Applying a heterogeneous random utility model to data from conjoint experiments

This study provides an alternative model, called the ``heterogeneous random utility model'' (HRUM), and the accompanying estimation method suitable for conjoint (survey) experiments. In the HRUM, estimation can be made for specific option combinations, and analyses can be conducted assuming linearity of attributes on the effect of options. Importantly, inference about probability of choosing becomes straightforward.Since the HRUM lacks an analytical solution, finding the maximum likelihood estimators (MLE) for option-specific parameters may incur substantial computational costs. It can be shown that the Hessian function of the log-likelihood of the HRUM is concave, facilitating the use of Newton's method for rapid optimization.An analysis using existing data revealed that estimates from the HRUM are more precise than those from the conventional logistic model. A related simulation study also showed that the HRUM exhibits similar large-sample properties as those from the MLE.