No abstract

Suppose S is presented, either simultaneously or successively, with two multidimensional stimuli and is asked to judge whether the two stimuli are "same" or "different." Egeth (1966) has outlined a number of plausible models showing how S might perform this task. The discussion that follows is similar to his analysis.A stimulus dimension will be said to be either "same" or "different" depending on whether or not the two stimuli match each other along that dimension. The act of comparing two stimuli along a given dimension and, thus, deciding if that dimension is "same" or "differen t" may be termed processing that dimension.Since the stimuli are multidimensional, S must process a number of dimensions in order to decide whether the stimuli are "same" or "different." Would S process these dimensions one at a time, one after another? Or might they all be processed simultaneously? Models of S's performance that assume the former may be called serial (process) models, while those that assume the latter may be called parallel (process) models. In the case of parallel models, even though the processing of each dimension is begun simultaneously with the other dimensions, it need not be finished simultaneously with the other dimensions.When the two stimuli are "same," S must process all relevant dimensions in order to be sure that the stimuli are "same." What about the occasions when the two stimuli are "different"? As soon as S has finished processing any "different" dimension, he has enough information to decide that the stimuli are "different." Will he conclude at that point that the stimuli are "different," or will he wait until processing has been finished on all dimensions before making a decision? Models that assume the former may be called self-terminating (process) models, while those that assume the latter may be called exhaustive (process) models.This analysis has generated four types of models. There are serial and parallel models and within each of these types there are both self-terminating and exhaustive models.The reaction times (RT) of S's judgments provide a means of testing these models. Suppose that S is asked to make one of two responses depending on whether the two stimuli are "same" or "different." One question of interest is: How does the mean RT of the "same" response compare with the mean RT of the "different" response?On a certain subset of trials, the two stimuli differ along exactly D stimulus dimensions. Suppose that the D dimensions that are to be "different" are chosen randomly on each trial. Another question of interest is: How does the mean RT of the "different" response vary as a function of D, the number of "different" dimensions?The term processing isochronality will indicate that, for each stimulus dimension, the processing time for that dimension is independent of whether the dimension is "same" or "different." Similarly, the term efferent isochronality will indicate that the amount of time intervening between the decision to respond and the completion of the response is independent...

No abstract

This article demonstrates how multinomial processing tree (MPT) models can be used as assessment tools to measure the source of cognitive deficits in clinical populations. The application of this type of modeling is illustrated with a model developed by Batchelder and Riefer (1980, Psychological Review) that uses the free recall of category pairs to separately measure storage and retrieval processes in memory. A special version of the model is described that incorporates order constraints in the model's parameters over repeated trails. Computer simulations of the model can be used to address such issues as bias and standard error in the parameter estimates, and how these are affected by individual differences. The article discusses the utility of conducting validity tests of MPT models when used for clinical assessment, and this is illustrated for the pair-clustering model in two experiments. Experiment 1 shows that presentation rate during study affects the storage of clusters but not their retrievability, while Experiment 2 shows that part-list cuing during recall hurts the retrieval of clusters, but does not affect their storage. Experiment 3 and 4 apply the model to two clinical populations: schizophrenics and alcoholics with organic brain damage. The model's analysis reveals that each clinical group suffers deficits in both storage and retrieval compared to a control group. The results suggest that the retrieval deficits are stronger and occur more consistently over trails, whereas storage deficits become significant only on later trails. In addition, the organic alcoholics exhibit no improvement in retrieval over trails, although their storage improves over trails at the same rate as that for the control group.