Convergent Validity of Two Implicit Racial Bias Measures for Individuals and Contexts

Abstract

Correlations among implicit measures of racial bias tend to be weak, raising questions about their validity. Weak individual-level correlations, however, could reflect a temporally unstable construct. According to the Bias of Crowds model, implicit bias can be fleeting for individuals but, when aggregated, reveals which concepts are widely accessible for a population. Using this framework, we hypothesized that correlations between two tests—the Implicit Association Test (IAT) and Affect Misattribution Procedure (AMP)—would be weak across individuals but stronger across counties. We also predicted that county-level racial bias should relate to meaningful contextual factors, like structural inequality. In support of these hypotheses ( N = 2,137 people, 103 counties), the IAT and AMP were correlated more strongly at the level of counties ( r = .47) than individuals ( r = .24). Furthermore, county-level bias was significantly related to structural inequality. This study shows that implicit measures show substantially more convergent validity at the context than the person level.