Does Race Matter in Neighborhood Preferences? Results from a Video Experiment

Abstract

Persistent racial residential segregation is often seen as the result of the preferences of whites and blacks: whites prefer to live with whites while blacks wish to live near many other blacks. The origin of these preferences and their social psychological underpinnings are hotly debated. Are neighborhood preferences colorblind or race-conscious? Does neighborhood racial composition have a net influence upon preferences or is race a proxy for social class? If preferences are race-conscious, is this more a matter of a desire to be in a neighborhood with one’s “own kind” or to avoid being in a neighborhood with another racial group?

We tested the racial proxy hypothesis using an innovative experiment that isolated the net effects of race and social class and followed it with an analysis of the social psychological factors associated with residential preferences. Face-to-face surveys using computer assisted interviewing were conducted with random samples of Detroit and Chicago residents. Respondents were asked how desirable they would rate neighborhoods shown in videos in which racial composition and social class characteristics were manipulated and they also completed—via computer assisted self-interviews—questions tapping into perceptions of discrimination, racial and neighborhood stereotypes, and in-group identity.

We find that net of social class, the race of a neighborhood's residents significantly influenced how it was rated. Whites said the all-white neighborhoods were most desirable. The independent effect of racial composition was smaller among blacks and blacks identified the racially mixed neighborhood as most desirable. Hypotheses about how racial group identity, stereotypes, and experiences of discrimination influenced the effect of race of residents upon neighborhood preferences were tested and show that for whites, those who hold negative stereotypes about African Americans and the neighborhoods where they live are significantly influenced by neighborhood racial composition. None of the proposed social psychological factors conditioned African American sensitivity to racial composition of neighborhoods.

INTRODUCTION

How people feel about living in integrated neighborhoods has been the subject of scholarly attention, largely because residential preferences are implicated as a central cause of persistent segregation (Farley et al. 1994; Bobo and Zubrinsky 1996; Clark 1991; Charles 2006). For the most part studies of residential preferences find that whites are willing to live with only a handful of African American neighbors (some put the figure at around 20 percent), while African Americans are open to quite a diverse range of neighborhoods, though a “50-50” neighborhood is routinely identified as the most attractive and all-black and especially all-white neighborhoods as the least attractive (e.g., Farley et al. 1994; Clark 1991, 1992; Bobo and Zubrinsky 1996; Charles 2006).¹ While studies typically focused on identifying the levels of racial integration or segregation that people find most desirable, recent analyses investigated what factors shape these preferences (Farley et al. 1994; Bobo and Zubrinsky 1996; Emerson et al. 2001; Harris 2001; Krysan 2002; Krysan and Farley 2002; Charles 2006). Some ask whether race matters at all in reactions to neighborhoods, or if it is instead neighborhood social class characteristics being masked by racial considerations (Emerson et al. 2001; Harris 1999, 2001; Taub et al. 1984; St. John and Bates 1990). Others ask whether race-specific preferences about neighborhoods are shaped by in-group preferences or out-group avoidance (Krysan and Farley 2002; Charles 2000; 2006; Bobo and Zubrinsky 1996; Farley et al. 1994). In short, if and how race matters in neighborhood preferences is the subject of contentious debate.

The solutions to the persistently high levels of racial residential segregation in our nation’s metropolitan areas (Lewis Mumford Center 2001) are quite different if underlying forces are class-, and not race-based and if race-based concerns are driven by “neutral” in-group preferences and not out-group prejudices or concerns about discriminatory treatment by an out-group. For example, if neighborhood social class characteristics are the only ones that matter, then the potential for future integration may hinge on reassuring current and future residents of the stability and high quality of the amenities of an integrated neighborhood. But if, despite equivalent social class characteristics, the race of a neighborhoods’ residents influences reactions to it, then “simply” fixing up the neighborhood will be insufficient to maintain demand for the neighborhood from people of all races. And how race matters has equally important policy implications. If racial residential preferences are driven less by racial prejudice and experiences with discrimination, and more by a “neutral” in-group preference, then the policy implications look quite different. In the case of the former, efforts would need to be made to reduce racial prejudice, eliminate racial discrimination, and in the case of some neighborhoods, overcome the real and/or perceived hostility toward residents who are from racial minority groups.

At the core of our analysis are two research questions: (1) Are neighborhood preferences colorblind or race-conscious?; and (2) If preferences are race-conscious, is this more a matter of a desire to be in a neighborhood with one’s “own kind” or to avoid being in a neighborhood with another racial group? Past research addressed these questions—usually separately—using a variety of methodological approaches including in-depth interviews (Feagin and Sikes 1995), open ended follow-up questions (Krysan 2002; Krysan and Farley 2002); multivariate regression analyses (Charles 2006; Bobo and Zubrinsky 1996; Harris 2001; Krysan and Farley 2002; Farley et al. 1994); and factorial experiments (Emerson et al. 2001; St. John and Bates 1990; Shlay and DiGregorio 1985).

In this paper, we develop a new measurement tool, video-CASI (Computer Assisted Self Interview) (Couper 2005), that embeds an experiment within a standard household survey designed to test the independent effect of the racial composition of a neighborhood on residential preferences, controlling for the class composition of the neighborhood. Specifically, we ask: Do people respond to racial cues in a neighborhood or are they shaped solely by its evident social class characteristics? After addressing the question if race matters, we then ask how it matters. What are the social psychological factors underlying racial residential preferences? Are whites’ and African Americans’ racial residential preferences driven by a desire to be around their own racial group, or a wish to exclude members of another racial group? Our survey includes a rich array of racial attitudes—in a self-administered format—permitting a test of the key theoretical explanations for racial residential preferences: in-group racial identity, racial stereotypes, and perceptions of racial discrimination. Thus, we improve on existing studies because we can both rigorously test class- versus race-based reactions to neighborhoods and then use the results of the experiment to test hypotheses about how race matters in shaping residential preferences.

BACKGROUND

Racial residential segregation declined modestly over the last three decades, with the largest drops occurring in the smaller and newer metropolises of the South and West (Logan et al. 2002). Residential segregation persists at high levels in places like Detroit, Chicago, and other older metropolises in the Midwest and Northeast and is a “structural linchpin” of racial inequality (Bobo 1989). Numerous studies published in the 1990s documented the deleterious consequences of residential segregation for outcomes among blacks including infant and adult mortality, educational attainment, employment, death rates from homicide, rates of single motherhood, and the accumulation of equity in homes (LaViest 1989, 1993; Polednak 1990, 1993; Peterson & Krivo 1993, 1999; Hart et al. 1998; Collins & Williams 1999).

