Abstract
Confidence reports are characterised as subjective assessments of performance given the quality of the perceptual evidence in the current trial. However, sequential effects in confidence recently reported by Rahnev and colleagues (2015) show that confidence is modulated by trial history. Specifically, the level of confidence in one trial is predictive of that in the subsequent (confidence leak). The proposed explanation involves subjects updating their expectation of the next trial's difficulty according to the current confidence level. Unlike Rahnev et al., we examined sequential effects in confidence with fixed difficulty, but also variable prior and reward structure. On each trial, participants reported the orientation (left/right) of a tilted Gabor (fixed to d'=1), followed by a confidence report (high/low). Feedback was provided for the orientation response. Different reward-prior combinations were tested in 7 separate sessions. Confidence in the current trial was predominantly dependent on the correctness of the orientation judgment in the current trial (confidence reflects performance) and the confidence reported in the previous trial (confidence leak). The correctness of the orientation judgment in the previous trial and the reward gained in the previous trial had weak and inconsistent effects on confidence. Subjects with larger confidence leaks, as estimated by a 1-back regression, show strong and idiosyncratic structure in the autocorrelation of confidence reports over long time scales. Confidence leak has a substantial effect on confidence assessments. When confidence leak was large, confidence judgments correlated with trials much earlier in the session. Thus, one must proceed with caution when interpreting or modelling confidence ratings based on single-trial evidence. To attenuate these effects in data, it would be preferable to measure confidence using a relative judgment ("Are you more confident in the current decision than you were in the previous trial?").
Meeting abstract presented at VSS 2018