Date of Award
Doctor of Philosophy (PhD)
In this dissertation, I present a quartet of experiments that studied confidence ratings and remember/know/guess judgments as indicators of recognition accuracy. The goal of these experiments was to examine the validity of these quantitative and qualitative measures of metacognitive monitoring and to interpret them using the continuous dual-process model of signal detection (Wixted & Mickes, 2010).
In Experiment 1, subjects heard or read items belonging to categorized lists and took an old/new recognition test over studied and new items while making remember/know/guess judgments after each recognition decision. Consistent with prior literature, remember judgments were more likely to be accurate than know judgments, and knows more accurate than guesses. Subjects were more likely to commit remember false alarms to nonstudied category members of higher response frequency for a category (e.g., eagle) than to items of lower response frequency (e.g., ostrich), although the overall proportion of false remembering was lower than the proportion often found using associative false memory procedures (e.g., Roediger & McDermott, 1995). Presentation modality did not affect recognition performance.
In Experiment 2, subjects provided both confidence ratings and remember/know/guess judgments following recognition decisions in an otherwise similar procedure. Overall, accuracy correlated with both confidence and remember/know/guess judgment, and remembered memories rated with high confidence were more accurate than either high confidence or remembered memories alone. These results suggested that confident retrieval of episodic and contextual information supported accurate recognition decisions. I also calculated confidence-accuracy correlations using four methods and found that confidence and accuracy were correlated for remembered and known memories, but that no correlation was found for guesses.
In Experiment 3, subjects studied category items in different screen positions (instead of in the center of the screen, as in the prior experiments). On the recognition test following, subjects were tested on whether items presented were old or new and also reported the screen position in which items were presented (i.e., a test of source memory). Confidence ratings followed these recognition + source decisions. A similar relationship was found between confidence ratings and remember/know/guess judgments when predicting both old/new recognition accuracy and source accuracy. This result contradicts predictions made by the continuous dual-process model, which states that only remember judgments and not confidence ratings should indicate source accuracy.
Experiment 4 was conducted to replicate and extend results of Experiment 3 and to examine the effects of the order of judgments provided during the test. In this experiment, subjects were asked to make old/new recognition decisions, old/new confidence ratings, source decisions, source confidence ratings, and remember/know/guess judgments, with test order counterbalanced among four between-subjects conditions. In this study, I found that the relationship between confidence and old/new and source accuracy as a function of remember/know/guess judgment was similar regardless of condition, reproducing the observations of Experiment 3. These results were also inconsistent with predictions made by the continuous dual-process model and suggested that the results of Experiment 3 were not due to confounding effects of judgment order.
Taken together, the results of these four experiments suggest that confidence and remember/know/guess judgments are valuable when used jointly and that both contribute individually as indicators of recognition accuracy. The results show that the continuous dual- process model of signal detection is a helpful way to consider the interaction of confidence ratings and remember/know/guess judgments, but they also imply that additional research is necessary to evaluate how the present results fit with the model. In particular, Experiments 3 and 4 failed to obtain Wixted and Mickes' (2010) finding of higher source accuracy for remember responses than for knows and guesses regardless of level of confidence.
The practical message is that researchers and rememberers should consider both quantitative and qualitative characteristics of a memory when attempting to infer its accuracy.
Chair and Committee
Henry L Roediger
Carl F Craver, Ian G Dobbins, Mark A McDaniel, Kathleen B McDermott
DeSoto, Kurt Andrew, "When Can We Trust Our Memories? Quantitative and Qualitative Indicators of Recognition Accuracy" (2015). Arts & Sciences Electronic Theses and Dissertations. 494.