Dynamic travel demand models incorporating un observed heterogeneity and First-order serial correlation 【Article】

Very little work using repeated cross-sectional data has been undertaken in transport research. This is especiallytrue for travel data gathered at multiple points in time, especially data that is gathered every 5-10 years such asUrban Area Travel Survey Data and Road Traffic Census Data in Japan. Accordingly, travel demand modelingbased on these types of data is not yet fully developed. This paper deals with methods for developing models whichinclude time series factors for predicting travel demand using three time-points travel data gathered in Hiroshima.As a result, it was shown that model parameters based on cross-sectional data were not stable over time by usingCovariance Analysis or T-Statistic. The existence of first-order serial correlation in residuals was confirmed byusing Generalized Durbin-Watson Statistics, while unobserved heterogeneity was checked by using Breusch-PaganStatistics. Fixed-effects models using these two factors were developed and it was shown that their predicting accu-racy was improved in comparison to traditional cross-sectional models.