Training Effects on Performance in Concept Identification Problems with Various Types of Solution Shifts

Abstract

Extensive training with same-class concept identification problems was expected to produce behavior that is consistent with hypothesis sampling models that assume rather severe memory limitations for previously tested hypotheses. After solving 0, 4, or 8 unidimensional concept problems, 256 undergraduates were transferred to a problem in which solution was (a) present on a single dimension—a no-shift problem, (b) reversed within a dimension after every second error—a reversal shift, (c) alternated between two dimensions after every second error—a dimensional shift, or (d) based on either of two dimensions—a relevant, redundant problem. Results showed that performance was better on the no-shift problem than the reversal shift problem but only at 0-pretraining. Performance was better on relevant, redundant problems than dimensional shift problems at 0- and 4-pretraining. There were frequent inconsistencies of hypotheses with previous information in the 0- and 8-pretraining conditions. The interpretation was that associative learning occurs among some inexperienced subjects and that with a moderate amount of training hypotheses are consistent with previous information. Memory is extremely limited when same-class problems are given repeatedly.