Demographers and sociologists focus on three main explanations for segregation. The first is that segregation is the outcome of racial differences in economic status. While studies using 1960–1980 census data refuted that hypothesis (Denton and Massey 1988; Farley 1977), more recent analyses suggest conflicting answers to this question and several identify a modest role for economics (Massey and Fischer 1999; Darden and Kamel 2000; Alba et al. 2000; St. John and Clymer 1999; Krivo and Kaufman 1999; Charles 2006). A second explanation is that segregation is the outcome of discriminatory practices in the housing market which persist, albeit in subtle forms (e.g., Munnell et al. 1996; Ross & Yinger 2002; Turner et al. 2002). A third key explanation emphasizes preferences: in essence, people live in segregated areas because they choose to do so. This explanation is the subject of much analysis with more recent attention focused on what these preferences mean. That is, whether they are in fact driven by race, or instead by social class characteristics.

Historically, the preferences explanation for segregation focused on whites’ desire to avoid living with blacks because of racial prejudice (Pettigrew 1973; Massey and Denton 1993). As such, race, per se, drove white preferences which, in turn, contributed to overall patterns of racial residential segregation. More recent arguments contend that white reluctance to living with blacks is rooted, not in antipathy toward African-Americans, but out of a fear of the liabilities whites associate with integrated neighborhoods, such as crime, deterioration, and the decline of property values (Clark 1986, 1988, 1989; Harris 1999, 2001; Ellen 2000; Taylor 1979; Taub et al. 1984). According to this argument, it is not race itself that drives preferences; instead, race is a proxy for social class. Others challenge this interpretation, arguing that racial composition persists as an important independent factor shaping preferences (Shlay and DiGregorio 1983; St. John and Bates 1990; Emerson et al. 2001).

It has been difficult to determine whether or not race, per se, matters because of how racial residential preferences have been measured. In the mid-1970s, Farley and his colleagues (1978) asked people how they felt about living in neighborhoods with people of different races by showing cards that illustrated various racial compositions and then asking how they would feel about living in such neighborhoods. This approach has been criticized because only a neighborhood’s racial composition was explicitly described and therefore it left unstated other neighborhood features. Thus the effect of race, per se, may be over-stated because of its conflation with social class (e.g., Harris 1999, 2001). Although the results of these studies indicated that people responded differently to the various levels of integration in a neighborhood, it was not clear whether racial composition was a proxy for social class characteristics. To understand, then, if race matters—independent of social class—a different methodological approach is necessary.

One approach for disentangling the effects of race and class on neighborhood preferences uses the neighborhood housing cards as its dependent variable and includes, among other predictor variables, measures of perceived racial differences in social class (Bobo and Zubrinksy 1996; Charles 2006). These studies report little effect of perceived social class or status differences on whether or not whites, blacks, Asians, or Latinos are open to or averse to sharing their neighborhoods with other racial/ethnic groups.

A second approach uses current neighborhood satisfaction as a dependent variable, and includes among its independent variables both the racial composition of the neighborhood and a range of social class characteristics (Stipak and Hensler 1983; Harris 2001). These studies conclude that differences in the social class characteristics of white and black neighborhoods explain much of the effect of racial composition. They also found that both whites and blacks express a desire to avoid high percentages of blacks.

A third approach uses factorial experiments that randomly assign individuals to evaluate different kinds of hypothetical neighborhoods as described in short vignettes. The vignettes are constructed in such a way that race and class characteristics can be isolated for their independent effects (Shlay and DiGregorio 1985; St. John and Bates 1990; Emerson et al. 2001). This technique does not suffer from the collinearity problems associated with using actual neighborhood satisfaction (St. John and Bates 1990). That is, because of high levels of residential segregation, studies of the effect of racial composition on actual neighborhood satisfaction rarely include information from white respondents living in largely black neighborhoods or black respondents living with many white neighbors.

The vignette studies confirm the independent effect of racial composition on neighborhood preferences, even after controlling for neighborhood features such as crime, deterioration levels, housing quality, and social environment (St. John and Bates 1990; Shlay and DiGregorio 1985; Emerson et al. 2001). For example, St. John and Bates (1990) find that the higher the percentage black in the neighborhood, the lower the evaluation by whites. For African Americans, increases in the percent black resulted in increasing levels of desirability only to a point, after which desirability dropped as the percentage black continued to increase. The approach used in these vignette studies is powerful, but raises questions about the mundane realism of the experiment, the possibility of an effect of social desirability, given the explicit nature of the questioning about race, and concerns about the high cognitive demands of imagining a neighborhood with multiple dimensions while answering a survey over the telephone. In addition, vignette studies have not included an array of racial attitude items so as to permit exploration of what social psychological factors underpin the influence that racial composition has on reported preferences. In the case of Emerson et al. (2001), the study was also limited to white respondents only.

Another key debate in the research on racial residential preferences focuses on the factors underlying racial residential preferences: Do preferences for segregation reflect a desire to be with one’s own group (Clark 1986; Thernstrom and Thernstrom 1997) or are they rooted in a desire to stay away from other racial groups (e.g., Charles 2006; Farley et al. 1994)? If we find that race, per se, influences residential preferences, then we can ask whether these racial preferences are driven by a desire to avoid a particular racial group out of prejudice (Charles 2006; Krysan 2002; Farley et al. 1994), or concerns about discrimination (Krysan 2002; Feagin and Sikes 1995), or whether preferences are due mainly to in-group attraction emanating from ethnocentrism (Clark 1991).

Research on the social psychological predictors of racial residential preferences has taken several approaches, all of which rely on the traditional showcard technique or a more modern version which nevertheless still focuses on racial composition without controlling for social class considerations. To date, the empirical and theoretical work on this topic has focused on three social psychological predictors of racial residential preferences: stereotypes, experiences of discrimination, and in-group identity.

Stereotypes

Several studies, all using results of survey questions asking whites explicitly about their racial residential preferences, examined the impact of racial stereotypes on racial residential preferences (Krysan 2002; Charles 2006; Farley et al. 1994). These studies report that racial stereotypes—as measured by a modern stereotype index (in which respondents evaluate different racial groups along several dimensions, such as hard-working or lazy; intelligent or unintelligent)—have a strong effect on whites’ expressed preferences for racially diverse or segregated neighborhoods. Whites who endorse negative racial stereotypes about blacks desire fewer African Americans in their neighborhoods than those who reject these stereotypes.

Ellen (2000) proposes an additional element of stereotyping that may drive segregation. Her neighborhood stereotyping hypothesis argues that white reluctance to move into an integrated—or integrating—neighborhood is driven by stereotypes about how the neighborhood will change. Krysan (2002) found indirect support for Ellen’s (2002) hypothesis in her analysis of open-ended follow-up questions to the traditional showcard question measuring racial residential preferences. Analyzing the reasons whites gave when they indicated they would move out of a neighborhood that became too heavily black, expressions of explicit racial prejudice and negative stereotypes about African Americans as a group existed, but were not widespread. However, negative stereotypes about integrated or all-black neighborhoods were very common. These neighborhood stereotypes included whites’ perception that integrated or black neighborhoods had problems with crime, property upkeep, and especially with declining property values, thus invoking many of the social class arguments of the racial proxy explanation. This analysis highlights the potential role of both stereotypes about groups and stereotypes about the neighborhoods in which groups live; a distinction missing in the quantitative models that have tested the effect of white stereotyping on preferences.²

Discrimination

An important factor shaping African American racial residential preferences are concerns about possible discrimination in whiter neighborhoods, as suggested by qualitative interview data and open-ended survey questions (Feagin and Sikes 1995; Krysan and Farley 2002). But quantitative analyses investigating the impact of experiences with discrimination have shown relatively modest and inconsistent effects of experiences with discrimination on residential preferences (Krysan and Farley 2002). And the statistical models seeking to explain variation in black racial residential preferences have been noteworthy for their relative lack of success: the explained variance has been modest at best. One limitation of the Krysan and Farley (2002) study was that it tapped just a few dimensions of discrimination (e.g., discrimination in the job market and the degree to which African Americans thought blacks missed out on good housing because of discrimination by real estate agents, banks, and individual whites), but not measures of experiences with housing discrimination specifically.

