The Effect of Blank Trials on Probability Learning

Abstract

This paper reports an experiment on the effect of blank trials on probability learning. A blank trial (Eo) is when neither reinforcing event (El or E2) is presented after a prediction response (Al or A2) is made. In this paper, only the noncontingent case of probability learning will be considered, where reinforcing events do not depend on responses. The experiment to be reported here can be considered to be of a more general interest than with probability learning per se. The data may be applicable to investigations of learning under almost any partial-reinforcement regime, especially that with concurrent schedules where subjects get different reinforcement probabilities on each schedule. Blank trials are characteristic of such schedules; for instance, the concurrent schedules VI-1 and VI-3 provide reinforcement after every minute (on average) on one schedule and after every three minutes on the other. Many responses on such schedules go unreinforced. Both in probability learning and with concurrent schedules, a matching of response probabilities to reinforcement probabilities is typically obtained. Matching has been of interest to operant researchers recently. Hermstein (1974) and Baum (1974) have specified some theoretical and logical properties of matching. In 1974, Baum argued that matching will not be obtained when the discrimination between alternative responses is poor. His arguments may be applicable to probability learning, and there may already be some direct tests of them in the literature on probability learning. Clearly, then, there would be an advantage gained if a close connection between matching in probability learning and with concurrent schedules could be established.