Fig.1. Proceduraldiagramshowingthereversalintheprobabilityofobtainingfoodoneachkeyatafixedtimesincethemostrecentfooddelivery.Because all foods were arranged according to a VI 45-s schedule (1.33 food deliveries per minute), food could be delivered at any time. If food was obtained, the time to the reversal reset, regardless of whether that food was obtained before or after the reversal.
No changeover delay (Herrnstein, 1961) was used. Conditions lasted for 21 sessions, and data from all sessions were used in the analyses.
In each session, a series of trials was arranged, during which two keys were lit and available. An exponential (constant probability) VI schedule arranged food deliveries at the rate of 1.33 foods/min. Each trial ended in a food delivery, and the subsequent trial started immediately after the food delivery ended. Thus, the end of food marked the start of a new trial, but food deliveries did not themselves signal additional information about the current contingencies. Because food deliveries were arranged on a VI schedule, trials were not fixed in duration. The VI schedule was held constant across conditions, and hence the average trial duration also remained constant across conditions.
Although food could be obtained at any time during a trial, the response more likely to produce it depended on how much time had elapsed since the start of that trial. At the beginning of each trial, one response (a peck to the left or right key, depending on the condition) was nine times more likely to produce food than the other response. Once a fixed amount of time had elapsed since the start of a trial, the probability of producing food reversed between the two responses. Thus, one key had a higher probability of delivering food before a reversal, and a lower probability of delivering a reinforcer after a reversal – we term this the Higher-to-Lower (H → L) key. The other, Lower-to-Higher (L → H) key had a lower probability of delivering food before a reversal, and a higher probability after a reversal.
Once food had been allocated to the next response on a key, it remained available on that key until collected. However, if food was arranged on a key, but not obtained before the reversal, the availability of this unclaimed food changed to the other key at the point of schedule reversal. This ensured the integrity of the abruptness of the reversal.
Fig. 1 shows a diagram of the reversal in probabilities. Assuming perfect discrimination of the contingencies, the search for the next food delivery might be conducted on the basis of a rule such as “if food is going to occur before the reversal, then it is nine times more likely to be obtained for a peck to the H → L key, whereas if food is going to occur after the reversal, then it is nine times more likely to be obtained for a peck to the other key, the L → H key.” The decision about whether or not the reversal has occurred would be made on the basis of time since the most recent food delivery.
We manipulated the discriminability of the stimuli signaling which response was more likely to produce a reinforcer in two ways: By changing the reversal time; and by changing the nature of the cue that accompanied the reversal. In some conditions, the reversal occurred 10 s after the most recent food delivery, whereas in others it occurred 30 s after the most recent food delivery. Because error in discriminating elapsed time increases in proportion to the duration to be estimated