Predictive Models as a Guide to Preference

Abstract

Multiattribute decomposition has been shown empirically to be superior to clinical judgment in evaluating such phenomena as student potential, ultimate psychiatric diagnosis, and longevity of patients with Hodgkins disease. This superiority is maintained even when attributes are combined in a nonoptimal manner. The reason is that people (expert and nonexpert) are much poorer at processing information than at perceiving and coding it. Thus it is hypothesized that multiattribute decomposition will also be superior to clinical judgment in deciding upon a single object or alternative when there is no external criterion of value. Examples are given. The multiattribute apprbach has the additional advantage that each single choice becomes a policy decision, thereby eliminating the types of bad decisions people make when they consider each choice as wholly unique.