An important problem in transportation planning is the modeling of patterns of trip-making-especially trips from home (origin) to work (destination) within a fixed geographic area. The standard tool used to study such OD flows is the so-called "gravity model," a Poisson log-linear regression model for studying the number of trips from origins within one element of a fixed partition area (traffic analysis zone, for example) to destinations within another.
We present an alternative: a gridless Bayesian hierarchical Poisson\slash gamma random field model, allowing us to incorporate spatial correlation explicitly. The models are fitted to a subset of data from the 1994/95 METRO survey of Portland, Oregon, and are internally validated on a reserved portion of the survey data. Bayesian posterior probability distributions are calculated using a Markov chain Monte Carlo integration scheme based on a novel method for simulating samples from gamma random fields.