Test Difficulty, True-Score Variance, and Predictive Power of Implicit Association Tests: Effects of Target Exemplar Stimuli

Following the test difficulty account, we manipulated the valence of the target exemplars of attitude Implicit Association Tests (IATs) with target categories that are generally considered positive or negative, using target exemplars that are either typically valenced (i.e., matching the respective target category’s valence) or atypically valenced (i.e., deviating from the respective target category’s valence). In doing so, we aimed to shift the IAT test difficulty from extreme test difficulty (i.e., mean IAT scores deviating strongly from zero) in the case of typically valenced exemplars to less extreme test difficulty (i.e., mean IAT scores deviating less strongly from zero) in the case of atypically valenced exemplars to increase the true-score variance and the predictive power of the IATs. We conducted three, pre-registered experiments (total n = 342) to test our hypotheses. We developed attitude IATs with the target categories environmental protection/environmental degradation, using both typically and atypically valenced target exemplars for each category. All data was analyzed using structural equation models. As hypothesized, in all three experiments the atypically compared to the typically valenced exemplars led to significantly less extreme test difficulties. However, contrary to our hypotheses, the less extreme test difficulty for IATs with atypical target exemplars did not result in a significant increase in true-score variance or predictive power in any of the experiments.