Discrete-trial probability learning in rats: Effects of local contingencies of reinforcement

Abstract

Three groups of rats were trained using a noncorrection procedure on a probability learning task in which responses on one lever produced reinforcement on 61% of the trials and responses on the other lever produced reinforcement on the remaining trials. For Group RAN, there were no sequential constraints, but for Groups REP and ALT, the lever on which reinforcement was produced on a given trial depended upon the lever that had been designated correct on the previous trial. For Group REP, the locus of reinforcement was repeated on approximately 80% of trials; Group ALT was required to switch to the other lever on approximately 80% of trials. The maximizing strategy for Group RAN was exclusive responding to the majority lever, for Group ALT, it was “win-shift, lose-stay,” and for Group REP, “win-stay, lose-shift.” Subjects from both the REP and ALT groups tended to maximize momentarily, thereby showing a lower relative frequency of majority responses than did subjects from the RAN group. The differences between groups cannot be accounted for by a molar analysis; a molecular analysis is needed.