Figure 5: Targeting false feedback to manipulate the magnitude of PEs: logic and results.

In a between-groups behavioural experiment, participants completed a recognition memory test in which they made an old/new decision and provided a continuous confidence rating. Veridical feedback was provided on the majority of trials. Probabilistic false positive feedback was targeted to specific response types (new or old errors) and levels of confidence (high or low confidence) across four groups. The expected response outcome (ERO) and memory strength (MS) alternatives make distinct predictions about the relative magnitude of PEs elicited by positive feedback targeted to different confidence levels for each response type (cf. Fig. 1). Thus, the two alternatives make distinct predictions about the relative magnitude of criterion shifts for the four groups, shown in a and b. Both alternatives predict that the two groups that received false feedback on new responses will show a conservative criterion (blue arrows) and that the two groups that received false feedback on old responses will show a liberal criterion (orange arrows). Any difference in criterion between the Low Confidence and High Confidence groups provides evidence consistent with a role for PEs in regulating recognition decisions. The specific pattern of results across the groups, specifically between the New–Low Confidence and New–High Confidence groups, differentiates between the two alternatives. (c) The empirical terminal criterion values from the four false-feedback groups. A control group that received completely veridical feedback adopted a neutral criterion, shown at the far right. Liberal criterion values correspond to negative values on the vertical axis. Error bars are s.e.m.