A flexible non-normal random coefficient multinomial probit model: Application to investigating commuter's mode choice behavior in a developing economy context

There is growing interest in employing non-normal parameter distributions on covariates to account for random taste heterogeneity in multinomial choice models. In this study, we propose a flexible, computationally tractable, structurally simple, and parsimonious-in-specification random coefficients multinomial probit (MNP) model that can accommodate non-normality in the random coefficients. Our proposed methodology subsumes the normally distributed random coefficient MNP model as a special case, thus eliminating the need to a priori decide on the distributional assumption for each coefficient. The approach employs an implicit Gaussian copula to combine the univariate coefficient distributions into a multivariate distribution with a flexible dependence structure. Using our new flexible MNP framework, we investigate the commute mode choice behavior for workers in the city of Bengaluru, a metropolitan city in southern India. Results from our analysis indicate that sociodemographic variables, commute characteristics, and mode-related attributes significantly impact the commute mode choice decision. Importantly, our results indicate the presence of unobserved taste heterogeneity in the sensitivities to the travel time and travel cost variables; moreover, the distribution of the travel time coefficient is found to be significantly non-normal. In terms of data fit, our proposed model statistically outperforms the traditional MNP model as well as an MNP model that imposes normality on the travel time coefficient. The pitfalls of ignoring non-normality in the distribution of parameters are also discussed, as are several policies to promote a shift from private modes of transportation to more sustainable public transportation/walk modes.