Poisson mixture regression for Bayesian inference on large over-dispersed transportation origin-destination matrices

Abstract

We propose a statistical modeling approach as a viable alternative to traditional transportation models concerning inference on origin-destination (OD) matrices. To this end we utilize Poisson mixtures in order to model a large over-dispersed OD matrix derived from the 2001 Belgian travel census. Bayesian methods are using a novel Poisson-inverse Gaussian model. As shown the model has desirable attributes both in its marginal and in its hierarchical form.