Stimulus Salience and Contingency Judgement: an Application of the Rescorla-Wagner Model

Abstract

The purpose of the present study was to evaluate the Rescorla-Wagner (1972) model interpretation of human contingency judgement. A program of research was carried out that examined the role of one of the learning rate-parameters critical to the model on the acquisition and extinction of human contingency judgement. The present dissertation presents a selective review of the human contingency judgement literature. Initially, the methodology used in the experimental component of the study is described and compared and contrasted with previous research. Following this is a discussion of the traditional, cognitive interpretation of contingency judgement performance and it is argued that none of the models derived from this perspective are able to provide a thorough explanation of the mechanism humans use to perform judgements of contingency. The associative interpretation of human contingency judgement is described. The foundations of the Rescorla-Wagner (1972) model are discussed and the model is formally specified along with a brief summary of evidence supporting the model's usefulness in predicting Pavlovian conditioning. The interpretation of contingency judgement that the model provides is then discussed along with supporting empirical evidence. This section ends with a discussion of the learning rate parameters of the Rescorla-Wagner model and how these have been applied in classical conditioning and how they were applied to the contingency judgement task used in the present study. A sound methodology was developed to permit explicit manipulation of variables expected to influence the critical parameters of the Rescorla-Wagner model within the context of an on-line, 2 x 2, human judgement task. Chapters 2 to 5 describe the empirical component of the thesis. In Experiment 1 the nature of contingency judgement acquisition within a 2 x 2, on-line task was investigated to establish that the present experimental methodology was both sound and sensitive enough to capture learning and replicate the contingency judgement research to date. The pattern of acquisition observed in this experiment was consistent with that predicted by the Rescorla-Wagner model although the influence of a manipulation predicted to effect learning rate was unclear. In Experiment 2 the perceptual salience of the cue was manipulated through three levels in order to examine the role of the learning rate parameters in a contingency judgement task. A corresponding effect on contingency judgement acquisition similar to that predicted by the Rescorla-Wagner model was observed. In Experiment 3, the effect of cue salience on acquisition was again examined however this time the task was more sensitive (taking subjects’ judgements after every trial). This allowed the influence of each cell type to be measured. The final experiment was similar to Experiment 3 but also included an extinction phase in which a positive contingency was immediately followed by a null contingency. Across the final two experiments a method of fitting the obtained data to that predicted by the original (Rescorla & Wagner, 1972), Revised (Van Hamme & Wasserman, 1994) Rescorla-Wagner models, and to a proposed Integrated model was implemented. The result of this analysis suggested that the Integrated model provided the most accurate model of human contingency judgement performance in a 2 x 2 task.