Updating and Extending the Disaggregate Choice Models in the Dutch National Model

Updating and Extending the Disaggregate Choice Models in the Dutch National Model


J Willigers, M de Bok, Significance / FEUP, NL


This paper presents the re-estimation and improvements of the Netherlands National Model System (LMS). These include integration of logsums from subsequent choices and combined RP/SP estimations for the MD models.


The Netherlands National Model System (LMS) is known as one of the first disaggregate national travel demand forecasting systems used in practice. The model system has been in use since 1986, and has been extensively updated and extended through its lifetime. Disaggregate discrete choice models are applied in the various modules of the modelling system. These modules simulate the different choices in travel behaviour: tour frequencies (TF), mode and destination choice (MD), time of day choice (ToD), secondary and lower level destinations and the choice of a train route.

This paper presents outcomes from the current update and extension of the modelling system. The re-estimation results are presented for the tour frequency models, the mode/destination choice models and time-of-day choices. These modules are re-estimated on more recent National Travel Surveys. The re-estimations include some enhancements to the estimation of the previous model system.

The coherence of the different choices in the modelling system is improved by inclusion of logsums from subsequent choices in the estimation of the modules. The re-estimated TF models include logsums from the MD models to test the responsiveness of travellers to changes in transport alternatives due to a transport policy (latent demand). Despite the intuitive importance of accessibility on trip generation, the current version of the LMS lacks significant accessibility attributes in the TF module. In this update it is tried to solve this shortcoming by inclusion of purpose specific logsums from the MD models in the trip frequency models.

A further advancement is made by integrating the MD and car ToD models into a single model. This model is estimated by a simultaneous SP/RP estimation; the RP travel survey data of MD choices are enriched with SP data on time-of-day and mode choice. This makes different nesting structures possible: ToD choice can be higher, lower or on the same level as destination choice. Additionally, the socio-economic segmentation of the MD model is improved, and the non-motorised modes are split into walking and cycling. The paper discusses these enhancements and their consequences.

For train travel a new model of station choice and access/egress mode choice is developed. The logsum from this model is used in the MD choice utility function to improve and the way train level-of-service affects MD and TF choices and to make the model system as a whole more consistent. The system is now thus specified as a single decision tree that comprises train access/egress mode choice, station choice, time of day choice, destination choice, main mode choice and tour frequency choice.

The long history of the modelling system and multiple updating efforts, allow a discussion of the evolution of key indicators over time. Elasticities and Values-of-Time from the re-estimated model are discussed and compared with results of previous model versions.


Association for European Transport