In-group Identity

The two preceding theoretical explanations about why whites and blacks would prefer to reside with greater numbers of co-ethnics focused on the perceptions of each group toward the other group. The third theoretical perspective turns attention to the in-group, and argues that racial residential preferences are driven less by a desire to avoid another group, and more out of an attraction to one’s own group. This argument suggests that whites and blacks alike prefer their own group, out of a sense of shared cultural experience and positive feelings toward it, quite distinct from any negative affect or perception of the out-group (Clark 1991; Thernstrom and Thernstrom 1997). To date, empirical assessments—using both multivariate regression models (Bobo and Zubrinsky 1996; Krysan 2002; Krysan and Farley 2002; Charles 2006) and analyses of open-ended survey data (Krysan 2002; Krysan and Farley 2002)—fail to find convincing evidence of the role of in-group identity in shaping racial residential preferences. With respect to the quantitative analysis, however, the indicator of in-group identity has been largely limited to a single survey question focusing on the perception that one’s fate was linked to the fate of one’s racial group.

Understanding the distinction between race-versus class-based motivations of neighborhood preferences and between social psychological factors focusing on the in-group versus the out-group is important both theoretically and for the development of policy. For example, if neighborhood social class characteristics are the only ones that matter, then the potential for future integration may hinge on reassuring current and future residents of the stability and high quality of the amenities of an integrated neighborhood. But if, despite equivalent social class characteristics, the race of a neighborhoods’ residents influences reactions to it, then “simply” fixing up the neighborhood will be insufficient to maintain demand for the neighborhood from people of all races. And how race matters has equally important policy implications. If racial residential preferences are driven less by racial prejudice and experiences with discrimination, and more by a “neutral” in-group preference, then the policy implications look quite different. In the case of the former, efforts would need to be made to reduce racial prejudice, eliminate racial discrimination, and in the case of some neighborhoods, overcome the real and/or perceived hostility toward residents who are from racial minority groups.

Our analysis, first, draws on a factorial experiment to answer the question: are individuals “color blind” when it comes to assessing a neighborhood’s desirability, or does race have an independent effect—beyond the observable social class characteristics—on perceived neighborhood desirability? The second section of our analysis engages the debate about what social psychological factors are associated with racial residential preferences. Using an innovative methodological tool, video-CASI (Couper 2005), combined with a broader array of measures of discrimination, stereotypes, and in-group identity than has been available in prior datasets, we learn much about if and how race shapes neighborhood preferences in two large metropolitan areas in the United States. Specifically, we test the following hypotheses:

• H1Controlling for social class of a neighborhood, white respondents will rate as most desirable neighborhoods with white residents only and evaluate as least desirable neighborhoods with black residents only. The desirability levels for the racially mixed neighborhood will fall in between these two extremes.
• H2Controlling for social class of a neighborhood, whites who hold negative stereotypes about blacks as a group or about neighborhoods with black residents, will be more influenced by the racial composition of a neighborhood than whites who do not hold such stereotypes.
• H3Controlling for social class of a neighborhood, whites who hold a stronger in-group identity will be more influenced by the racial composition of a neighborhood than will whites who do not strongly identify with whites as a group.
• H4Controlling for social class of a neighborhood, African American respondents will rate as more desirable a neighborhood that is racially mixed (black and white) as compared to the all-black and all-white neighborhoods. However, the all-black neighborhood will be evaluated as more desirable than the all-white neighborhood.
• H5Controlling for social class of a neighborhood, African Americans who report and perceive more discrimination will be more influenced by the racial composition of a neighborhood than will blacks who report and perceive less discrimination.
• H6Controlling for social class of a neighborhood, African Americans who report a stronger in-group identity will be more influenced by the racial composition of a neighborhood than will blacks who report a weaker in-group identity.

DATA AND METHODS

This analysis is based on data gathered in face-to-face surveys conducted in metropolitan Chicago and Detroit. The Detroit Area Study (DAS) and the Chicago Area Study (CAS) were multi-stage area probability samples of adults 21 years and older living in households in the Detroit Tri-County Area (Wayne, Oakland and Macomb Counties) in Michigan, and in Cook County (which includes Chicago), Illinois. The two areas were first stratified by racial/ethnic composition, using Census 2000 counts, so that over-samples could be drawn of African Americans, Latinos (in Chicago), and those living in racially mixed neighborhoods.³ A total of 734 completed interviews were obtained in Detroit, for an unweighted response rate (AAPOR RR2) of 56%. In Chicago, there were 789 completed interviews with a 45% unweighted response rate. All analyses use a weight that incorporates a selection weight and an adjustment for nonresponse (which was the inverse of the response rate in each primary sampling unit). Although Latinos were over-sampled in Chicago, due to sample size limitations, we restrict this analysis to non-Latino whites and non-Latino blacks.

Interviews were conducted from April through October 2004 in Detroit, and from August 2004 through August 2005 in Chicago. Identical survey instruments were used in the two sites, save for minor changes to questions that were specific to each area. The survey was conducted primarily as a computer assisted personal interview (CAPI), but in the middle of the interviewer-administered survey, the laptop computer was turned over to the respondent and a section of the survey became Video-CASI (Computer Assisted Self Interview) (Couper 2005). During this time, the respondent viewed four or five 27 to 44 second neighborhood videos (with an average duration of 35 seconds) and answered, in private using the computer keyboard, several questions about each neighborhood they had viewed. Their responses to a question about overall desirability of each neighborhood they viewed are the dependent variables for this analysis.

A major innovation in this study is that rather than rely on respondents to imagine neighborhoods of varying racial compositions and social class characteristics, as done in previous vignette experiments on residential preferences, we present short videos that show them actual neighborhoods. We created 13 videos that were shown in both Chicago and Detroit. Samples of the actual videos used can be viewed at http://www.psc.isr.umich.edu/tmp/das/.⁴ The videos included five different social class levels, which we refer to as: lower working class, upper working class, blemished middle class, unblemished middle class, and upper middle class. As St. John and Bates (1990) note:

People gain a picture of neighborhood lifestyle and behavior from observing visual cues in neighborhoods, such as housing type and maintenance and general neighborhood upkeep. Consequently, the environmental cleanliness and housing quality dimensions probably serve as cues for impressions of social class (p. 52).

