The adequacy of conventional linear regression models in trip generation analysis is examined in this study. Simulation experiments are conducted to determine whether model coefficients can be accurately estimated by least-squares estimation when the dependent variable is a nonnegative integer. Following this, nonlinear, two-stage model systems are estimated by using an empirical data set to examine whether more elaborate representation of the decision process underlying trip generation will lead to improved prediction. The results of this study indicate that linear regression models of trip generation offer consistent coefficient estimates and accurate predictions, and improved performance may not be obtained by adopting more complex model systems. This paper appears in transportation research record no. 1220, Forecasting.
Abstract