Improving the Predictive Power of Conjoint Analysis: The use of Factor Analysis and Cluster Analysis

A method is derived to improve the predictive accuracy of conjoint analysis by grouping respondents with similar preferences. The often-used model of cluster analysis is shown to be inadequate because real respondents do not form densely packed clusters in preference space. The author then derives the best method of weighting respondents in the sense of maximizing predictive accuracy in conjoint analysis. This method turns out to be a form of Q -type factor analysis. This “optimal weighting” method is shown to perform better than cluster analysis and individual-level analysis in Monté Carlo examples and in real data. Optimal weighting is contrasted with other methods for improving conjoint analysis, and recommendations on their use are made.