We selected different neighborhoods to convey the different social class levels, relying on this assumption that respondents infer social class based on features such as home and property size, upkeep of the houses, and other cues gleaned from observations. Each of the different neighborhoods had, in turn, three variants in terms of the race of the individuals shown in the neighborhood:⁵ (1) all residents are white; (2) all residents are black; (3) three residents are white and two residents are black. The key independent variable (neighborhood racial composition) was manipulated by randomly assigning respondents to view different racial compositions of what were otherwise exactly the same neighborhoods.

The videos were filmed specifically for this research. Five “residents” were shown in each video. These residents were actors hired by the research team; however, throughout the paper we will refer to them as “residents.” Across each condition (that is, racial composition) the residents were shown doing exactly the same thing—though the activities varied across the neighborhoods of different social classes. The only difference for a neighborhood of a given social class was the race of the people doing the activities. The residents were shown performing routine activities such as walking down the sidewalk, standing in the driveway, chatting with neighbors on the front lawn, or getting their mail. In each neighborhood, there was one scene where three individuals were shown together talking in the driveway, front yard, at the mailbox, or surrounding a car that was being repaired. Residents wore short sleeve shirts and no hats to increase the likelihood that the respondents could detect their racial/ethnic identity. Residents within each neighborhood social class level were matched on approximate age, gender, and dress style, so that the only difference readily apparent to respondents was their race.

Prior to fielding the survey, we showed the videos to focus groups to assess whether people could detect the intended race/ethnicity of the residents and to validate that our neighborhoods were of five distinct social class levels. There was consensus about the racial composition of the black, white, and mixed black-white neighborhoods: focus group participants identified the intended racial composition of the neighborhoods. In addition, the focus group participants agreed with the ranking of the neighborhoods on social class levels intended by the investigators.

Respondents viewed four videos in Detroit and five videos in Chicago.⁶ The first video each respondent saw was the upper working class neighborhood with no people visible; then each respondent viewed the lower working class, blemished middle class,⁷ unblemished middle class, and finally the upper middle class neighborhoods, in that order.⁸ Respondents were randomly assigned to the racial composition they saw for each of the different social class levels. Racial composition varied across each video for every respondent; in other words, each respondent saw only one neighborhood with black residents; one neighborhood with white residents; and one with a mix of black and white residents. After viewing each neighborhood on the computer screen, respondents were asked questions using CASI, including our measure of racial residential preferences: “Overall, how would you rate the neighborhood you just saw as a place to live?” (1=very undesirable; 7=very desirable).

This factorial experiment forms the foundation of our answer to the first research question—asking whether race, independent of social class, influences neighborhood preferences. Our method builds on existing factorial research in several ways. First, instead of presenting vignettes as verbal descriptions, respondents were shown a visual image of an actual neighborhood. This mirrors what individuals do when searching for a place to live—they drive around and look at neighborhoods (Krysan, 2008). In addition, the manipulation of race of residents in our study is more subtle than in prior vignette studies. In the videos, the residents appear only briefly, and are not explicitly mentioned in the survey questions. The residents appear almost incidental to the main focus of the video, which are the homes and yards that make up the neighborhood. Finally, the cognitive demands of vignette studies—particularly in a telephone interview study—are substantial. Over the telephone, respondents were asked to imagine a neighborhood and keep in mind six or seven different elements of that neighborhood. In our study, respondents were simply asked to react to a clear visual image of a real neighborhood and then answer questions about it.⁹ Finally, the most sophisticated of the vignette studies (Emerson et al. 2001) focused exclusively on white respondents; in our study, we test hypotheses for both African Americans and whites.

Most studies of racial attitudes and residential preferences rely on either telephone or face-to-face interviewing. Our respondents, by contrast, viewed the videos in private, and answered the questions about neighborhoods directly on the computer, thus minimizing social desirability pressures (Krysan 1998). The interviewer never knew how respondents answered any of the questions. Although there may still be social desirability pressures created by the presence of an interviewer, our private reporting environment likely increased the level of honest reporting on sensitive topics. In addition, the neighborhood videos came early in the survey, prior to any questions with explicit racial content.

Our analyses exploit the within-subjects experimental design of this study through the use of hierarchical linear models. We use the PROC MIXED procedure in SAS, with residual maximum likelihood (REML) estimation (West et al., 2007). Our dependent variables are the ratings of desirability of the four neighborhoods—and thus our unit of analysis is the video. Given that each respondent saw and rated the same baseline video—an upper working class neighborhood with no residents—we include the ratings of this neighborhood as a respondent-level control. In this way, the ratings of the remaining videos (with residents visible) can be viewed as adjusted for each respondent’s individual level of rating of the initial neighborhood video devoid of residents.

We fit a series of three-level models. The first-level predictors—those predictors with values that vary across conditions—are the videos’ experimentally-manipulated characteristics. These include the race of the residents—a series of dummy variables, “white”, “black”, and “mixed”, where “mixed” is the omitted category—and the social class of the neighborhood. This was also a series of dummy variables: “lower working class”, “blemished middle class”, “unblemished middle class”, and “upper middle class”, with “upper middle class” as the omitted category.

There are two sets of second-level predictors, that is, predictors with fixed values for the respondents. First is the respondent’s desirability rating of the first video, which is the evaluation of the upper working class neighborhood with no residents. The second set of second-level predictors is the social psychological indicators. ¹⁰ The exact question wording for all measures is in Appendix A.

At the third level, we include random effects associated with the sampling error computing units (SECUs) to reflect the primary stage selection in the sampling design (where sampled respondents are nested within SECUs, which are themselves comprised of census block groups). This adjusts the estimates to account for the clustered nature of the sample design. In addition, all analyses are weighted to reflect the stratified design and differential selection probabilities.

Because our hypotheses differ depending on the race of the respondent, we fit models separately for whites and African Americans.¹¹ Our analysis presents two models for whites and three for African Americans. The first model tests the overall effects of a neighborhood’s racial composition on respondents’ preferences, net of neighborhood social class characteristics (Hypotheses 1 and 4). The second model (and third in the case of African Americans) tests hypotheses about the social psychological factors that shape racial residential preferences (Hypotheses 2, 3, 5 and 6). We do the latter by testing the degree to which racial stereotypes (for whites), experiences with discrimination (for blacks), and in-group identity (for both whites and blacks) influence the effects of neighborhood racial composition on evaluations of the neighborhood’s desirability. This is accomplished using cross-level interactions between the first-level predictor (neighborhood racial composition) and the second-level predictors (social psychological measures).

