Subjects were tested with standard verbal tests of recognition memory. They were given lists of common words to study and after a short interval (three minutes filled with chit-chat) were then given test lists composed of both new and old words. For each word on the recognition test, the subjects were asked to rate their confidence that the word was old or new.
Results were analyzed with the receiver operating characteristic (ROC) curve, a statistical tool widely used to evaluate the accuracy of a classifier. In this case, the ROC curves are confidence plots where the points are derived from rates of true-positive versus false-positive answers (saying "yes" correctly to a word that was on the previous list v. saying "yes" incorrectly to a word that wasn't).
Some theories of recognition, said Wixted, an expert on signal-detection models of memory, hold that the shape of the ROC is significant: A symmetric curve indicates that only the process of familiarity is involved, while an asymmetric curve indicates that both familiarity and recollection are at work. Also, the greater the asymmetry, the more significant the role of recollection.
After controlling for memory strength by testing the brain-damaged patients, who have diminished abilities, with shorter (and therefore easier) word lists than those given to controls the recognition accuracy of both groups was almost identical. Critically, once accuracy was equated, the asymmetry of the ROC curves was identical as well.
This is the first ROC study, Wixted noted, to age-match the subjects and to "strengthen" the memories of the hippocampal patients so that their ROCs could be meaningfully compared to that of the controls. The similarity of the ROCs suggests that recollection is operative even in patients with ext
Contact: Inga Kiderra
University of California - San Diego