In Experiments 1, 2, and 3 four recognition tasks are compared: single, cued, pair, and associative. Five recognition memory models are applied to representative data from Experiment 2: Gillund & Shiffrin's (1984) SAM model, Hintzman's (1984) MINERVA 2, Humphreys' (1978) Independent Cue Model (ICM), Murdock's (1982) TODAM, and Pike's (1984) Matrix model. MINERVA 2 and the ICM predict that cued recognition is superior to single item recognition. SAM predicts that cued recognition performance cannot surpass single item recognition performance. TODAM and the Matrix model can predict positive or negative cuing effects, depending on parameters. However, TODAM and the Matrix model predict that as associative recognition performance increases, the cuing advantage also increases. Experiments l, 2, and 3 showed small negative cuing effects that did not vary with presentation time, and levels of associative recognition performance that rose dramatically with presentation time. MINERVA 2 and the Matrix model clearly did not fit the data. The fit of the ICM was better than for MINERVA 2 or Matrix models. SAM and TODAM fit the data reasonably well, with TODAM fitting the best. Strengths and weaknesses of the models are discussed.
Experiments 4 and 5 showed negative cuing effects only for low frequency words presented for short study durations, and showed strong positive cuing effects for high frequency words at longer presentation rates. Positive cuing effects are not predicted by the SAM model. Modifications are discussed that will allow SAM to predict a cuing advantage, and allow other models to predict a cuing deficit even when associative recognition performance rises.