RESULTS

Initial Analyses: Model Fit and Intra-Class Correlation

As an initial step, we fit an unconditional (or variance components) model without the fixed effects of any of the predictor variables to examine the components of variance at each of the three levels. We find that the SECU-level random effects—reflecting the clustering of respondents in block groups—account for a small proportion of the total variance (0.015 for whites and 0.012 for African Americans). The estimated proportion of total variation attributed to within-respondent variation—that is, differences in ratings across the videos—is about 0.90 for whites and 0.73 for African Americans. This is evidence that the major component of the evaluations respondents gave reflected what they saw in the videos, and did not reflect the characteristics of where they lived (which are assessed by the SECU effects).

Adding the characteristics of the videos (race of the residents in the neighborhood and social class of the neighborhood) to the base models adds significantly to the explained variance. Compared to the unconditional model, characteristics of the videos explain about 0.50 of the within-respondent variation for whites, and 0.46 for African Americans. These results show that both the race and the social class represented in the neighborhood videos significantly affected respondents’ ratings of the desirability of that neighborhood. But our interest is in the race-specific effects of neighborhood racial composition on desirability ratings, and we turn to those results now.

Does race matter in white neighborhood preferences?

Model 1 in Table 1 provides our test of Hypothesis 1, which takes the following form:

• DESIRABLE_ijk = γ₀₀₀ +γ₁₀₀ (VIDRACE = white) +γ₂₀₀ (VIDRACE = black) +γ₃₀₀ (VIDCLASS = LWC) +γ₄₀₀ (VIDCLASS = BMC) +γ₅₀₀ (VIDCLASS = UMC) +γ₀₁₀ (DESIRE_VID1) + r_0jk + u_00k + e_ijk

Table 1.

Ratings of Desirability of Video Neighborhoods, White Respondents Only (Standard Errors in Parentheses).

	Model 1	Model 2
Intercept	4.375 (0.126)**	4.393 (0.126)**
Video-level effects:
Video Class   Lower Working Class   Blemished Middle Class   Unblemished Middle Class   Upper Middle Class	−2.069 (0.063)** −1.442 (0.068)** −0.142 (0.069)** ---	−2.060 (0.062)** −1.460 (0.068)** −0.158 (0.069)* ---
Video Race   White   Black   Mixed	0.158 (0.057)** −0.247 (0.057)** ---	0.137 (0.057)* −0.260 (0.057)** ---
Respondent-level effects:
Rating of video 1	0.312 (0.028)**	0.313 (0.028)**
Stereotypes		0.068 (0.056)
N’Hood Stereotype		0.055 (0.053)
Feelings Toward Groups		−0.174 (0.056)**
Cross-level Interactions:
Stereotypes*White Stereotypes*Black Stereotypes*Mixed		−0.098 (0.065) −0.167 (0.065) ---
NhoodStereo*White NhoodStereo*Black NhoodStereo*Mixed		−0.088 (0.063) −0.163 (0.063)** ---
Feelings*White Feelings*Black Feelings*Mixed		0.228 (0.065)** 0.098 (0.065) ---
−2 Res Log Likelihood	6267.5	6269.1
Tests of fixed effects:
Video race Video class Stereotypes*video race NhoodStereoe*video race Feelings*video race	F=25.5, d.f.=2, 1199, p<.0001 F=468.3, d.f.=3, 1199, p<.0001	F=24.94, d.f.=2, 1193, p<.0001 F=468.3, d.f.=3, 1193, p<.0001 F=3.30, d.f.=2, 1193, p=.037 F=3.39, d.f.=2, 1193, p=.034 F=6.25, d.f.=2, 1193, p=.0020

^*p<.05;

^**p<.01

Video 1=Upper working class with no residents

The first finding is that whites are affected by the social class of the neighborhood: the neighborhood with the highest social class level was rated as the most desirable; the lowest was the least desirable. For example, where a score of seven meant very desirable and a score of one meant very undesirable, the adjusted mean desirability rating for the lower working class neighborhood was 3.51 as compared to the upper middle class neighborhood, which had a mean rating of 5.58 on the desirability scale, based on the coefficients in Model 1 (here and elsewhere, the adjusted means are not shown in the tables).

But we also find that race of the residents had a significant and independent effect—controlling for the social class of the neighborhood—on whites’ reported desirability of the neighborhood. The variable capturing the race of the neighborhood residents was significant, and the effects are in the hypothesized direction: the highest predicted mean desirability is for white neighborhoods (adjusted mean=4.85), followed by mixed neighborhoods (adjusted mean=4.69), and then black neighborhoods (adjusted mean=4.45).

What predicts which whites are influenced by racial composition when they rate neighborhoods?

We now test whether out-group perceptions and/or in-group identity affect whether racial composition shapes how whites rate the desirability of a neighborhood (Hypotheses 2 and 3). Model 2 in Table 1 reports the results for an equation that includes all of the predictor variables needed to test Hypotheses 2 and 3¹²:

• DESIRABILITY_ijk = γ₀₀₀ +γ₁₀₀ (VIDRACE=WHITE) +γ₂₀₀ (VIDRACE=BLACK) +γ₃₀₀ (VIDCLASS=LWC) +γ₄₀₀ (VIDCLASS=BMC) +γ₅₀₀ (VIDCLASS=UBMC) +γ₀₁₀ (DESIRE_VID1) +γ₀₂₀ (STEREOTYPES) +γ₀₃₀ (N'HOOD STEREOTYPE) +γ₀₄₀ (FEELINGS TOWARD GROUPS) +γ₁₂₀ (STEREOTYPES*VIDRACE=WHITE) +γ₂₂₀ (STEREOTYPES*VIDRACE=BLACK) +γ₁₃₀ (NHOODSTEREO*VIDRACE=WHITE) +γ₂₂₀ (NHOODSTEREO*VIDRACE=BLACK) / +γ₁₄₀ (FEELINGS*VIDRACE=WHITE) +γ₂₂₀ (FEELINGS*VIDRACE=BLACK) +r_0jk + u_00k + e_ijk

The measures to test Hypothesis 2 include two different kinds of racial stereotypes. The first stereotype scale (stereotype) measures the degree to which white respondents think blacks, as a group, have negative traits—that they are less intelligent, more likely to prefer to live off welfare, are involved in gangs and crime or do not supervise their children well, as compared to whites. ¹³ We hypothesize that whites who endorse negative stereotypes about African Americans will, more so than those who do not endorse stereotypes, be influenced by neighborhood racial composition. The items comprising the stereotype scale are described in Appendix A. The stereotype scale is standardized, with a mean of 0 and standard deviation of 1. To test Hypothesis 2, we include in our model a cross-level interaction.

Our hypothesis is supported by the data; the interaction between the stereotypes scale and the race of the video is statistically significant (F=3.3, d.f.=2, 1193, p<.05) in Table 1 (Model 2). The slope associated with the stereotype scale for the Mixed Neighborhood is 0.068. The net slope for the stereotype scale for the White Neighborhood is (.068 − .098) = −0.03, while that for the Black Neighborhood is (.068 − .167) = −0.099. This suggests that the negative slope for the stereotype scale on ratings of neighborhood desirability is stronger when the video showed black rather than white residents. Specifically, a standard deviation increase on the stereotype scale was associated with a reduction of only 0.03 in the rating of desirability for the video with white residents, but a much larger decrease of 0.10 for the video with black residents. Whites who held negative racial stereotypes about blacks were most likely to downgrade the desirability of black neighborhoods—while evaluations of the desirability of the mixed and the white neighborhood were unaffected by respondent’s level of stereotyping.

