Second responses in paired-associate learning

Mathematical models of memory can be evaluated using a continuous paired-associate learning paradigm in which a second response is requested immediately following an error. This procedure provides discriminability among the models on the basis of the relationship between the first- and second-response error probabilities. Theoretical predictions are derived for a variety of models, including finite-state and noncontingent operator models, formulations based on the theory of signal detectability, the multicomponent memory model of Bower (1967b), and the long- and short-term model of Atkinson and Shiffrin (1968). None of these is unequivocally acceptable, and many are decisively rejected by the data of a single-trial experiment. The most satisfactory models are the multicomponent model and the Atkinson and Shiffrin model, both of which assume a greater degree of internal structure than the rejected models. It appears that internal structure of this sort will be a necessary part of any satisfactory model of second responses.