AN INFORMATION PROCESSING PROBE INTO CONJOINT ANALYSIS*

Abstract

Abstract Protocol analysis was used to identify the actual choice rule used by respondents in a conjoint analysis interview under two different card‐sorting instruction conditions. In both conditions respondents used a noncompensatory rule (conjunctive or lexicographic) and not the additive rule assumed in many conjoint analysis studies. Predictions based upon either the additive rule or the noncompensatory rule were, however, found to be consistent with preference ranks in a holdout sample. The protocols were further analyzed to provide partial validation of the conjoint weights and to investigate the nature of prediction errors observed in the conjoint analysis.