Modeling discrete choices with large fine-scale spatial data: opportunities and challenges

Discrete choice models have played a pivotal role in modeling spatial & regional systems in the past few decades. Various extensions for the classic discrete choice models have been developed, including spatial discrete choice models and dynamic discrete choice models. The two categories of models represent methodological developments in the spatial dimension and the temporal dimension of spatial data modeling, respectively. With the growing availability of spatial data in large and more refined scales, spatial and dynamic discrete choice modeling techniques face methodological and computational challenges. By comparing different existing estimation methods, we show that the Bayesian MCMC (Markov chain Monte Carlo) method is the most computationally efficient estimation method for explicitly modeling the spatial relationships with large data sets. As far as dynamic discrete choice models are concerned, most of the existing research effort focuses on improving the model performance in capturing the inter-temporal structures of decision making. In addition, there is an anticipated need to integrate the spatial dimension and the temporal dimension in spatial data modeling. It will require (1) effective solutions to address the computational challenges; and (2) theoretical and methodological innovations in characterizing spatial data processes. We also discussed opportunities and challenges related to empirical analysis with spatial big data, including reviewing several open-source software packages for estimating spatial discrete choice models. Lacking continuous development support and limited applicability are the main issues that prevent them from being widely used.