Our second single-item measure taps a different racial stereotype—this one about neighborhoods where blacks live rather than blacks themselves (NhoodStereo). This builds on Ellen (2000) and Krysan (2002) who argue that stereotypes about the neighborhoods in which blacks live are an important factor shaping racial residential preferences. Our measure asked respondents the degree to which they believe that “if a few black families move in, property values in a white neighborhood will decline”. When evaluating a neighborhood’s desirability, the slope of the neighborhood stereotype variable for the Mixed Neighborhood is 0.055. The net slope for the neighborhood stereotype variable for the White Neighborhood is (.055 − .088) = −0.033, while that for the Black Neighborhood is (.055 − .163) = −0.108. Thus, the negative slope for neighborhood stereotyping on ratings of neighborhood desirability is stronger when respondents saw black rather than white residents. A standard deviation increase on the perception that property values are likely to fall is associated with a decrease of 0.033 in the rating of desirability for the video with white residents, but a decrease of 0.108 for the video with black residents. Again, the interaction is statistically significant (F=3.39, d.f.=2, 1193, p<.05), as shown in Table 1.

Our results show that perceptions about the out-group (in this case blacks) are a significant predictor of whether or not racial composition, net of social class, will shape neighborhood desirability ratings. We now move to a test of Hypothesis 3. Our survey included three different measures of in-group identity, thus affording a fuller test of the role of in-group identity than has been possible in earlier tests of this hypothesis. Two of the measures were explicitly about whites’ views of their own racial group (how close they feel to whites, and how much they view their fate as being linked to the fate of other whites). Neither of these single-item measures (see Appendix A for question wording) shows an effect on neighborhood desirability either alone or in conjunction with the race of the residents portrayed in the neighborhood (analyses not shown). They are, therefore, not included in the final model (Model 2, Table 1).

Model 2, Table 1, shows the results for our third group identity measure, which taps the degree to which whites report feeling warmer toward whites than blacks. This was assessed using a difference score based on thermometer ratings toward whites and blacks. We use a standardized version of this “feeling score,” with a mean of 0 and a standard deviation of 1. Both coefficients (for the main effect and for the interaction between the feeling score and the race of the residents) are statistically significant, as shown in Table 1.

Specifically, the slope of the feeling score toward whites versus blacks for the Mixed Neighborhood is −0.174. The net slope of the feeling score for the White Neighborhood is (−.174 + .228) = 0.054, while that for the Black Neighborhood is (−.174 + .098) = −0.076. Thus, a standard deviation increase on the difference in the feeling score (indicating more positive feelings for whites than for blacks), is associated with a modest increase of 0.054 in the rating of the desirability of the neighborhood with white residents, but a 0.076 decrease in desirability for the neighborhood with black residents and a 0.174 decrease in desirability for the neighborhood with both white and black residents. In short, whites who felt more warmly toward whites than blacks were especially sensitive to changes in the racial composition of the neighborhoods—net of social class characteristics. And they are especially likely to find all white neighborhoods more desirable than all black or mixed neighborhoods.

Does race matter in African American neighborhood preferences?

We now turn to a parallel analysis for black respondents. Table 2, Model 1 shows that social class also matters for African Americans, and in the same way it does for whites: African Americans rated as most desirable the neighborhood that was of the highest social class (mean rating of 5.85); and the lowest social class neighborhood was rated as lowest (mean rating of 3.63). However, we also find that race of residents, independent of social class, shapes neighborhood preferences. That is, race of residents in the video is a significant (F=4.87, d.f.=2, 1128, p=.008) predictor—above and beyond social class. The mixed neighborhoods, as hypothesized, were rated as most desirable, followed by the black neighborhoods then the white neighborhoods. The individual contrasts (not shown) reveal that the evaluation of desirability of mixed and white neighborhoods are different (adjusted p=.0056) but not that of black and white (adjusted p=.160) or mixed and black (adjusted p=.42).

Table 2.

Ratings of Desirability of Video Neighborhoods, Black Respondents Only (Standard Errors in Parentheses)

	Model 1	Model 2	Model 3
Intercept	4.601 (.138) **	4.613 (0.14)**	4.600 (0.14)**
Video-level effects:
Video Class   Lower Working Class   Blemished Middle Class   Unblemished Middle Class   Upper Middle Class	−2.097 (.070) ** −1.194 (.077) ** −0.161 (.076) * ---	−2.095 (0.070)** −1.196 (0.077)** −0.154 (0.076)* ---	−2.090 (0.070)** −1.200 (0.077)** −0.160 (0.076)* ---
Video Race   White   Black   Mixed	−0.195 (.063) ** −0.080 (.063) ---	−0.185 (0.064)** −0.088 (0.065) ---	−0.195 (0.063)** −0.080 (0.063) ---
Respondent-level effects:
Rating of video 1	0.313 (.028) **	0.312 (0.028)**	0.314 (0.028)**
Discrimination		−0.019 (0.044)
In-Group Identity			0.038 (0.070)
Feelings Toward Groups			−0.021 (0.056)
Cross-level Interactions:
Discrimination*White Discrimination*Black Discrimination*Mixed		−0.039 (0.052) 0.031 (0.052) ---
In-Group*White In-Group*Black In-Group*Mixed			−0.120 (0.082) −0.040 (0.081) ---
Feelings*White Feelings*Black Feelings*Mixed			0.100 (0.065) −0.005 (0.065) ---
−2 Res Log Likelihood	6171.4	6182.6	6187.4
Tests of fixed effects:
Video race Video class Discrimination*video race In-Group*video race Feelings*video race	F=4.87, d.f.=2, 1128, p=.008 F=370.35, d.f.=3, 1128, p<.0001	F=4.13, d.f.=2, 1126, p=.016 F=370.9, d.f.=3, 1126, p<.0001 F=0.91, d.f.=2, 1126, p=.40	F=4.87, d.f.=2, 1124, p=.0078 F=366.4, d.f.=3, 1124, p<.0001 F=2.08, d.f.=2, 1124, p=0.12 F=1.65, d.f.=2, 1124, p=0.19

^*p<.05

^**p<.01

Video 1=Upper working class with no residents

What predicts which African Americans are influenced by racial composition when they rate neighborhoods?

