Whites as the target group needed to assess racial bias among white liberals
I tweeted about the possibility that whites are the proper target group for assessing racial bias among white liberals, so I thought I'd check available data to assess whether there is evidence for this. The recent Iyengar and Westwood 2015 AJPS article measuring different types of discrimination seemed a good place to look.
In the Iyengar and Westwood 2015 racial discrimination experiment, respondents were given a choice between two high school seniors competing for a scholarship, with names and clubs intended to signal race:
- Arthur Wolfe, President of the Future Investment Banker Club
- Jamal Washington, President of the African American Student Association
For some respondents, the two applicants had the same GPA (3.5 or 4.0), and, for other respondents, one of the applicants had a 3.5 GPA and the other had a 4.0 GPA.
Here are the results for white liberals and white conservatives:
EQUALLY QUALIFIED
Liberals: 73% selected the black target [n=34] CI: [58, 89]
Conservatives: 40% selected the black target [n=55] CI: [27, 53]
Difference between the 73% and the 40%: two-tailed p=0.001BLACK TARGET MORE QUALIFIED
Liberals: 92% selected the black target [n=12] CI: [73, 110]
Conservatives: 56% selected the black target [n=23] CI: [35, 78]
Difference between the 92% and the 56%: two-tailed p=0.013WHITE TARGET MORE QUALIFIED
Liberals: 44% selected the black target [n=18] CI: [19, 70]
Conservatives: 16% selected the black target [n=19] CI: [-2, 34]
Difference between the 44% and the 16%: two-tailed p=0.061
There were substantial differences in the estimates, with white liberals on average favoring the target with the black name when the targets were equally qualified.
Here are the results for white Democrats and white Republicans:
EQUALLY QUALIFIED
Democrats: 62% selected the black target [n=53] CI: [49, 76]
Republicans: 47% selected the black target [n=45] CI: [32, 62]
Difference between the 62% and the 47%: two-tailed p=0.125BLACK TARGET MORE QUALIFIED
Democrats: 75% selected the black target [n=32] CI: [59, 91]
Republicans: 50% selected the black target [n=16] CI: [22, 78]
Difference between the 75% and the 50%: two-tailed p=0.104WHITE TARGET MORE QUALIFIED
Democrats: 59% selected the black target [n=18] CI: [37, 81]
Republicans: 21% selected the black target [n=19] CI: [1, 41]
Difference between the 59% and the 21%: two-tailed p=0.012
Results indicated that white Democrats on average favored the target with the black name in all three scenarios, even when the white target was more qualified. The point estimate for white Republicans never crossed 50% in any scenario.
Very interesting. However, I would argue that those club associations signal a lot more than just race, possibly skewing the results.
For instance, conservatives may be more friendly towards investment bankers than liberals. And investment banking skills differ a lot from managing a social club.
What do you think?
Hi Joel,
I think that is a realistic alternate explanation for the results: respondents could be reacting to the target's club instead of (or in addition to) the target's race. The bias is in the expected direction, so I think that these results are ambiguous.
I did not see a way to address the conflation of club and race in the data for the Iyengar and Westwood 2015 AJPS experiment, but this morning I checked the data from a set of different experiments that manipulated race, described here. Race was manipulated in several ways across these experiments, such as manipulating the target's name or photograph.
There were several choices to make in the analysis of these experiments, such as whether to limit the sample to persons who passed attention checks and whether to include persons who only weakly identify with a particular ideology or political party. I did not conduct a sensitivity analysis, so I'll only describe the basic results from the analysis that I conducted, with the caveat that these could shift with different research design choices.
Data on respondent ideology and partisanship were available for most of the experiments, with exceptions of Oliver and Lee 2004 and Cottrell and Neuberg 2004 (which did not have data on ideology or partisanship) and Van Boven et al. 2006 (which had data on only partisanship).
Meta-analyses of the experiments with available data on respondent ideology and partisanship indicated:
* among white liberals, there was a pooled statistically significant discrimination in favor of the black target, with a small effect size estimate about 0.1.
* among white Democrats, there was a pooled discrimination in favor of the black target with the lower end of the 95% confidence interval at zero and a small effect size estimate about 0.1.
* there was no statistically significant pooled discrimination among white conservatives or white Republicans, with respective point estimates at -0.02 and 0.03.
Interesting. So in this meta-analysis, we find that liberals and Democrats are more likely to discriminate, however in favor of black applicants. I would not have guessed that.