Reference effects in multiattribute evaluations

Abstract

The studies reported here examined the differences in evaluative ratings given to pairs of multiattribute alternatives that were, respectively, at the "worst" and "best" plausible levels on one target dimension, but at the same constant level on the remaining dimensions. Results of the first study indicated that evaluative differences were significantly greater when nontarget dimensions were held constant at their "typical" (usually intermediate) levels rather than at their worst levels. In the second study such differences were larger still when the nontarget dimensions were maintained at their best levels. The results are interpreted as evidence of a systematic violation of requirements for linear representations of subjective multiattribute evaluation policies. Supplementary data analyses and simulations of decision situations were conducted to assess the implications of the revealed reference effects for decision analysis procedures. Those results suggested that ignoring reference effects and assuming a linear model lead to prescriptions that may well be inappropriate.