Pigeons were trained on a two-component multiple schedule in which each separate component consisted of a three-link chain schedule. After initial baseline training, the stimuli correlated with the terminal links of each chain were presented in a successive discrimination, with one stimulus continuing to be associated with reinforcement while responses to the alternative stimulus were extinguished. Subjects were then returned to the original chain schedule, but with extinction in effect in both components of the multiple schedule. In two separate experiments, extinction of initial-link responding was not affected by which terminal link had been extinguished during the separate discrimination training, indicating that devaluation of the terminal link was not transmitted directly to the initial link of the chain. There was also no effect of the devaluation procedure during the first session of testing on responding in the middle link of the chain, but an effect did develop with continued extinction of the entire chain when the terminal components were presented during extinction. When the terminal components were omitted, however, the latter effect did not occur. Also, when the terminal link was omitted, extinction occurred more rapidly in the middle component than in the initial component, indicating a backward pattern of extinction.A major development in the study of operant behavior has been the appreciation that subjects are sensitive to the specific nature of the reinforcer contingent on the operant response. In a variety of studies (e.g., Adams & Dickinson, 1981;Colwill & Rescorla, 1986), animals have been trained with an operant contingency, and then the reinforcer contingent on the response has been devalued (typically by taste aversion conditioning) in the absence of availability of the response. When the response is then tested without further exposure to reinforcement, its rate is substantially and immediately reduced, relative to control conditions where the reinforcer contingent on a second response has not been devalued. Such results have sponsored the view that "response strength" does not adequately characterize the determinants of the level of operant behavior, and that some notion of direct response-reinforcer associations must be invoked instead (e.g., Dickinson, 1980). This research was supported by grants from the National Science Foundation. Reprint requests should be addressed to the first author, Department of Psychology, University of California, La Jolla, CA 92093-0109 (e-mail: bawilliams@ucsd.edu).
-Accepted by previous editor, Vincent M. LoLordoWhile the results noted above indicate that operant responding may be sensitive to the current value of its contingent reinforcer in simple training procedures, the role of such effects in more complex situations remains to be determined. Consider a chain schedule of reinforcement. Traditional analyses (e.g., Keller & Schoenfeld, 1950) of chain schedule performance have assumed that the reinforcer maintaining such behavior is not the food at the ...