THE COMPUTATIONAL MODELING APPROACH

Decision researchers have struggled for a long time with the fact that preferences are highly changeable and vary in complex ways across contexts and tasks. For example, reversals have been observed when preferences are measured using binary versus triadic choice sets or when preferences are measured by choice versus price methods. Several theoretical approaches have been developed to understand this puzzling variability in preferences. One approach is to modify the classic utility model by allowing the weights or values that enter the utility function to change across contexts or tasks. For example, Tversky, Sattath, and Slovic (1988) believe that the decision weights for attributes change across choice versus price tasks. A second approach is to use different heuristic rules to form preferences, depending on task and context. For example, Payne, Bettman, and Johnson (1993) propose that decision makers switch from compensatory to noncompensatory types of rules when the number of options increases or as time pressure increases. Both of these approaches are well established and have made a large impact on decision research.

This chapter presents a computational approach to understanding how preferences change across contexts and tasks. According to this approach, preferences are constructed from a dynamic process that takes decision contexts as inputs and generates task responses as outputs. Computational models are formed by a collection of microprocessing units, each of which performs an elementary cognitive or affective evaluation.