Multiple Discrete-Continuous Choice Models with Flexible and Partially Monotonic Utility Functions

Abstract

The multiple discrete-continuous extreme value (MDCEV) model estimates individuals’ preferences for multiple alternatives and their usage, while accounting for satiation–the diminishing marginal utility from consuming additional units of each alternative–in a closed form. However, existing models lack flexibility due to specific assumptions about the utility function (e.g., monotonicity and parabolicity), leading to poor finite sample properties and prediction errors when the true data-generating process deviates from these assumptions. This study relaxes these assumptions by specifying the satiation parameters using lattice networks (LN), piecewise linear functions that flexibly model nonlinear attribute effects and employ multilinear interpolation to capture complex attribute interactions. The proposed MDCEV-LN demonstrated high predictive accuracy in budget allocation in a Monte Carlo study. At the same time, it maintains interpretability with added flexibility to accommodate various functional forms, including traditional loglinear satiation trend. Thus, MDCEV-LN offers an accurate, flexible and interpretable framework for discrete-continuous choice analysis.