Modeling United States railcar flows, 1985--2002 using Bayesian methods with Markov random fields

The U.S. rail system provides an interesting focal point for examining flow modeling and network structure. Little work has been done on rail flows and network relationships since the passage of the Staggers Act in 1980. Much of the geography literature on transportation has moved to the, perhaps, trendier topics of airlines, urban mass transit, or environmental issues. Recently, pedestrian transportation has received more attention from transportation geographers than railroads. While disconcerting that a major transportation backbone within the United States has received such little academic interest in recent times, this lack of inquiry allows this study to make a unique, and hopefully timely, contribution to the literature on modeling flows and on railroad dynamics.;This study proceeds from the belief that an examination of railroads and railcar flows provides important insights into economic geography and to the economics of transportation. These flows are examined using carload waybill survey data involving a relatively new modeling methodology. The waybill dataset provides an excellent historical reference of flows and flow patterns across the United States for the period after the Staggers Act.;The new modeling methodology, spatial Markov random fields applied to a network combined with a Bayesian auto-model framework, allows for a robust predictive estimation of railcar flows that is shown to be consistent over time. This method represents an improvement over standard gravity or spatial interaction models; models may be structured as simple spatial interactions or using more complex model forms.;The flow estimation results and patterns are then compared to the expected results that implied by three different models of economic geography. These models are the "new" economic geography represented by Fujita, Krugman and Venables, an alternative specified by Glaeser and Kohlhase, and an `old" economic geography model provided by Taaffe, Morrill and Gould. Support is found for the Taaffe, Morrill and Gould model, while implications for the two "new" economic geography models are mixed.