Do Racial and Ethnic Stereotypes Depend on The Sex of Target Group Members? Evidence from a Survey-Based Experiment

Bibliographic Data

ID2277249
AuthorsJ M Timberlake (0000-0003-3143-2810, University of Cincinnati, corresponding author), Sarah Beth Este, Sarah Beth Estes (University of Arkansas at Little Rock)
Year2007
Volume48
Issue3
Pages399-433
Publication date2007-08-01
Peer ReviewedYes
Open AccessYes
TypeARTICLE
VenueSociological Quarterly (JOURNAL)
Journal identifiersISSN: 0038-0253 • E-ISSN: 1533-8525
PublisherInforma UK Limited (PUBLISHER • GB)
DOI10.1111/j.1533-8525.2007.00083.x
OpenAlexW2062103072
LanguageEN
Citations received26
References cited49

Several social scientific perspectives suggest that racial and ethnic stereotypes vary by the sex of minority group members. However, prior research has not used survey data to test the hypothesis that the public holds strong gendered racial/ethnic stereotypes. We exploit a between-subjects experiment in the Multi-City Study of Urban Inequality to investigate the extent to which respondents' beliefs about seven attributes of four racial/ethnic target groups depend on the sex of target group members. Consistent with the "contingent" perspective of intersectionality theory, we find that effects of target group sex depend upon the extent to which particular racial/ethnic stereotypes have been gendered in American culture. However, we also find that effects of target group sex are consistently much smaller than those of race/ethnicity. We conclude by discussing implications for survey research on stereotypes across multiple status dimensions. ACKNOWLEDGMENTSWe thank the University of Chicago's Social Sciences Division Computing Services for providing MCSUI data and codebooks, Irene Browne for information on the MCSUI, Jennifer Malat, Ken Rasinski, and two anonymous reviewers for their insightful comments, Neal Ritchey for methodological advice, and Joseph Michael and Alana Yoder for their research assistance.NOTESNotes1 As with "Asian" and "Hispanic," we capitalize "White" and "Black" to emphasize the shared geographic and cultural heritage of people of African and European descent. We also use the terms "racial/ethnic" as an adjective and "race/ethnicity" as a noun to capture broadly the reality that individuals sort themselves (CitationHarris and Sim 2002) and are sorted (CitationHarris 2002) into different social groups based at least in part on geographic or cultural heritage.2 There is a relatively long-standing debate about the correct meaning and use of the terms "sex" and "gender" (CitationLaner 2000). In this article we use the term "sex" to denote whether individuals are male or female and "gender" when referring to stereotypes. That is, an attribute that respondents believe is characteristic of men or women would be a "gendered" stereotype.3 CitationWinship and Radbill (1994:241) caution against using sampling weights in ordinary least squares (OLS) regression models because weighting generates heteroskedasticity. They suggest that in cases where models cannot be specified so as to obviate the need for weights, researchers should use standard error estimators that account for heteroskedasticity. The Stata 8.0 procedure for estimating regressions with complex survey data (svyreg) makes "minimal assumptions about the nature of the sample.... Hence, the variance estimators ... produce variance estimates that generally will be either approximately unbiased or biased toward more conservative estimates, i.e., larger standard errors" (CitationStataCorp 2003:346–47). We fit the models with several types of estimators, and we can confirm that the standard errors estimated with the svyreg procedure are neither substantially nor downwardly biased.4 Reviewers had several concerns regarding these items. First, a reviewer worried that there might be social desirability effects; that is, the tendency for respondents to respond consistently to questions about similar groups so as not to appear racially prejudiced. In the MCSUI, social desirability concerns were muted in two ways. First, as explained above, most respondents were interviewed by a same-race interviewer. Second, the experimental design ensured that respondents received questions about only one category of sex. While there could still be social desirability with respect to race/ethnicity, this would not be a problem for the primary focus of our analysis; that is, whether or not there were significant differences in ratings across categories of sex. A second concern was that the stereotype items do not capture the extent to which context can affect the activation of stereotypes. It is certainly true that social psychological research has found large effects of various sorts of priming mechanisms on response patterns. We argue that both survey- and laboratory-based research feature strengths and weaknesses. Clearly, the great strength of survey research is its ability to generate inferences to a population; however, it is limited in the way that laboratory research is not in its ability to manipulate the context within which respondents interpret questions.5 For example, suppose Whites' average ratings on some stereotype item were 5 for White men, 4 for White women, 4 for Black men, and 3 for Black women. Using the differencing procedure, Whites' average ratings would be −1 (4 − 5) for Black men (i.e., relative to White men) and −1 (3 − 4) for Black women (i.e., relative to White women). Subtracting these two scores would yield no difference between the ratings of men and women (−1 − [−1] = 0); however, in reality, Whites would have rated women one point lower than men for each group.6 Although after applying sampling and nonresponse weights, it turns out that Asians are more evenly divided between the three cities and that Hispanics are overwhelmingly from Los Angeles (see Table 2).7 All MCSUI items were originally developed to elicit racial/ethnic stereotypes; hence, we do not expect any of the items to fall in the two left-hand quadrants of Figure 1.8 Consider the following regression model: where the βnk are effects of conventionally coded (i.e., 0, 1) dummy variables (the Xnk) and the γc are effects of control covariates (the Zc) coded in deviation units. As a result of this coding scheme, the intercept β0 is interpretable as the covariate-adjusted (i.e., controlling for the Zc) mean of Y for the omitted category of the Xn. The βnk are interpreted as covariate-adjusted increments or decrements to that mean for the included categories of the Xn

Ethnic group · Intersectionality · Race (biology) · Social group · Social identity theory · Sociology · White (mutation) · Gender Studies · Psychology · Social and Cultural Dynamics · Social and Intergroup Psychology · Social Psychology · Urban, Neighborhood, and Segregation Studies

Tools

Open DOISci-Hub