THE VALIDITY OF FEELING OF KNOWING IN REC ALL AND RECOGNITION

The traditional experimental procedure used to investigate feeling of knowing (FOK) consists of three phases. In the first phase, subjects' knowledge is tested, usually with a recall test. In the second phase, subjects are asked to predict subsequent memory performance on items t hat failed to be recalled. Finally, in the third phase, the accuracy o f the FOK predictions is evaluated against a criterion test. The crite rion test usually consists of a forced-choice recognition test. Gammas are computed to measure degree of agreement between each individual s ubject's FOK and performance on the criterion test. Many studies based on this paradigm show that FOR can predict performance on the criteri on test. However the correlations obtained in these studies are rather weak: Typical gammas are around .30. One goal of the present study wa s to determine whether similar results obtain when all items are consi dered, instead of only those having previously led to recall failure. Another goal of the present study was to compare the predictive accura cy of individual FOK with that of average FOK and average item difficu lty in different memory tasks. To achieve these goals, modifications w ere made to the traditional experimental paradigm. Subjects had to est imate their FOK prior to the first recall test. They were not informed of the accuracy of their responses, so that all items could also be t ested in the following criterion test. In one experiment, the criterio n test consisted of a forced-choice recognition task; in the other exp eriment, it consisted of a cued-recall test. The same material, consis ting of a set of over 100 definitions of rare words, was used in the r ecall, cued-recall, and recognition tasks. Results show individual FOK to be a poor predictor of performance in the forced-choice recognitio n task, the average gamma (-.03) not being significantly different fro m 0 when computed on previously failed items. Gammas increased and rea ched an average of .52 when the validity of individual FOK was compute d over all items. However, the validity of the other predictors also i ncreased, so that average item difficulty remained the best predictor of performance in the recognition task, with an average gamma of .72. Finally, there was no difference in predictive accuracy between indivi dual and average FOK in recognition task. By contrast, individual FOK was always the best predictor of performance in the recall tasks. Gamm as computed over all items averaged .80 in the recall task and .92 whe n the first letter of the target was given as a cue. In comparison, ga mmas for average item difficulty averaged .68 and .74 in the recall an d cued-recall tasks, respectively. Average FOK was the worst predictor of performance in the recall (.61) and cued-recall tasks (.70). The w ide range of gammas obtained calls for caution in drawing conclusions about the validity of feeling of knowing. Individual FOK did not predi ct subjects ability to discriminate targets from distracters in the fo rced-choice task, when the analyses were restricted to unrecalled item s. Similar analyses in previous studies have yielded more reliable gam mas. However, subjects in previous studies were implicitly or explicit ly informed of their recall errors and could therefore adjust their FO K estimates and/or performance in the criterion test. Whatever may be the case, item difficulty was the best predictor of performance in the forced-choice task in the present study as well as in previous ones. Such results hardly support a theory of FOK as monitoring target avail ability. It is argued instead that FOK monitors accessibility to possi ble answers which satisfy to varying degrees the constraints of the de finition presented. FOK cannot determine which of many candidate answe rs is the target because the memory contents, which FOK monitors, may or may not match the dictionary contents used to define the targets. F OK was found to be very accurate in predicting recall performance when a cue was provided concerning the expected target.