Effectiveness of Bayesian Updating Attributes in Data Transferability Applications

Transportation Research Record: Journal of the Transportation Research Board, No. 2344, Transportation Research Board of the National Academies, Washington, D.C., 2013, pp. 1–9. DOI: 10.3141/2344-01 T. H. Rashidi, School of Civil and Environmental Engineering, University of New South Wales, Room CV113, Building H20, Sydney, New South Wales 2031, Australia. J. Auld and A. Mohammadian, Department of Civil and Materials Engineering, University of Illinois at Chicago, 842 West Taylor Street, Chicago, IL 60607. Corresponding author: T. H. Rashidi, [email protected]. rich literature behind them. Recently, data transferability models have been more frequently employed by small and midsize local areas as an alternative (1, 2). The most commonly used formulation in transferability modeling is the Bayesian updating method (3, 4). A simple conjugate normal–normal Bayesian updating procedure is a typical formulation that has been employed for updating; in it, both the prior distribution from which to transfer and the posterior transferred result are assumed to have a normal distribution for the parameter of interest (3). Unfortunately, many attributes of the Bayesian updating method have generally been overlooked in both these updating models and data transferability studies in the transportation field, and this fact could potentially limit their effectiveness and applicability. For example, the effectiveness of nonconjugate distributions, noninformative priors, and many other alternative types of Bayesian updating formulations have not been studied and discussed in the literature. This study examines several Bayesian updating scenarios in which different issues about Bayesian updating are addressed. One fundamental aspect of Bayesian updating is the capability of incorporating prior information about the dependent variable. However, it is possible that the use of inappropriate prior information may result in deceptive findings and would not necessarily improve the final result. Therefore, several updating scenarios with different levels of prior information, including current and out-of-date national information, were used to investigate this possibility. Another issue is the determination of whether the use of more complex Bayesian updating formulations, such as the inclusion of random effects or nonconjugate prior distributions, will produce a better model fit. The results of the study generally show that Bayesian updating is a tool that should be cautiously employed. It can improve the model fitness and lead to better results; however, it can also lead to unintended consequences and reduced model performance if employed improperly. Therefore, the Bayesian updating method should be used with great care and consideration, and the strengths and weaknesses of the method should be taken into account. The major objective of this study is to first demonstrate the potential of Bayesian updating to improve data collection efforts when large samples are unavailable while also analyzing some of the commonly used types of Bayesian updating attributes so as to find a yardstick for validating the strengths and weaknesses of each of them in real data transferability applications. The rest of the paper is organized as follows. Initially, a literature review of travel data transferability models and Bayesian updating procedure is presented. Following that, data sources that are used in this study are introduced. Then, the modeling methodology and results are presented and discussed. Finally, conclusions and future research directions are presented. Effectiveness of Bayesian Updating Attributes in Data Transferability Applications