES ABOUT RACE
Black Perceptions of Discrimination
Prior to the passage of civil rights legislation, blacks perceived the United States to be a highly prejudiced and discriminatory society. In 1963, more than three-quarters of a nationally representative sample of black Americans perceived significant racial discrimination in U.S. job markets (Schuman et al., 1997). Asked whether they had as good a chance as whites to get jobs for which they were qualified, a resounding 77 percent said “no.” In the following years, the percentage of black respondents perceiving employment discrimination fell, reaching 64 percent in 1978 and 55 percent in 1989. As late as 1997, more than half (53 percent) of all African Americans said that blacks still did not have as good a chance as whites to get jobs for which they were equally qualified. Moreover, in 1996, 63 percent of African Americans nationwide continued to view discrimination as a primary cause of disadvantage among blacks (Schuman et al., 1997).
Between 1997 and 2001, the Gallup Organization and Princeton Survey Research Associates public opinion polls asked nationally representative samples of African Americans to report any discrimination or unfair treatment they had experienced within the past 30 days (Smith, 2001). On average, 26 percent of respondents said they had experienced discriminatory treatment while shopping, 16 percent at the workplace, and 16 percent while on public transportation. A national survey of African Americans sponsored by the Washington Post during 2000 found that 30 percent had at least sometimes been “called names or insulted” and 17 percent had been “physically threatened or attacked” because of their race. Rates for non-blacks were substantially lower, with only 18 percent of the general population reporting a racial or ethnic insult and 11 percent a physical threat or attack. Clearly, African Americans, more than other Americans, still perceive significant discrimination in public life and view it as a significant barrier to their social and economic advancement.
In the Multi-City Study of Urban Inequality, each respondent was asked: “In general, how much discrimination is there that hurts the chances of blacks to get good-paying jobs?” In Atlanta, 60 percent of black respondents answered “a lot,” compared with 57 percent in Boston, 62 percent in Detroit, and 69 percent in Los Angeles. When respondents who answered “some” to the same question were added, the percentage of African Americans who perceived racial discrimination in job markets rose to well over 90 percent in each metropolitan area (Kluegel and Bobo, 2001). Of course, the implicit definition of discrimination used by respondents may not be the same as the legal definition currently recognized by U.S. courts.
White Perceptions of Discrimination
Not long ago, white respondents were willing to admit their support for racial discrimination to survey researchers. In the early 1940s, for example, 68 percent of whites nationwide said they thought blacks and whites should attend separate schools, 55 percent said whites should have priority over blacks in hiring, and 54 percent agreed that separate sections should be reserved for blacks and whites on buses and streetcars. By the 1960s, such segregationist attitudes had moderated considerably, with 30 percent of whites still favoring racially separate schools and 11 percent approving of white preferences in hiring. In 1970, 12 percent of whites admitted to favoring racial segregation in public transportation (Schuman et al., 1997).
In the years since the civil rights era ended, these percentages have fallen even further; fewer and fewer whites are willing to express open support for racial discrimination. By 1995, only 4 percent reported they believed that blacks and whites should attend separate schools, just 13 percent said there should be laws against black–white intermarriage (as late as 1963 the percentage was still 62 percent), and over 90 percent of whites endorsed the principle that blacks have a right to live wherever they can afford to live (Schuman et al., 1997). Although white support for racist principles had fallen to low levels, it had not disappeared entirely. As late as 1993, 15 percent of whites agreed that blacks should respect the rights of whites to exclude blacks from their neighborhoods if they so desire (Schuman et al., 1997).
Moreover, while fewer whites openly support principles of racial discrimination, many remain ambivalent in their attitudes about race. For instance, surveys show that only 13 percent of whites support a ban on black–white intermarriage, whereas 33 percent still disapprove of the practice personally. A mere 2 percent object to sending their children to a school where “a few” students are black, while 19 percent object to sending their children to one where half are black. Likewise, just 2 percent of whites said they would move out of their home if black neighbors moved in next door, but 25 percent said they would leave if blacks entered their neighborhood “in great numbers.” As of 1995, nearly a quarter of whites (23 percent) said they would object to having a black dinner guest (Schuman et al., 1997).
Ambivalent Attitudes About Race
At the end of the twentieth century, open support for principles of racial discrimination had fallen to very low levels among whites, with only 10 to 15 percent endorsing discriminatory actions or policies. Although whites may have come to support a nondiscriminatory society in principle, however, they remain substantially uneasy about its implications in practice.
White respondents, for example, continue to express reservations about racial mixing in social institutions where such mixing formerly did not occur, and racial segregation generally remains high in schools, churches, neighborhoods, and marriages (see Emerson and Smith, 2000; Farley, 1996; Massey and Denton, 1993; Orfield and Eaton, 1996). For instance, although most whites now agree that blacks should be able to live wherever they choose, they still want blacks to choose to live somewhere else.
Racial attitudes of individual Americans may not correspond with their views about or support for public policies or practices. Loury (2002) draws a useful distinction between the “social meanings” associated with racial classification and the “racial attitudes” held by individuals. Social meanings refer to the unexamined beliefs that influence how citizens understand and interpret the images they glean from the larger social world. For example, the meaning of a policy regarding job preference may be sensitive to the race of those affected: Veterans have been seen as acceptable beneficiaries, whereas the application of such a policy for blacks has been thought to violate meritocratic principles (Skrentny, 1996). Likewise, views about welfare policies may depend on the race of local recipients, so that more blacks being on the local rolls is associated with greater hostility toward recipients (Luttmer, 2001).
Sniderman and Piazza (1993) illustrate the difference between racial attitudes and racial meanings in their “mere mention” experiment. As noted by Loury (2002), these survey researchers found that “the mere mention” of affirmative action, in the context of soliciting from white respondents their views about racial stereotypes, made those whites more likely to agree with negative racial generalizations (e.g., most blacks are lazy). Compared with two groups of similar whites, the ones to whom affirmative action was “merely mentioned” showed a significantly higher tendency to affirm negative stereotypes about blacks than did those to whom affirmative action was not mentioned at all. The researchers concluded that the respondents’ expression of these anti-black sentiments had been “caused” by their dislike of affirmative action, and not the other way around. That is, whites were expressing primarily their ideological views about policy, which, when affirmative action was mentioned, then spilled over to affect their views about race.7
Regardless of how it is conceptualized, the discrepancy between the acceptance of nondiscrimination in principle and the discomfort about its implications in practice can be traced, at least in part, back to the persistence of anti-black stereotypes. In the mid-1980s, 61 percent of whites nationwide said that blacks on welfare could get a job “if they really tried,” 42 percent said that black neighborhoods are more rundown because blacks “don’t take care of their property,” and 43 percent endorsed the view that blacks would be as well off as whites “if they would just try harder” (Sniderman and Piazza, 1993; Loury, 2002). As of 1991, moreover, 34 percent of whites described blacks as “lazy” and 21 percent labeled them as “irresponsible” (Sniderman and Piazza, 1993). Likewise, according to the 1993 Multi-City Study of Urban Inequality, whites perceived blacks to be significantly less intelligent, less rich, less self-supporting, and less easy to get along with than Asians, Hispanics, or whites (Bobo and Massagli, 2001), and whites perceived African Americans as the least desirable potential neighbors (Charles, 2001).