Confidence-accuracy inversions in scene recognition: a remember-know analysis.

S. E. Clark (1997) offered a modified signal-detection explanation of the confidence-accuracy inversions observed in E. Tulving's (1981) experiments. In addition to replicating E. Tulving (1981), we had participants make "remember-familiar" judgments. Confidence and accuracy dissociated across subjective reports. Response confidence differed only for judgments based on familiarity, whereas accuracy differed only for "remember" responses. S. E. Clark's model does not predict this, nor can it mimic "remember" performance across all conditions. We propose that although "knowing" can be accommodated within an equal variance signal-detection account, "remembering" is governed by contextual constraints that influence the distinctiveness of information upon which participants rely during reports. The current paradigm is a pictorial analogue to H. L. Roediger and K. B. McDermott's paradigm (1995) in that participants claim to explicitly remember thematically related items that were not actually seen during study.