Abstract
The majority of demand models used at present are based on cross-sectional data. Behavior, however, is temporally related. Using three waves of a panel data of 2200 commuters in Southern California, this paper conducts a comparative analysis of three types of models: an ordered probit model, a two-period joint choice probit model, and a two-period dynamic beta logistic model. The choice behavior modeled is the choice between driving alone and sharing a ride for the work commute. Prediction tests both on a hold-out sample as well as on a forecast sample were conducted. With the hold-out sample, all three models performed similarly. With the forecast sample, however, the beta-logistic model performed better than the other two models in aggregate predictions, and approximately the same as the joint choice probit model in disaggregate predictions, while both of these models performed better than the ordered probit model.
