Two experiments examined the effect of category variability on updating of category beliefs. In Experiment 1, participants learned about categories via exposure to numerical distributions of attributes. These distributions had the same mean value, but had either high or low levels of variability around the mean. After category learning participants were shown new category exemplars that contradicted the acquired beliefs (i.e. drawn from a sample with higher or lower means) and were asked to

more »

... stimate the category central tendency. Importantly, this new data was purportedly sampled ten years after the original sample. Beliefs about the attributes of low variable categories were more strongly influenced by disconfirming information than beliefs about the attributes of high variable categories. In Experiment 2, the same procedure was followed except that half the participants were told that the training data and the new, contradictory exemplars were drawn from the same sample (i.e. no ten year interval between training and test exemplars). Under these conditions, beliefs about the attributes of the categories were still influenced by disconfirming information, but the effect of variability was reversed. This suggests that temporal construal is an important factor in determining responses to disconfirming information. The results are discussed in terms of a Bayesian model of belief revision.