Score contribution per author:
α: calibrated so average coauthorship-adjusted count equals average raw count
The logit‐mixed logit (LML) model advances choice modeling by generalizing previous parametric and semi‐nonparametric specifications and allowing retrieval of flexible taste distributions. Using standard operating conditions in the field, we report results from Monte Carlo experiments designed to assess the finite sample bias‐variance tradeoff for the LML using as a benchmark conventional Mixed logit models (MXL) under asymmetric and multimodal taste distributions. The LML specification always outperforms the MXL in terms of bias, but when the variance around modes is high the mean squared error (MSE) is lower than that of MXL only at sample sizes larger than usual and with some nuances. D‐error minimizing experimental design predicated on multinomial logit significantly reduces MSE, but no clear winner is found between polynomial, step, and spline functions for the multidimensional grid function. Analysis of empirical data from a choice experiment on tap water shows that multimodality emerges only if higher number of node parameters are used in the LML.