A joint mode/time-of-day choice model using combined revealed preference and stated choice data

The factors influencing trip departure time are taking more importance in practice since urban congestion is increasingly being addressed by travel demand management (TDM) strategies. In this paper we formulate and estimate a joint travel mode-departure time model for commuting trips using combining revealed preference (RP) and stated choice (SC) data. The information was gathered through a RP/SC/attitudinal survey applied to nearly 500 people that travel to work in the Santiago Metropolitan Area. The RP data considers nine alternative modes and up to 11 time periods, and the level-of-service data were obtained at unusual precision levels using GPS measurements. Some relevant results are that the travel time, cost and cost divided by the wage rate coefficients were fairly similar in both the RP and SC environments, yielding equal error variances for both datasets. The only parameters that differed between each type of data were those associated with the schedule delay early (SDE) and late (SDL) variables required by Small’s scheduling model (Small, 1982). This was attributed to the different temporal perspectives between the RP choices (long term decisions) and the SC decisions (short term), and also due to the context presented in the SC experiment (implementation of a congestion charging policy and a flexible working hours scheme). On the other hand, it was found that the best way to introduce scheduling flexibility into the models was by interacting the schedule delay terms (SDE and SDL) with dummy variables for the scheduling flexibility of each individual. It was also shown that when using time periods with a higher resolution (i.e. 15 min instead of 30 and 60 min intervals), the model goodness of fit decreased and the values of time increased. Finally, from a simple exercise of forecasting the impacts of a hypothetic congestion charging scheme, it was found that the schedule delay coefficients derived from the SC context produce a smoother and less-peaked temporal distribution of travel demand than the RP parameters.