Hypotheses about reinforcement in human learning

Abstract

The effects of reward magnitude were studied in human paired-associate learning. Experiment 1 tested predictions from the Expectancy hypothesis (Estes, 1969) in a modified version of the task used by Keller et al. (1965). The rate and level of learning varied with the magnitude of reward in both noncorrection and correction procedures and the results did not support the Expectancy hypothesis. In Experiment 2 tests were made of predictions from the Effect hypothesis (Thorndike, 1935) and the Incentive-motivation hypothesis (Spence, 1956). During acquisition, learning varied with the similarity of the eliciting and reward stimuli and of the instrumental and goal responses. In a transfer phase of Experiment 2, subjects were required to learn a new response while reward values were held constant. Rate of learning was similar for different reward values during transfer, particularly for subjects changed to instrumental- and goal-response similarity. These findings did not support the Effect hypothesis but were consistent with the Incentive-motivation hypothesis. Learning differences for different reward magnitudes were regarded as deriving mainly from the acquisition of some anticipatory component (r_g) of the response to reward.