A case for the generalizability of attribute importance: The constant ratio rule of effects

Abstract

In many environmental and decisional systems evaluative judgments may be viewed as functions of system attributes. Researchers in a variety of applied areas are interested in the importance of such attributes in determining the behavior of individuals interacting with the system. Relative importance can be assessed by the amount of variance accounted for by each attribute. Unfortunately, the generalizability of attribute importance has been questionable. In this paper a rule is proposed that allows for the generalizability of attribute importance across information sets. This rule states that the ratio of the variances accounted for by any two attributes is a constant across subsets of attributes. The empirical support for the invariance of the ratios implies that results may be generalized across informational context. The constant ratio rule and its corollary, the product rule, were supported in an experiment on rating job applicants. In addition, a review of 22 other data sets shows consistent support for the constant ratio rule. The implications of this type of generalizability are discussed in terms of applied research methods and theoretical models of judgment.