When Choice Models Fail: Compensatory Models in Negatively Correlated Environments

Abstract

Linear compensatory models, which involve tradeoffs between product attributes, have been argued to provide reasonably good predictions of choices made by noncompensatory heuristics, which do not involve tradeoffs. This robustness to misspecification of functional form may fail, however, when there are negative correlations among attributes in a choice set. A Monté Carlo simulation demonstrates that certain noncompensatory rules are poorly fit by linear models, even in orthogonal environments, and that this fit diminishes further in nonorthogonal environments. Two laboratory experiments assess the extent to which such model failure might arise in natural contexts. The first, a process-tracing analysis, examines the decision strategies consumers use in nonorthogonal choice environments. The second explores the ability of a compensatory choice model calibrated on actual choices to predict decisions made in orthogonal and nonorthogonal contexts. The authors conclude with a discussion of the work's implications for current research in applied choice modeling.