Estimation of Duration Model Based on Aggregated Distributions of Arrivals and Departures

Abstract

In order to estimate a duration model describing the staying duration of trip makers, disaggregate data of arrival time and departure time for every sample are generally used. But such disaggregated survey needs a lot of labor and cost. This paper proposed two estimating methods for the duration model containing time varying factors, solely based on the aggregated distributions of arrivals and departures. One is a grid search method of parameters minimizing the summation of errors squared, the other is a maximum likelihood method for the traffic counts reproduced by Detroit method as to meet the aggregated observations. Those two methods were applied to the staying duration model which explains the staying days number of the excursionist who visits Shikoku from Honshu, using the traffic count data across the Honshu Shikoku Bridges, and the applicability of the proposed methods was shown. The estimated models clarified influence by holiday arrangements and the weather on the duration.