Choice-Behavior and Distribution of Reinforcement among Response-Alternatives

Abstract

According to the statistical associative model frequently applied to modified verbal-conditioning,1 each reinforcement of one of the k-possible responses results in a specifiable change in the probability of alternative Al. That is, there is added to the existing probability (Pi) and increment 0 (1 P), if Ai is reinforced, or a decrement, -OPi, if a response other than Ai is reinforced. This version of statistical associative theory, when applied to a situation in which there are more than two responses, incorporates the simplifying assumption that these changes depend only upon the probability of Ai and are invariant with the distribution of the remaining probability among the alternative responses. In contrast to this formulation, it may be proposed that in a situation involving more than two possible responses, the size of the decrement in the probability of Ai may depend upon which of the other alternatives is reinforced. Likewise, the incremental effects of a reinforcement of Ai may depend upon the distribution of values of the alternative responses. Such a position is suggested within the context of Hull-Spence theory when it is assumed that the probability of a given response is a function of the probability that the momentary strength of that response is greater than the momentary strength of any one of the alternatives.2 According to this view, concentrating reinforcements in order to build up a single strong alternative should be more effective in reducing the probability of Ai than distributing those reinforcements among several weaker alternatives. Two studies which have questioned the adequacy of the statistical associative model might be interpreted as supporting the latter suggestion. Both Gardner, and Cotton and Rechtshoffen have found that the asymptotic frequency of the most often reinforced response is greater if the