Intuitive inference strategies in interval learning tasks as a function of validity magnitude and sign

Abstract

Nine groups of ten subjects each performed for 200 trials on interval learning tasks in which the stimulus, response and reinforcement continua were continuous scaled variables. The nine experimental conditions were defined by the size and sign of the product-moment correlation between the stimulus and reinforcement (criterion) distributions. That is, γeranged from −.80 to +.80 in steps of .20. There was little or no evidence for matching behavior with correlational performance measures. Subjects learning positive γe relationships tended to overmatch (γs > γe), with the tendency to do so increasing as the size of γe decreased. Subjects who learned negative γe values consistently undermatched (γs < γe), apparently due to a general set to regard "relatedness" as a positive phenomenon. Analysis of subjects' conditional response arrays showed that subjects' conditional response distributions had greater dispersion than the corresponding conditional reinforcement distributions. The size of this effect increased as a function of the absolute magnitude of γe and was most evident with high and low stimulus arrays. A tendency for subjects to displace the array means followed a similar pattern.