Importance of trials versus accumulating time across trials in partially reinforced appetitive conditioning.

Abstract

Four experiments with rats examined partial reinforcement in appetitive conditioning. In Experiment 1, adding nonreinforced trials to a continuous reinforcement schedule slowed acquisition, whereas deleting reinforcers did not. Trial massing suppressed performance and learning. In Experiment 2, conditioning with a short conditioned stimulus (CS) was rapid, and partial reinforcement with a short CS was as effective as continuous reinforcement with equal accumulated time in the CS. In Experiment 3, conditioning was nevertheless influenced by the probability of reinforcement. In Experiments 3 and 4, conditioning was especially disrupted when nonreinforced trials preceded reinforced trials closely in time. The results underscore the importance of temporal variables in conditioning but are more consistent with trial-based accounts than time-accumulation accounts of conditioning.

Pavlovian conditioning and cumulative reinforcement rate.

“…Thus our experiments did not confirm the difference in responding that was evident, although not statistically significant, in Experiment 2 of Bouton and Sunsay (2003). This is despite the fact that our within-subjects design should be more sensitive to differences between conditions than their betweensubjects design.…”

“…However, the comparison of particular interest is between rats' responses to the long 100% CS and the short 33% CS, since these have the same cumulative rate of reinforcement (one US per 60 s of CS). If, as predicted by some real-time trial-based models of conditioning and suggested by the results reported in Bouton and Sunsay (2003), conditioning is affected by trial structure, then D there should be a difference in the rats' response rates to the two CSs. However, if trial structure is irrelevant and rats simply accumulate learning about non-reinforcement across CS exposure, then they should show similar levels and rates of conditioning to the two CSs…”

“…Alternatively, given that the PREE is usually absent in within-subjects designs (Pearce, Redhead, & Aydin, 1997), this raises a possibility that the results we present here are specific to our within-subjects designs. It is thus possible that there would be differences in conditioning between CSs with matched cumulative reinforcement rate but differing in trial-based reinforcement if they were compared between subjects (as in Bouton & Sunsay, 2003).…”

“…In another experiment (Experiment 3), Bouton and Sunsay (2003) compared two groups that were conditioned with the same CS-US interval (10 s) but which differed in the length and number of non-reinforced exposures of the CS. For one group, reinforced trials were intermixed with an equal number of long (40 s) nonreinforced presentations of the CS; for the other group, for every reinforced trial there were four short (10 s) nonreinforced trials.…”

“…This result is consistent with a model like SOP which predicts that the two groups would differ in the mean associability of the CS during non-reinforcement-for the 40-s group, much of the non-reinforced exposure would occur when the CS has low associability, whereas for the other group each moment of non-reinforcement would occur when associability was still high (early in the CS presentation). Bouton and Sunsay's (2003) experiments suggest that the effect of CS-US interval on delay conditioning is not simply a function of cumulative reinforcement rate, but depends on trial structure as implied by trial-based models of learning. This conclusion is at odds with our recent demonstrations that the relationship between response rate and reinforcement rate is the same regardless of how reinforcement rate is distributed across trials (Harris & Carpenter, 2011).…”