Human Performance on a Signal Detection Task: Discriminability and Sensitivity to Reinforcement

Abstract

Ten human participants worked on a computerbased discrete-trials signal detection task in which stimulus disparity and the ratio of reinforcers for correct responses were manipulated. During each trial, a 12 × 12 stimulus array consisting of an unequal number of randomly arranged circles and squares was presented. Participants responded by indicating whether the stimulus contained more circles or more squares. Two levels of stimulus disparity, high and low, were arranged by changing the numbers of circles and squares in the stimulus. At both levels of disparity, the ratio of reinforcers for correct circle and square responses was varied across 5 different levels ranging from 1:9 to 9:1. Reinforcers were assigned using a controlled reinforcer procedure that held obtained reinforcer ratios close to the programmed ratios. For all participants, discriminability was higher when the stimulus disparity was high but was not affected by changes in the reinforcer ratio. Response bias varied as a function of the reinforcer ratio. Estimates of the sensitivity of bias to changes in the reinforcer ratio varied with changes in disparity; however, the variations were typically small and their direction differed across participants. These findings are consistent with previous human research (Gallagher & Alsop, showing that when a controlled reinforcer procedure is used, bias and sensitivity are independent of discriminability.

Divided stimulus control depends on differential and nondifferential reinforcement: Testing a quantitative model

“…Likewise, generalization between responses (from B 1 to B 2 and vice versa) is quantified by d br , which also ranges from 1 to ∞. Manipulations that degrade stimulus or contingency discriminability, such as reducing the physical disparity between stimuli or inserting a delay between responses and reinforcers, decrease d sb and d br , respectively (Davison & Nevin, 1999; see also e.g., Davison & Jenkins, 1985;Everly, 2016;Godfrey & Davison, 1998). This model is easily extended to the reinforcement-for-errors procedure, in which B 21 and B 12 responses are reinforced, and such reinforcers (R 21 and R 12 ) generalize via the same mechanisms.…”