| A basic principle of probability is the conjunction rule, prob(A) (GREATERTHEQ) prob(A & B). Past research shows that this rule is consistently violated for certain problems in which the representativeness or availability heuristic would predict the opposite judgment.;This dissertation proposes a conflicting-heuristic model to account for conjunction-rule violations and for the effects of training on such violations. According to the model, the rule is always violated when both (a) the set-inclusion relation is unrecognized and (b) a heuristic, such as representativeness, which conflicts with the set-inclusion relation is applicable. The rule is never violated when both (a) the set-inclusion relation is recognized and (b) the logical rule based on set inclusion is cognitively available. Under combinations of these conditions, violations occur with varying probabilities.;The conflicting-heuristic model predicts that training in the use of set-inclusion relations should decrease violations of the conjunction rule. In Experiment 1, subjects read a brief training document, then attempted twelve probability problems, followed by six similarity problems. As predicted, training reduced conjunction-rule violations, compared with a control group. Similarity and probability judgments correlated strongly for the control group but were partly dissociated by training.;Experiment 2 replicated the training effect found in Experiment 1 and also tested a linguistic-misunderstanding model that attributes conjunction-rule violations to the misunderstanding of "B" as "B and not A". Such misunderstandings may indeed occur to some extent: rewriting the "B" term as "B and may or may not be A" reduces the violations somewhat. The data show clearly, however, that this misunderstanding cannot be the main cause of the violations, and they largely support the conflicting-heuristic model.;Experiment 3 tested whether presentation of Venn diagram exercises in connection with each problem would enhance the training effect. The data indicate that such an enhancement is not large, if it exists at all. However, commission of conjunction errors is correlated with poor Venn diagram performance. |
