Predictive Validity of Conjoint Analysis Results based on Best-Worst Scaling compared with Results based on Ranks Data

Abstract

The time-consuming nature of Best-Worst deriving scales for a conjoint analysis study leads the authors to question if ranked data, which could be translated into equivalent B-W scores, might be nearly as useful. Respondents answered a B-W scaling task regarding a familiar, highly involving product category, and then two-weeks later did a sorting task on the same fractional factorial array. Conjoint analysis coefficients were calculated and values were forecast for four holdout profiles. Kendall's tau measured the extent that the holdout ranks were reproduced in conjoint predictions. Results suggest that overall the rank data were comparable to the B-W data, but with greater variance.