Associative and rule-based accounts of contingency judgment.

Abstract

Two types of model may account for how people learn and make judgments about contingent relationships between pairs of events. Rule-based accounts hold that people act as intuitive statisticians, mentally tallying frequencies of event occurrences, non-occurrence, and co-occurrences, and integrating these frequencies at the time of judgment. Associative accounts hold that people form psychological associations between pairs of events during learning and they base their contingency judgments directly on these associations. The present research was conducted to determine which type of model provides a better explanation for a cue-interaction effect, when the judged degree of contingency between an outcome (e.g., a disease) and a cue (e.g., a symptom) is affected in systematic ways by the actual degree of contingency between the outcome and a second cue. Prototypically, subjects were presented with two problems in which they predicted (with feedback) whether or not each of 36 fictional patients had a particular disease, given information about the presence or absence of each of two potential causes of that disease. For both problems, the disease was moderately contingent on one cause (the target cause). In the strong-context problem, the outcome was highly contingent on the other cause (the context cause). In the weak-context condition, the disease was non-contingent on the context cause. The cue-interaction effect known as overshadowing was said to have occurred when subjects then judged the disease-target contingency to be stronger in the weak-context condition than in the strong-context condition. Across five experiments, overshadowing occurred and failed to occur in better accord with the predictions of the associative account than with those of the rule-based account. However, certain anomalous results--from both the present set of experiments and past studies--suggest that both associative and rule-based processes may be operative, but under different task conditions. Accordingly, a comprehensive theory of human contingency learning and judgment, which includes both types of processes, is outlined.