Erring on the side of caution: The influence of base rates, payoffs, and discriminability on face identification performance.

Unfamiliar face identification is challenging: An individual’s appearance can vary across images, and images of different individuals can look similar. Matching identity in unfamiliar faces is often thought of as a perceptual problem—driven primarily by differences in sensitivity (d′). This assumption ignores effects of base rates (proportion of match vs. mismatch trials), payoffs (relative cost of misses vs. false alarms), and discriminability (difficulty)—effects deemed significant by economic models. We examined whether participants optimized performance in the context of unequal base rates and payoffs, and varying levels of discriminability—and how these parameters influenced d′ and criterion (c). Across two studies, participants (Study 1: n=252; Study 2: n=336) completed two rounds of an identity-matching task. Round 1 (Studies 1 and 2) comprised an equal number of match and mismatch trials; accurate responses (hits and correct rejections) earned 5 points while misses (responding different on match trials) and false alarms (FAs, responding same on mismatch trials) cost 5 points. In Round 2, participants were assigned to a high base rate (Study 1: 80% match trials; Study 2: 80% mismatched trials) or a costly error condition (Study 1: -30 points for FAs, -2 points for misses; Study 2: -2 points for FAs, -30 points for misses). In Study 2, we also manipulated discriminability (50% of participants performed the task with own- and other-race faces). As predicted by the expected value function, the manipulations in Round 2 shifted criterion in the optimal direction (e.g., more conservative when FAs were costly or when mismatches were more common), with no effect on d′. Importantly, shifts in criterion were largest when discriminability was poor—both in terms of individual differences in d′ and discriminability of stimuli. These studies have implications for applied settings and theoretical models of face identification.