Purchase this article with an account.
Dawn Weatherford, Barret Schein; Mismatch prevalence influences response bias and discriminability in unfamiliar face matching. Journal of Vision 2015;15(12):697. doi: 10.1167/15.12.697.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Determining if an individual matches their identification card holds important consequences on tasks such as granting age-restricted access to goods and services as well as airport security. Unfamiliar identity matching represents a complex visual search task, with diagnostic facial cues indicating matches and mismatches. Recent evidence supports that, like other visual search tasks, low prevalence (e.g., 10%) of mismatched identities between face pairs increases error rates for mismatch trials. Binary match/mismatch decisions have indicated that criterion shifts, and not changes in discriminability or motor execution, underlie the effect. However, confidence-based responding has yet to be explored. Across 140 trials, participants either viewed a high (90%), medium (50%), or low (10%) prevalence of mismatches. Using a 1-6 confidence scale, participants rated a face in the context of an identification card as either a match or mismatch to an isolated face that was either the same or different person. To encourage accuracy, incorrect decisions were followed by corrective feedback. Within-participants analyses replicated previous findings demonstrating sensitivity to contextual prevalence cues as the trials progressed, such that criterion shifts, but not discrimination, differed among prevalence conditions. However, receiver operating characteristic (ROC) curves showed some evidence for greater discriminability in high and medium prevalence conditions as compared to the low prevalence condition, but only for high confidence responses. This is new finding deserves further exploration. Looking at ROC curves can provide important insight into discriminability that may be masked by binary response restrictions. Additionally, increased error rates due to low mismatch prevalence in real-world situations are incredibly important, as misidentifying individuals may have serious ramifications. Because real world mismatch prevalence is likely much lower than the present conditions (i.e., < 10%), increasing discriminability as a function of perceptual training may serve a useful role in maximizing identity matching performance.
Meeting abstract presented at VSS 2015
This PDF is available to Subscribers Only