Models of contingency judgment

A contingency judgment entails an evaluation of the predictive relation between a cue and an outcome. Recent research on contingency judgment has sought to understand how people judge these relations. Several important findings of cue interaction indicate that the relationship between a cue and an outcome is not judged in isolation. Rather, the judged predictiveness of one cue is dependent on the predictive strength of other cues which are present at the same time. Any model of contingency judgment must be able to account for cue interaction. Three classes of models correctly predict cue interaction in contingency judgment: associative models, network models, and statistical models. Each of these models assigns predictive strengths to cues so as to minimize the discrepancy between the outcome which actually occurs and the outcome predicted on the basis of all cues present. These classes of models make some differential predictions, however. Associative models predict that the order in which trials are presented will affect the resulting strengths of predictive cues. Statistical models predict no order effect but rather predict that contingency judgments will be perfectly retrospective. The predictions derived from network models are more adaptable; the effects of trial order depend on the details of the particular network model under inspection. In the studies reported here, college students were presented with trial-by-trial data regarding several medical symptoms and a disease and were asked to judge the predictive relationship between each symptom and the disease. Experiment 1 illustrated a cue interaction phenomenon called blocking. Experiments 2 and 3 demonstrated that trial order affects the blocking phenomenon. Specifically, a forward blocking trial order resulted in more cue interaction than did a backward blocking trial order. Experiment 4 demonstrated that a backward blocking procedure does produce some cue interaction. Experiment 5 established that trial order affects another cue interaction phenomenon, conditioned inhibition, in a manner similar to the effect on blocking. These experiments indicate that trial order influences contingency judgments but that some retrospection does occur. Neither the associative nor the statistical models can fully account for this data pattern. A possible network account is discussed