Recognition judgments under risk: Low confidence when certainty is low

Prior research has shown that misleading information about memory probes impairs accuracy in recognition memory tests. Such misleading information also impairs confidence in recognition responses, but for correct rejections only, not for hits. It is unknown whether such effects are preserved when participants face different outcomes according to performance. In the current study, after studying a series of words, participants performed a recognition memory task in which each memory probe could be preceded by a cue forecasting probabilistically the probe's study status (“Likely old” or “Likely new”). Seventy five percent of the cues forecasted accurately the study status of the probes (i.e., valid cues), and 25% inaccurately (i.e., invalid cues). In addition, participants gained or lost points according to whether they made correct or incorrect memory judgments, and received a comestible reward after accumulating 10 of such points. This accumulation scheme was administered in half the study-test blocks. The accuracy/confidence dissociation demonstrated in prior experiments is replicated here. Furthermore, we demonstrate that the decrease in confidence for novel items preceded by invalid cues is accentuated when response accuracy can yield gains or losses. We interpret this confidence decrease for invalidly-cued/rewarded correct rejections as reflecting a combination of loss aversion, as postulated by Prospect theory, and absence of recollection, as postulated by dual process models of recognition.