On the effectiveness of approach-avoidance instructions and training for changing evaluations of social groups

Prior evidence suggests that White participants who repeatedly approach images of Black people and avoid images of White people can exhibit a reduction in implicit racial bias (Kawakami, Phills, Steele, & Dovidio, 2007). In contrast, a recent study by Van Dessel, De Houwer, Gast, and Smith (2015) showed that mere instructions to perform approach-avoidance training in an upcoming phase produces a similar change in implicit evaluations of unfamiliar but not familiar social groups. We report 4 experiments that examined the replicability and generalizability of these findings for well-known social groups. Experiment 1 was a replication of the study by Kawakami et al. (2007) in a different domain (i.e., Flemish students' bias toward Turkish people) showing relatively weak evidence for small approach-avoidance training effects on implicit evaluations and explicit liking ratings. Experiment 2 replicated the finding of Van Dessel et al. (2015) that approach-avoidance instructions do not influence implicit evaluations of social out-groups and found no instruction effects even when participants first completed training with nonsocial stimuli. Experiment 3 established the presence of a small approach-avoidance training effect on implicit (but not explicit evaluations) in a large online sample. Experiment 4 directly compared approach-avoidance training and instruction effects, corroborating (a) the effect of training on implicit evaluations which was both small and subject to boundary conditions and (b) the absence of such an effect of instructions. There were again no effects on explicit evaluations. Whereas the current findings provide supportive evidence for training-based approach-avoidance effects (on Implicit Association Test [IAT] scores: meta-analytic effect size current experiments: d = 0.18, Bayes Factor = 65.22; current and prior experiments: d = 0.23, Bayes Factor = 4404.42) and evidence for the absence of instruction-based effects (Bayes Factors < 0.19), they also illustrate that there is still much uncertainty regarding the boundary conditions of these effects and the underlying mental processes. (PsycInfo Database Record (c) 2020 APA, all rights reserved).