The modeling of individuals' departure time choices has gained increasingly interest over the past few decades. Understanding drivers' preferences for different departure times not only has important implications for policy evaluation (e.g., road pricing, peak spreading), but the resulting parameters obtained from the discrete choice model can also be used as key inputs for traffic demand models. There is a long tradition of estimating departure time choice models using SP data, due to the scarcity of good quality RP data.  In the more recent literature there are an increasing number of studies that combine both types of data. While these studies focused more on the econometric issues of joint SP-RP estimation, less attention has been paid on the data issue, particularly the quality control of the RP observations and the comparability of choice situations between SP and RP data.

In the current study, we used the data collected from Dutch ‘Spitsmijden' project, which involved rewarding commuters for avoiding car trips in the peak hours. The RP data was well validated by using different technological means of data collection, and the choice scenarios of the SP survey were specifically designed for the comparison purpose. Furthermore, the information provided from GPS data was incorporated to enrich the RP observations. We show that this is crucial since estimation results from door-to-door travel time data differ from estimations based on only a part of the trip. The dataset gives us numerous opportunities for investigating many key features in joint SP-RP estimations. In the present paper we will report a series of joint SP-RP mixed logit models. In particular, the issues of scale heterogeneity between SP and RP, unobserved heterogeneity in the taste parameters, and the panel structure of the SP and RP dataset will be taken into account in the model specification.