Violations of the Local Independence Assumption in Categorization

The Threshold Theory of categorization posits that prior to making a categorization decision a respondent assesses the similarity between item and category and compares it against a personal threshold. Only if the item-category similarity exceeds this threshold should the respondent endorse the item as a category member. The Threshold Theory thus assumes that a single latent variable, item-category similarity, suffices to explain all categorization patterns. We put this assumption, known as local independence, to test by providing respondents with sets of four items to categorize. These items are equated in terms of their similarity to the target category, but are by design comprised of two pairs of similar items, such as rollerskates-skateboard and horse-mule for the category of VEHICLES. Contrary to the local independence assumption, the items within the pairs are more likely to receive the same categorization decision than the items across pairs. We explain how these categorization patterns can be accommodated within the Threshold Theory framework, either by giving up the assumption that everyone assesses item-category similarity in the same way for differently weighted category information, or by explicitly incorporating item-item similarity in addition to item-category similarity.