An Extended Random Coefficients Model, with Application to Metric Conjoint Analysis

The authors present a modeling technology that extends the standard random coefficients model (RCM) by allowing for (1) error variance heterogeneity, (2) a parsimonious, factor-analytic representation of coefficient heterogeneity, and (3) the estimation of one or more unobserved predictors. The RCM and its extensions are motivated and assessed in the context of metric conjoint analysis. The value of each extension is investigated systematically by fitting to three data sets different combinations of model specifications which are generated according to an experimental design. We find that all three extensions are essential to adequately represent the data. In addition, using models that lack these extensions often alter conclusions about substantive issues such as which factors drive consumer preferences. A robust implementation of the extended random coefficients model allows researchers to assess and accommodate departures from the model's parametric assumptions