I am pretty sure but not 100%: obtaining, interpreting and presenting eyewitness confidence statements

Abstract

Eyewitness identification confidence is typically studied using scales (generally numeric); in practice, eyewitnesses typically provide confidence in their own words. Verbal and numeric confidence similarly predict accuracy, but verbal confidence is difficult to interpret reliably (Mansour, 2020). To minimize miscommunication, eyewitnesses could provide scale ratings after verbal judgements or vice versa, but we do not know if the order in which such confidence statements are obtained affects the confidence-accuracy relationship. I tested the utility of requesting both verbal and numeric confidence and whether order effects exist. Participants (N = 198) viewed a mock-crime video with two perpetrators. After a delay, they viewed two simultaneous lineups with one perpetrator each and provided confidence for each perpetrator verbally (in their own words) and then numerically (0-100%) or numerically and then verbally. Numeric confidence in identifications was higher when provided first, t(393.82) = 2.40, p = .02, d = 0.24. Confidence-accuracy characteristic (CAC) curve analysis indicates the effect is driven by medium-confidence judgements (numeric range). No order effect was found for verbal confidence (p = .32). However, for low and high numeric confidence, verbal followed by numeric was better calibrated than numeric followed by verbal. When the numeric judgement came first, none of the subsequent verbal judgements could be categorized as high confidence using our coding scheme. These data provide preliminary evidence that eyewitnesses should provide only a single confidence judgement.