The role of shifting in choice behavior of pigeons on a two-armed bandit.

Pigeons (Columba livia) were confronted with a problem in choice known as the two-armed bandit (TAB): two concurrent discrete trials random-ratio schedules in which the good side changes randomly from day to day. In the first experiment a substantial proportion of the birds' choices were contrary to those predicted by immediate maximization. Under a variety of parameter settings the pigeons chose the bad side more often than expected by immediate maximization. The hypothesis was advanced that shifting, that is, the tendency to avoid the most recently visited location, was responsible for limiting the number of choices on the good side of the TAB. Experiments 2-4 examined this hypothesis. The first of these experiments compared the accuracy on win-shift vs win-stay trials, and lose-shift vs lose-stay. When the response keys were at opposite sides of the operant chamber (keys-far), as in the first TAB experiment, there was a significant tendency to shift. This tendency was not present when the keys were close together (keys-close), in Experiment 3. Experiment 4 compared TAB performance in the original keys-far situation with that in the keys-close chamber. The proportion of choices on the good side of the TAB increased faster and attained a higher level for the keys-close group, and the proportion of choices which were shifts from one side to the other was smaller. A higher proportion of individual choices in the keys-close condition could be accounted for by immediate maximization.