Range Sensitivity of Attribute Weights in Multiattribute Value Models

Abstract

In decision analysis, multiattribute value functions are normalized relative to the best and worst outcomes in the local decision context. With this normalization, attribute weights (scaling constants) should vary as a function of the range of outcomes on each attribute in the local context. Other things being equal, the greater the range of outcomes for attribute X, the greater the weight for attribute X should be. Two experiments tested this normative range-sensitivity principle for three weight assessment methods: direct importance weights, trade-off weights, and swing weights. Weights elicited using the direct importance weight method were range-insensitive, contrary to the standard normative model. Weights elicited using the swing weight and trade-off methods both displayed statistically significant range sensitivity, although both methods were less range-sensitive than predicted by the normative model. The studies revealed another bias as well. Trade-off judgments gave greater weight to the most important attribute than did direct importance ratings or swing weight assessments.