Interaction Context is Key: A Meta-Analysis of Experimental Evidence on Interventions against Algorithm Aversion

Abstract

Algorithm aversion is a barrier to the adoption of advanced technologies, and individuals prefer human judgement over superior algorithmic decisions in certain contexts. Previous literature has looked at various interventions against algorithm aversion, but all studies have been domain specific. Therefore, this study investigates whether experimentally tested interventions are effective across various domains. We conducted a meta-analysis of 32 experimental studies with 89 effect sizes, demonstrating that these aggregated interventions significantly reduce algorithm aversion (overall effect size=0.23). In line with current research, we split the analysis into human, algorithm and context-specific subsamples and find that modifying the interaction environment shows the highest effectiveness (g=0.55) in overcoming algorithm aversion. Future research should test the intervention approaches identified here as most promising, such as providing information about how many other people found the algorithm useful, or simply framing the task in a more objective way to reduce bias against algorithms.