No correlation, no evidence for attention shift in category learning: different mechanisms behind illusory correlations and the inverse base-rate effect.

Abstract

When common and rare attributes are equally prevalent for frequent and infrequent categories, the frequent categories (e.g., majority groups) are associated more strongly with the common attributes (e.g., positive valence) than infrequent categories (e.g., minority groups). Such a frequency-based illusory correlation (IC) effect has been shown to arise through unbiased learning, which is less complete for infrequent than for frequent categories. On the other hand, when frequent categories are always paired with corresponding common attributes and infrequent categories with corresponding rare attributes, an inverse base-rate effect (IBRE) arises. The association of the infrequent categories with the rare attributes is stronger than the association of the frequent categories with the common attributes. Recently, it has been proposed that the attention shift mechanism that produces the IBRE is also essential to explaining IC effects (Sherman, Kruschke, Sherman, Percy, Petrocelli, & Conrey, 2009). No evidence was found for this explanation of standard IC effects across 4 experiments and related computational modeling of attention shift (using a model called EXIT, derived from "extended ADIT" model, where the name ADIT is an acronym for "attention to distinctive input"). In a fifth experiment, evidence for attention shift was found for perfect category-attribute correlations. In sum, incomplete learning continues to offer a sufficient and parsimonious account of IC effects.