Hypotheses 5 and 6 test whether perceptions of racial discrimination or in-group identity shape how neighborhood racial composition influenced overall desirability of the neighborhood for African Americans. These are both tested in Models 2 and 3 in Table 2. Unlike previous studies, ours included an expanded set of questions about perceptions of discrimination, and three items tapping in-group identity, rather than just a single item. Looking first at the impact of perceptions of racial discrimination, we find that experiences or perceptions of discrimination on the job, by police, in the neighborhood, and in the housing market did not differentiate between those who were and were not affected by the race of the residents in the video.¹⁴

Hypothesis 6 focused on two aspects of in-group identity.¹⁵ The first is a two-item scale that taps perceptions of the in-group and the respondent’s connection to it—either in the sense of linked fate, or in closeness and similarity of feelings and thoughts. African Americans with stronger in-group identity were no more or less likely to be affected by the racial composition of a neighborhood in their assessment of its desirability. A second measure that gauges African Americans’ warmth toward other African Americans—vis a vis their warmth toward whites—also revealed no significant effects on whether or not the respondent was influenced by a community’s racial composition. How blacks rated neighborhoods was influenced by the quality of housing and the race of residents but not by their perceptions of racial discrimination or their feelings about whites or their own identification with blacks.

DISCUSSION AND CONCLUSION

This study used a video vignette experiment to address questions about if and how race matters in neighborhood preferences. Our fundamental conclusion is that race, per se, shapes how whites and, to a lesser extent, blacks view residential space. Residential preferences are not simply a reaction to class-based features of a neighborhood; they are shaped by the race of who lives there. To be sure, a neighborhood’s social class matters.¹⁶ Both whites and African Americans evaluated upper middle class and middle class neighborhoods as much more desirable places than lower and upper working class neighborhoods. But controlling for social class characteristics did not eliminate the influence of racial composition, thus refuting the racial proxy hypothesis.

More specifically, for whites living in metropolitan Chicago and Detroit, neighborhoods portrayed as having only black residents were viewed less favorably than identical neighborhoods with either only white residents or a mix of white and black residents. Both the racially mixed and the all-black neighborhood were rated by whites as significantly less desirable than the all-white neighborhood. The presence of African Americans in a neighborhood resulted in a downgrading of its desirability.

Further analysis of how race matters, however, reveals some indirect support for the racial proxy hypothesis. But it is a racial proxy hypothesis with a caveat. Whites who believed that the arrival of a few black families into an all-white neighborhood would likely drive down property values were influenced by neighborhood racial composition. At the same time, traditional racial stereotypes—the belief that African Americans lack intelligence and prefer to live off welfare, for example—also shaped how whites were influenced by the racial composition of the neighborhood. The caveat to interpreting this as support for the racial proxy hypothesis is that, as Krysan (2002) has argued, the difference between these two kinds of stereotypes—for the purposes of their role in perpetuating racial residential segregation—is largely semantic. That is, it may matter quite little, for the perpetuation of segregation, if the stereotypes whites hold are about blacks themselves or about neighborhoods with black residents. If both kinds of stereotypes result in avoiding neighborhoods with African Americans, and/or moving from those neighborhoods that are experiencing an increase in African American populations, then both kinds of beliefs are segregation-promoting.

We included two dimensions of in-group identity, thus providing a test of the idea that ethnocentric tendencies—how one feels about one’s own group—were influential in shaping racial residential preferences. Whites who reported feeling particularly close to whites, and those who felt a common fate identity with other whites were no more or less likely to be influenced by a neighborhoods’ racial composition. In-group attraction, then, is not driving whites’ reactions to black neighbors. Our second in-group identity measure—tapping the degree to which whites felt more warmly toward whites than blacks—did influence the degree to which neighborhood racial composition shaped whites’ desirability ratings. Because this feeling score measure is comparative, it is at least as much about negative feelings toward the out-group as it is about positive feelings toward the in-group. The different results for these racial group identity questions thus points to the power of exclusion and anti-black feelings as driving whites’ residential preferences, as opposed to a closeness or attraction to one’s own group, i.e., ethnocentrism.

We emphasize another finding for whites: the effect of neighborhood racial composition and the social psychological factors underlying it, suggest that white and mixed neighborhoods were often indistinguishable. For example, racial stereotypes did not influence reactions to the all-white or mixed neighborhoods; only to the all-black neighborhood. There may be some cause for optimism in this finding. But this optimism is cautious since the racially mixed neighborhood was nevertheless rated as less desirable than the all-white neighborhood.

Turning to results for African Americans, we find that race matters as well. But it matters less strongly—and differently—than it does for whites. The differences in the magnitude of the effect of neighborhood racial composition on neighborhood evaluations for blacks versus whites is apparent in the mean differences in the levels of desirability across the three neighborhood types. That is, for white Chicago and Detroit area residents, the difference in overall ratings between the most desirable (white) neighborhood and the least desirable (black) neighborhood was 0.41 points; for African American Chicago and Detroit area residents, the difference between the most desirable (mixed) and the least desirable (white) was less than half that, or 0.13 points. Although African Americans are responsive to the racial composition of the neighborhood, the degree of their responsiveness is significantly less than that of whites.¹⁷

Also, race matters differently for African Americans. When neighborhoods have identical observable social class characteristics, it is the all-white neighborhood that was evaluated as least desirable by African Americans. The racially mixed and the all-black neighborhood evaluations were generally indistinguishable from each other and both were rated more highly than the white neighborhood. It is not the case, then, that African Americans evaluate all-black neighborhoods in the same way that whites do—once social class characteristics are controlled. Here our findings depart from prior studies contending that whites and blacks are both averse to having black neighbors (Stipak and Hensler 1983; Harris 2001).

But our results also depart from prior studies (Farley et al. 1994) that conclude the 50-50 neighborhood is the most attractive for African Americans. Once controlling for the neighborhood’s class characteristics, black respondents did not distinguish between the all-black and racially mixed neighborhood. This raises the possibility that earlier studies have over-stated blacks’ lack of interest in all-black neighborhoods. But, it is also the case that we did not find that black respondents found all-black neighborhoods to be significantly more desirable than racially mixed neighborhoods. To some extent, both this pattern, and the more general finding that the effect of racial composition on blacks’ desirability ratings was small (as compared to whites), is consistent with recent studies arguing that African Americans’ residential preferences are more flexible and fluid than is often asserted (Krysan and Bader 2008; Krysan and Farley 2002). Finally, the finding that the all-white neighborhoods were least desirable for African Americans is nevertheless consistent with prior research showing that being a pioneer in an all-white neighborhood is an unattractive option among African Americans (Farley et al. 1994).

It is worth pointing out that although our video vignette methodology has several strengths relative to prior measures of residential preferences, a shortcoming is that we cannot pinpoint the levels of segregation or integration that whites and blacks find most desirable. Specifically, our mixed neighborhood was a majority white neighborhood (three out of five residents were white). But given the subtlety of the experimental manipulation of racial composition, it is not clear whether this neighborhood was interpreted as having “token” black representation or nearly “equal” black representation. Thus, although we have evidence from other studies that whites typically say they would be comfortable living in 20 percent African American neighborhoods and a higher representation of blacks makes them more uncomfortable, it is unclear whether our respondents interpreted the mixed neighborhood as falling into that range. Similarly, we cannot pinpoint African American’s preferred levels of integration, since we had just three different options. It is noteworthy, though, that the mixed neighborhood we portrayed—which blacks in Detroit and Chicago largely evaluated as comparable to the all-black neighborhood—had a slight majority white population (3 out of 5 residents). Again, we cannot be sure how African Americans inferred the overall racial composition of this neighborhood; but it is clear that the presence of some African Americans in the neighborhood was sufficient to differentiate the evaluations of it as compared to the all-white neighborhood.

