ITEM EFFECTS IN RECOGNITION AND FRAGMENT COMPLETION - CONTINGENCY RELATIONS VARY FOR DIFFERENT SUBSETS OF WORDS

Item effects were examined in recognition-memory and word-fragment completion tasks, using the 192 words and fragments from Tulving, Schacter, and Stark (1982). Independent variables included various subjective ratings taken on the words and various quantitative characteristics of the word fragments. Experiment 1 provided "unit analysis" data on the degree to which these variables correlate with recognition and fragment completion and allowed the selection of lists that were predicted to yield different contingency relations between the two tasks, on the assumption that the contingency is at least partly a function of the set of words used. Experiment 2 confirmed the predicted ordering of relations over lists and provided further evidence that contingency relations are different for different subsets of words. Implications for the use of contingency analyses in successive-testing experiments are discussed. It is argued that measures of association can be severely limited by suppressor variables and can very widely depending on the sample of items. Consequently, little of theoretical importance can be learned by testing the stochastic independence of tasks.