Simulation Experiments in Choice Simplification: The Effects of Task and Context on Forecasting Performance

Abstract

Experimental and protocol-based research has demonstrated convincingly that consumers frequently use simplification heuristics prior to making choices. Consequently, quantitative choice models incorporating simplification strategies recently have received much research attention. Given the additional computational cost and complexity involved in estimating these models, the authors investigate how robustly the standard logit model accommodates choice simplification processes. The results show that when consumers screen brands using an elimination-by-aspects strategy, logit forecasts choices reasonably well, but the parameter estimates are severely biased. However, when consumers screen through brands previously purchased, augmented by brands promoted, logit both forecasts poorly and produces biased parameter estimates, even when dedicated screening variables are included in the logit utility function. The predictive validity of logit improves somewhat when the universal set size is small and when purchase event feedback is less important.