Measuring implicit attitudes and stereotypes: Increasing internal consistency reveals the convergent validity of IAT and priming measures

Implicit measures in social psychology were developed to assess automatic associations between social categories and attributes. Tests designed to measure the same implicit associations should correlate, but relations among these measures can be attenuated by sub-optimal measurement reliability. Therefore, techniques that adjust for measurement error are needed to test the convergent validity of these measures. Five experiments examined the internal consistency (reliability), convergent validity, and effect sizes of three latency-based implicit measures: the Implicit Association Test (IAT) and two types of priming methods, category priming (CP) and its variant response-window priming (RW-CP); these tasks were designed to measure age attitudes in Experiments 1 (a and b) and 2, and gender stereotypes in Experiments 3 and 4. In Experiments 1a and 1b, raw correlations among IAT, CP and RW-CP measures of age attitudes were small (rs ≤ .21), but the priming measures also had low internal consistency (αs < .3). Experiments 24 examined modifications to priming procedures that increased the salience of social categories (i.e., young and old age categories in Experiment 2 and male and female gender categories in Experiments 3 and 4). These modifications—which included treating exemplars of social categories (i.e., age, gender) as categorized targets (rather than primes), use of special instructions drawing attention to category primes, and manipulation of time intervals between primes and targets to increase visibility of category primes—were found in some cases to increase priming effect sizes, internal consistency, sensitivity to known-group differences, and convergence with related measures. Across experiments, IAT effects were larger, had greater internal consistency (αs ≥ .7), and were more sensitive to known-group effects than the two priming measures. Strong correlations among the three measures emerged (confirming their convergent validity), especially when categories were salient and confirmatory factor analysis was used to correct for measurement error. Experiment 4 also demonstrated the discriminant validity of the implicit measures by showing that they tapped constructs that were distinct from explicit measures. Results supported the construct validity of IAT and priming measures, but also indicated that researchers must take steps to improve the reliability of priming measures to increase their utility.