IRRELEVANT INFORMATION IN PROBABILISTIC CATEGORIZATION

Using a probabilistic category learning paradigm, 6 experiments explor ed irrelevant information and 4 current models. Utilization of relevan t configural information was lowered by the presence of an irrelevant dimension, both if that was the only relevant information and if a dim ension was also relevant. An irrelevant cue value lowered the utilizat ion of relevant cue values. An additional relevant dimension had a lar ger degrading effect on the utilization of a relevant dimension than a n additional irrelevant dimension, thereby suggesting that the effect of irrelevant information is due to the complexity of the environment rather than to factors particular to the irrelevant nature of the info rmation. The current models failed to fit the findings. However, resul ts showing that memory errors account for salience effects provide a d irection for revising one of the models.