Not So Averse After All: Behavioral Mechanisms Underlying the Use of Algorithms in Managerial Forecasting

Abstract

Although algorithms potentially outperform human judgment in managerial decision-making and forecasting, practitioners frequently refrain from using them. Extant literature provides mixed evidence on algorithm aversion and suggests very different underlying behavioral mechanisms. We contribute to existing research in two ways. First, we replace the traditional experimental instrument of measuring algorithm aversion, which asks participants to abandon their personal forecast in favor of a second, usually superior algorithmic forecast, with a new approach, in which participants receive human-made and algorithmic information simultaneously to form their judgments. Second, we derive competing empirical implications from potential mechanisms, allowing for an evaluation of their relevance. Using two factorial survey experiments with a forecasting task, we find very limited algorithm aversion. Regarding mechanisms underlying algorithm aversion discussed in existing research, we find that selective attention is most relevant: Users are more likely to notice superior performance of a human than of an algorithm.HIGHLIGHTS Algorithm aversion as measured in extant literature is largely due to psychological ownership effects concerning outputs produced by users. Algorithm’s superior performance is perceived less than superior performance of a human. Users act on their perceptions of performance in the case of humans, but not in the case of algorithms.