Controlling relations in conditional discrimination and matching by exclusion.
Punishment can trim exclusion errors, but you must apply the contingency across many stimuli or the old pattern returns.
01Research in Context
What this study did
Adults played a matching-to-sample game on a computer. They learned to pick the correct picture when given a sample. New samples appeared and the adults used exclusion to choose. Later, wrong choices cost points. The team watched if the point loss changed how people played.
What they found
Point loss cut the exclusion choices, but only for the exact pictures used in punishment. When new pictures showed up, people went back to old ways. To get lasting change, the team had to punish wrong picks across many pictures.
How this fits with other research
O'Hora et al. (2014) later showed the same thing with written rules. They proved that consequences still control 'understood' rules, backing up the 1987 point-loss effect.
Davison et al. (1989) added a twist: a high or low tone told adults which class the sample belonged to. That paper extends the 1987 work by showing context can switch class membership, not just stop exclusion.
Cooper et al. (1990) removed all points and prizes. Some adults still built classes, but results were shaky. It conceptually replicates the training phase of 1987, warning us not to trust equivalence without checks.
Why it matters
If you use timeout or response cost to kill an error pattern, apply it to every stimulus you care about. One-item punishment will not travel. Probe generalization with new items before you call the problem fixed.
Want CEUs on This Topic?
The ABA Clubhouse has 60+ free CEUs — live every Wednesday. Ethics, supervision & clinical topics.
Join Free →When you punish an incorrect match, rotate in at least three more unknown items and punish errors on them too, then probe with completely new samples.
02At a glance
03Original abstract
Normally capable adults learned two-choice identity matching of three-digit numerals and arbitrary matching of physically dissimilar nonsense syllables. The stimuli were displayed on a computer terminal, and responses consisted of typing on the terminal's keyboard. In Experiment 1, every trial displayed a sample numeral, a comparison numeral, and three equal signs (= = =). The comparison stimulus was to be selected if it was identical with the sample; otherwise the equal sign was to be selected. This "single comparison" method was then used to show that arbitrary matching could be based upon either sample-S+ or sample-S- relations. In Experiment 2, a series of probe trials displayed a novel sample, a comparison stimulus from the arbitrary matching baseline, and = = =. Subjects typically selected = = =; they apparently were excluding the baseline comparison stimulus. Experiments 3 through 5 investigated which variables in training would lead to the selection of baseline comparison stimuli in response to novel samples. Behavior was usually unchanged when baseline training included relating comparison stimuli to as many as four different samples. Punishment contingencies were effective, but performance did not generalize unless those contingencies were applied in relation to more than one baseline comparison stimulus.
Journal of the experimental analysis of behavior, 1987 · doi:10.1901/jeab.1987.48-187