When a choice set consists of a distribution of alternatives with correlated benefits and costs, consumers often exhibit single-peaked preferences-they prefer an alternative having moderate costs and benefits. Theories disagree about how adding additional lower benefit/lower cost or higher benefit/higher cost alternatives to this choice set will affect relative preferences for the initial set of alternatives. Prototype theory predicts that adding alternatives should produce assimilation, whereas multiattribute range-frequency theory predicts that it creates contrast. We reconcile these two theories by assuming that single-peaked preferences reflect a composition of underlying benefit and cost valuations. Moreover, we claim that the correlational structure of the benefit and cost dimensions in the contextual stimuli determines whether these stimuli will exert an assimilation or contrast effect. We show that when benefits and costs are correlated (uncorrelated), adding alternatives that extend the range of offerings produces assimilation (contrast) for preference judgments. We propose a cost-benefit trade-off model that incorporates elements of single-peaked preference theory and range-frequency theory to explain the complex fashion in which contextual stimuli affect consumer ideals.

for assistance with the data collection. The authors also thank Bryan Orme of Sawtooth Software for providing the software and assistance related to the experiments and the two anonymous JMR reviewers for their constructive comments. John Hauser served as associate editor for this article.

Three experiments are presented in which different aspects concerning Heit and Barsalou's (1996) instantiation principle were investigated. Mean typicalities of subordinate categories within superordinates were predicted very accurately for all investigated concepts. Multiple instantiations were shown to yield somewhat better predictions than single instantiation. The instantiation principle also successfully predicted mean typicalities on a different level (i.e., in lower-level concepts). An alternative account of Heit and Barsalou's findings was also proven wrong. Finally, correspondence between empirically obtained and predicted standard deviations is argued to be dubious, because of several possible sources of bias in the observed and predicted values.