A Comparative Investigation of Seven Implicit Measures of Social Cognition

Abstract

The present research compared the psychometric qualities of seven indirect attitude measures across three attitude domains (race, politics and self-esteem) with a large sample (n = 23,413). We compared the measures on internal consistency, sensitivity to known effects, relationship with indirect and direct measures of the same topic, reliability and validity of single-category attitude measurement, ability to detect meaningful variance among people with non-extreme attitudes, and robustness to the exclusion of misbehaved or well-behaved participants. All seven indirect measures correlated with each of the other indirect measures, and with direct measures of the same topic. For all seven measures, these relations were weak for self-esteem, moderate for race and strong for politics. This suggests that some of the source of variation in reliability and predictive validity of indirect measures is a function of the concepts rather than the methods. The Implicit Association Test (IAT) and Brief IAT (BIAT) showed the best overall psychometric quality, followed by the Go-No go Association Task, Single-target IAT (ST-IAT), Affective Misattribution Procedure (AMP), Sorting Paired Features task, and Evaluative Priming. The AMP showed a steep decline in its psychometric qualities when people with extreme attitude scores were removed. Single-category attitude scores computed for the IAT and BIAT showed good relationship with other attitude measures, but no evidence of discriminant validity between paired categories. The other measures, especially the AMP and ST-IAT, showed better evidence for discriminant validity. In a follow-up study, we improved some of the AMP’s psychometric qualities by increasing the number of trials.