Reinforcement for errors in a signal-detection procedure.
Reinforcing errors, even rarely, reliably lowers discrimination accuracy in a matching-law fashion.
01Research in Context
What this study did
Researchers used pigeons in a signal-detection task. Birds pecked one key if a light was bright, another if it was dim.
They changed how often wrong pecks earned food. Error payoffs ranged from zero to almost the same as correct pecks.
What they found
When errors paid off more, accuracy dropped. Even tiny error rewards hurt discrimination.
A simple matching equation predicted the exact drop. Birds matched their choice ratio to the payoff ratio.
How this fits with other research
Rider et al. (1984) added a delay between sample and choice. Accuracy still followed the matching rule, showing the law holds across time gaps.
Thrailkill et al. (2025) seems to disagree. They showed that higher payoff on one element later drew more attention to it. Yet both studies obey the matching law; one shows harm from reinforcing errors, the other shows benefit from reinforcing features.
Campbell (2003) found that at extreme payoff ratios pigeons stopped looking at the sample and just side-biased. Together these papers warn: any unintended payoff for wrong responses can quickly erode stimulus control.
Why it matters
If a token, praise, or escape accidentally follows an incorrect response, you are paying for errors. Even once in a while is enough to weaken discrimination. Check your data sheets for accidental reinforcement. Make sure only correct responses contact reinforcement until the skill is solid.
Want CEUs on This Topic?
The ABA Clubhouse has 60+ free CEUs — live every Wednesday. Ethics, supervision & clinical topics.
Join Free →Audit your error-correction procedure: ensure incorrect responses never produce tokens, praise, or escape.
02At a glance
03Original abstract
Six pigeons were trained on a signal-detection procedure. They were required to peck the left key when a 5-second white light had been presented, and the right key when a 10-second light had been presented. These two correct responses were followed by food reinforcement with a probability of .7. Errors, left responses after the 10-second stimulus or right responses after the 5-second stimulus, were initially followed by a 3-second blackout of the chamber. In nine subsequent experimental conditions, errors were followed by food reinforcement with increasing probability while the probability of reinforcement for correct responses was kept constant. The percentage of correct responses decreased as error reinforcement probability increased. A matching model of detection performance, in which discrimination is a joint function of stimulus discriminability and stimulus-reinforcement association, provided a convincing fit to the data and to two sets of published data. The model also fitted published data on multiple and multiple-concurrent free-operant performance. This description of detection performance in terms of matching offers both accurate prediction of complex behavior and measures of discriminability with wide generality.
Journal of the experimental analysis of behavior, 1980 · doi:10.1901/jeab.1980.34-35