Our study also sought to better understand the social psychological factors underlying African American racial residential preferences. Prior research highlighted the importance of experiences with discrimination; but our study finds little support for this. Neither experiences with housing discrimination, perceptions of the effect of housing discrimination on residential segregation, nor discrimination experiences outside the realm of housing had any predictive power in explaining which blacks were or were not influenced by neighborhood racial composition in their desirability ratings. At the same time, contrary to speculations about the causes of racial residential segregation that focus on the role of African American in-group identity as an important factor shaping black residential preferences (Clark 1991; Patterson 1996; Thernstrom and Thernstrom 1996), we found that none of our measures of in-group identity were statistically significant predictors of African American reactions to racial composition.¹⁸

We leveraged methodological innovations related to how we asked and measured racial attitudes and neighborhood preferences to shed substantive light on important questions related to racial residential segregation in general, and residential preferences in particular. Our findings demonstrate that race shapes perceptions of neighborhood desirability for both whites and blacks—even when that “space” is identical on all other dimensions. These results mirror those reported in the labor market by recent field experiments examining the impact of applicant’s race on employer hiring decisions in both entry-level jobs in Milwaukee (Pager 2003) and white collar occupations in Boston and Chicago (Bertrand and Mullainathan 2004). In both studies, the findings demonstrated that even controlling for a variety of demographic and labor market characteristics, being black—or having a “black” first name—significantly reduced the chances that an applicant would be called back by the prospective employee.

The racial proxy and related neighborhood stereotyping hypotheses are not an accurate answer to the question of what underlies responses to neighborhood racial composition, and this is particularly clear for whites. To be sure, we found that whites who believe that property values fall when a black family moves into the neighborhood are more influenced by neighborhood racial composition. But we also found that traditional negative group stereotypes shaped these reactions. And, of course, the overall finding is that the racial composition of the neighborhood did matter significantly, even when observable social class characteristics were identical. All of this suggests that race matters, and it is not just a function of class-based concerns for which race serves as a proxy. Policy solutions to address racial residential segregation, then, need to bear in mind that “simply” fixing up the neighborhood is not sufficient to overcoming whites’ reluctance to move into, or remain in, integrating neighborhoods. And the finding that racial stereotypes—both of the neighborhood and of blacks themselves—operate to shape white responses to integrated neighborhoods means that efforts to dispel these stereotypes are also important.

Measures of racial residential preferences in the past can be rightly criticized for the possibility that race was conflated with social class characteristics. In this video experiment, we addressed this criticism by holding the physical characteristics of neighborhoods constant and providing rich visual cues about neighborhood quality as a way to disentangle the effects of race and class on residential preferences. That whites are influenced by neighborhood racial composition suggests that whites are not, in fact, color blind but class conscious. Rather, they are both class- and color-conscious.

APPENDIX A: Variable Descriptions

Stereotype Scale (STEREO) is comprised of four items in which white respondents were asked to place whites and blacks, as groups, at specific points on a seven point scale. The numerical values for this scale were generated by taking the difference between the score the respondents gave whites as a group versus the score they assigned blacks as a group. Respondents with high values on this scale endorsed negative stereotypes about African Americans, vis a vis whites. The four characteristics were “tends to be intelligent” versus “tends to be unintelligent”; “tends to prefer to live off welfare” versus “tends to prefer to be self-supporting”; “tends to be involved in street crime or gangs” versus “tends not to be involved in street crime or gangs”; and “tends to do a good job supervising their children” versus “tends to do a bad job of supervising their children”. The scale is comprised of a sum of the differences for all four of these items; the scale is then standardized to a mean of 0 and a standard deviation of 1.

Perceptions of Property Values and Race (PROPFALL) is a single question asking respondents: “When a few Black families move into an all-White area, how likely is it that property values will go down? Would you say it is…?” The response options were 1 = not at all likely; 2 = not very likely; 3 = somewhat likely; and 4 = very likely.

Feelings Toward Groups (FEELOTH) is constructed by subtracting how warmly a respondent feels toward their own group (on a 100 point feeling thermometer) from how warmly they feel toward the other race (on a 100 point feeling thermometer). The scale has been standardized to a mean of 0 and a standard deviation of 1.

Common Fate Identity (FATE) is a single question asking respondents: “How much do you think what happens generally to [R’s Race] in this country will affect you?” The response options were 1=not at all; 2=not very much; 3=some; 4=a lot.

Close Race (CLOSRACE) is a single question asking respondents “How close are your ideas and feelings to the ideas and feelings of other [R’s race] people?” 1=not at all close; 2=slightly close; 3=somewhat close; 4=pretty close; 5=very close; 6=extremely close.

Experiences with Housing Discrimination is a summary scale derived from four different questions asking respondents about their experiences with housing discrimination. Respondents were asked if any of the following had happened to them: (1) Have you ever felt that you were denied housing because the landlord or a real estate agent didn’t want to sell or rent to you because of your race or ethnicity? (2) Have you ever felt that a real estate agent was showing you only homes in certain neighborhoods because of your race or ethnicity? (If the Respondent never used a real estate agent, he/she was given the value of 0); (3) Have you ever lived in a neighborhood where neighbors made life difficult for you or your family because of your race or ethnicity?; (4) At any time when you were applying for a mortgage loan, have you ever felt that you were treated unfairly because of your race or ethnicity? (If the Respondent never applied for a mortgage loan, he/she was given the value of 0). The values on the scale range from 0 to 4, with a four meaning they had experienced all four kinds of discrimination. Missing values were imputed so there are no missing data.

Experience with Job Discrimination is measured by the response to a single question asking: “Have you ever felt that you had been unfairly denied a promotion at your job because of your race or ethnicity?” (yes/no).

Experience with Police Discrimination is measured by the response to a single question asking: “Have you ever felt that you had been unfairly stopped, searched, or questioned by the police because of your race or ethnicity?” (yes/no)

Discrimination Causes Segregation is a summary scale of three items in which respondents were asked whether they believed each kind of discrimination is or is not a reason why blacks live in less desirable housing than whites. Three kinds of discrimination were asked about: “because landlords will not rent apartments to blacks”; “because real estate agents will not show, sell, or rent to blacks”; and “because banks and lenders will not loan money to Blacks to purchase a home”. For each item, respondents were asked if this happened “almost never” “rarely” “sometimes” “often” or “very often”. The summary variable is the mean value of the sum of these three questions and so has a range from 0 to 5 (where 0 is assigned to respondents who deny that blacks live in less desirable housing) and a 5 would be assigned to respondents who believe that all three of these types of housing discrimination “very often” are the reason why blacks live in less desirable housing than whites.