ABSTRACT
The interaction of two different types of information used in item recognition is examined in three experiments using a study-test procedure. On each trial, the subject studies 16 words presented singly. The test list consists of single words to which the subject has to respond old or new by pressing either of two buttons. Both old and new words in the test list may be tested once or twice. The data of main interest are reaction time and accuracy for the second test of new words as a function of lag between the first and second tests. At lag 0, reaction time is short and accuracy high; at lags 2, 3, and 4, reaction time is long and accuracy, low; at longer lags, reaction time decreases and accuracy improves. This non-monotonicity is inconsistent with unelaborated versions of several models of memory retrieval and forces the addition of a process that allows a response to a test word to be based on the subject's memory of the previous response to the word. A three-boundary random walk model is proposed to explain the data, and the relation of this model to other current models in the item recognition literature is considered.
