The Learning of Multiattribute Judgment Policies

Abstract

In this article we explore the process by which consumers learn multiattribute rules of judgment. We report on two experiments in which subjects are asked to learn, through induction, a multiattribute rule that defines product quality in a novel category. In the first experiment we find that subjects are capable of predicting the outcomes of a single multiattribute rule with as few as four bits of feedback, with the primary method of evaluation being comparisons to examples. There is, however, a decided bias in these judgments: subjects are able to learn the attributes associated with a good option more rapidly and with greater accuracy than those associated with a bad one. The second experiment reinforces these earlier results, and offers two additional insights: the ability to learn rules is largely independent of the degree to which subjects can control their own method of learning, and subjects use configural judgment policies to arrive at predictions even when the true generating rule is additive in nature. We discuss implications of the findings for current research in consumer choice analysis.