Role of Errors in Concept Identification

Abstract

216 Ss were given a concept-identification problem wherein Ss were required to make an overt button-pressing response, as an observing response, in order to view each dimension of an instance. On each trial, Ss were allowed to view any dimensions they desired with the constraint that they were allowed to view only half of the available dimensions. On those trials in which the one relevant dimension was not observed, Ss were given misinformative feedback. Three levels of misinformative feedback, designed to vary the probability of S's classification response being called correct ( p), were used. It was found that mean number of errors to solution was a function of problem difficulty and saliency but independent of variations in p as predicted by Trabasso and Bower (1968). The results did not support the strength theories in concept identification.

Thinking

“…Using this procedure, Chatfield, Van Dyke & Powell (20) found that increasing the number of times a: subject is "correct" when the relevant dimension is not sampled (b y providing misinformative fe edback) has little effect on errors to criterion, adding weight to the standard assumption that one error is sufficient to eliminate an H from consideration (at least temporarily) no matter how many times it has been correct in the past. However, Matthews & Patton (112) fo und that the probability of not changing an H after a disconfirmation (a "lose-stay") increased with the length of a criterion run of H-probes, reaching .67 after 8 such trial blocks.…”