Re-assessing the incremental predictive validity of Implicit Association Tests

Abstract

Indirect measures of attitudes or stereotypes, such as the Implicit Association Test (IAT), assess associations that are relatively automatic, unintentional, or uncontrollable. A primary argument for the IAT's use is that it can predict relevant outcomes beyond parallel direct measures, such as self-report (a claim referred to as demonstrating incremental predictive validity). Prior work on this issue relied primarily on least squares linear regression analyses, which are unable to correct for measurement (un)reliability and may then seriously inflate false positive rates in claims of incremental predictive validity. Properly accounting for the impact of measurement reliability requires using Structural Equation Modeling (SEM). In a pre-registered analysis, we investigated 10 IATs and 250 outcomes variables (N > 14,000), and found that 69.6% of outcomes were reliably correlated with the IAT. Among outcomes that were associated with both the IAT and self-report, the IAT showed incremental predictive validity in 58.6% of cases using least squares linear regression analysis and 59.2% of cases when using SEM, with the two analytic approaches reaching the same conclusion 91.4% of the time. Though the two analysis strategies largely converged, discrepancies were large enough to suggest a non-trivial percentage of conclusions drawn from least squares linear regression will be erroneous. As only SEM properly accounts for measurement reliability, it should be adopted in future analyses. To facilitate that goal, we provide tools for researchers to complete SEM analyses on tests concerning the incremental predictive validity of the IAT.