The Effects of Category Generalizations and Instance Similarity on Schema Abstraction

Three experiments were designed to differentiate two models of schema abstrac- tion. One model, called the generalization model, proposes that category gen- eralizations, defined as feature combinations that occur frequently across study items, are abstracted during learning and used to classify transfer items. Ac- cording to the other model, called the instance-only model, transfer items are classified according to their similarity to studied items. Study materials were constructed that either yielded category generalizations (generalize condition) or did not (control condition). Transfer items differed on whether they were classifiable by category generalizations and on their similarity to study items. In Experiments 1 and 3, accuracy and confidence on transfer items were better in the generalize condition than in the control condition. Experiment 2 manip- ulated the order in which generalizable study items were presented for study: Items were either blocked, so that those contributing to a category generalization occurred close in the study sequence, or randomly ordered. Study items were learned faster and transfer performance was better with blocked presentation than with random presentation. In all three experiments, there was an effect for the similarity of transfer items to study material. There was some evidence indicating better transfer performance on novel items that partially, rather than completely, fit a category generalization. The results support a schema abstrac- tion model in which transfer is a function of similarity both to specific category instances and to higher order category information abstracted from those in- stances.