Dynamic Heterogeneity in Discrete Choice Experiments

Abstract

In discrete choice experiments, respondents’ decisions are studied based on repeated tasks. Especially when conducting longer sequences of tasks to gain more information, learning or fatigue may come into play and affect the identification of the effects. These effects may be exacerbated by customer-specific heterogeneity. We introduce Bayesian multinomial logit models with heterogeneously time-varying coefficients constructed as tensor products of random effects and penalized splines to capture time variation and customer-specific heterogeneity. In an empirical application, we find evidence for the presence of such effects. Our approach outperforms benchmark models in fit and predictive accuracy, as also demonstrated through a simulation.

Keywords: Choice experiments, Functional random effects, Multinomial logit model, Penalized splines, Tensor products