This study addresses the question of whether conventional models based on cross-sectional data alone account for trip generation adequately. Alternative model formulations are examined using a panel data set to determine whether elements associated with past time points should considered in trip generation analysis and, if so, which elements should be introduced into the model. The analysis shows that estimated model coefficients and t-statistics differ substantially depending on model specification, and that allowing for serial correlation and incorporating a lagged dependent variable both significantly improve the model's fit. The results indicate that serial correlation should be incorporated whenever feasible for efficient estimation and improved fit and that it is safer to ignore state dependence(dependence of observed trip generation on that from a previous time point) and incorporate serial correlation, than to ignore serial correlation and incorporate state dependence. This significance of serial correlation, which presumably is due to omitted variables that are longitudinally correlated, suggests that important determinants of trip generation lie outside the set of variables that have traditionally been considered in travel behaviour analysis. This paper appears in Transportation Research Record No. 1203, Demand Forecasting and Trip Generation-Route Choice Dynamics.
Samenvatting