The Remember Response: Subject to Bias, Graded, and Not a Process-Pure Indicator of Recollection
Rotello, Caren M., Macmillan, Neil A., Reeder, John A., Wong, Mungchen, Psychonomic Bulletin & Review
Recognition memory judgments have long been assumed to depend on the contributions of two underlying processes: recollection and familiarity. We measured recollection with receiver-operating characteristic (ROC) data and remember-know judgments. Under standard remember-know instructions, the two estimates of recollection diverged. When subjects were told they might need to justify their remember responses to the experimenter, the two estimates were more likely to agree. The data support the conclusion that remember responses are generally based on a continuous underlying process but that specific task instructions can produce data that appear consistent with a high-threshold recollective process. Models based on signal detection theory provide a better account of these data than does the dual-process model (Yonelinas, 1994) or process-pure interpretations.
Recognition judgments are widely believed to be based on two underlying processes: recollection and familiarity. Here, we focus on two behavioral techniques for identifying these components: the remember-know paradigm and item recognition rating experiments.
In the remember-know paradigm (TuIving, 1985), subjects are asked to distinguish the subjective experience of remembering something specific about the occurrence of a study item from that of knowing that it was studied despite their failure to retrieve any specific details. A variety of empirical dissociations of remembering and knowing have been taken as evidence that the judgments are based on different underlying processes or memory systems (for a recent review, see Gardiner & Richardson-Klavehn, 2000). This process-pure interpretation underlies the majority of empirical remember-know studies.
An influential quantitative version of the process-pure hypothesis, the dual-process model (Yonelinas, 1994), proposes two converging operations for measuring the amount of recollection, which we denote RDP. (1) In the remember-know paradigm, remember judgments draw only on recollection, know judgments only on familiarity (Yonelinas, 2001). RDP should equal the proportion of old items that are remembered, but to allow for small amounts of nonmemorial noise, it is better estimated when the proportion of remembered new items is subtracted from that proportion (Yonelinas, 2002).' (2) In rating experiments, receiver-operating characteristic (ROC) curves predicted by the model have a y-intercept that equals R^sub DP^. Thus, ROC curves and remember judgments should produce essentially identical estimates of the underlying recollection process (Yonelinas, 2001, 2002).
Two studies from Yonelinas's laboratory that allowed estimation of R^sub DP^ from both remember judgments and ROC data simultaneously showed this equivalence (Yonelinas, 2001, Experiment 1; Yonelinas, Dobbins, Szymanski, Dhaliwal, & King, 1996, Experiment 2).2 The subjects in these studies followed almost all highest confidence (and few lower confidence) old judgments with remember responses; know responses occurred at a variety of confidence levels. This pattern of responding leads directly to the similar estimates of R^sub DP^.
In the only other experiments for which both ROCs and remember-know judgments were reported for the same subjects, a different pattern was found. Rotello, Macmillan, and Reeder (2004, Experiments IA and IB) observed remember judgments to words recognized with a variety of confidence levels. When the dual-process model is fit to these data, the estimates of R^sub DP^ from ROCs and from remember judgments diverge: In Experiment IA, the ROC analysis concluded that R^sub DP^ equals . 18, whereas the data from the remember judgments imply that R^sub DP^ is .26 (estimated from highest confidence responses) or .29 (overall); in Experiment IB, the ROC implies that RDP equals .09 and the data from the remember judgments indicate that R^sub DP^ is .23 (highest confidence) or .25 (overall). According to these data, if the ROC y-intercept measures recollection, the remember judgments do not. …