The Effect of Forecast Inconsistency and Explicit Uncertainty Estimates on Trust and Decision-Making

Abstract

The goal of this dissertation was to evaluate whether inconsistency is a substantial threat to trust in sequential predictions from a single advisor, and to test whether providing explicit uncertainty estimates can preserve trust in the context of inconsistency. Four experiments were conducted in which participants decided, according to a sequence of snow accumulation forecasts, whether or not they should close schools due to an impending snowstorm. Since prioritizing consistency in weather forecasts can be at a tradeoff to forecast accuracy, Experiments 1-3 used highly controlled forecast stimuli to directly compare the relative effects of inaccuracy and inconsistency on trust. In Experiment 4, findings from the highly controlled experiments were retested and extended with realistic forecast stimuli varying naturally in terms of inaccuracy and inconsistency. In addition, half of participants were randomly assigned to receive reliable uncertainty estimates for a threshold level of snow with the deterministic forecast. Across all four experiments inconsistency was less detrimental to trust than inaccuracy and appeared to confer decision-making benefits. Access to reliable uncertainty estimates amplified these benefits.