The purpose of this paper is to propose a new method to estimate parameters of discrete choice model disaggregate and aggregate data. There exists some kinds of aggregate data which could be utilized for transportation studies in Japan: the national census, road traffic data, air, rail, and bus passenger data and so on, but travel demand models with these data often contain less information required for transportation planning. The Bayesian approach was employed for combining aggregate data with discrete choice model. Updated parameters of the model could be calculated with the following prior distribution and a likelihood function composed of aggregate data. The prior distribution was assumed to be that of discrete model parameters which were calibrated singly with disaggregate data. In this study, two types of estimation methods are presented: one is the above and another is an approximate technique that is equivalent to the generalized least squares. The property of the estimators were analyzed with person trip survey data in local cities in Japan. The estimators were compared with those of alternative estimation methods: choice-based sampling and constant term modification. Results indicated that new techniques were so effective to build behavioural models when additional aggregate data were available. Furthermore, expansion of the method to the practical usage was also discussed.
Abstract