Departure Time Modelling Applicability and Travel Time Uncertainty



Departure Time Modelling Applicability and Travel Time Uncertainty

Authors

M Börjesson, WSP. SE

Description

A mixed logit departure time choice model is estimated, combining SP and RP data. We test for correlation of scheduling sensitivity across RP and SP choices within individuals and for RP/SP scale differences. Travel times are assumed being uncertain.

Abstract

The full-scale experiment of introducing time of-day-varying tolls in Stockholm has focused the need to model drivers' trip timing decision.
In the paper we estimate a departure time choice model, combining stated preference (SP) and revealed preference (RP) data, to be used for a full-scale application. We explicitly account for response scale differences between RP and SP data. No joint RP and SP analysis of trip timing choice has previously been published. The analysis therefore provides a unique opportunity to compare observed and stated trip timing behaviour.

A mixed logit framework was applied, using Johnson's SB as mixing distribution, explicitly accounting for correlation in unobserved heterogeneity over repeated SP choices. This was fundamental for accurate estimation of the substitution pattern. We also explicitly test for correlation of scheduling sensitivity across RP and SP choices within individuals. Estimation indicates that response-scale is larger in the RP data, and that scheduling sensitivity differs between RP and SP choices within individuals, implying systematic differences in the RP and SP data. We discuss several causes for the differences.

The model takes into account that travellers face an uncertain travel time. Few previous studies have used RP data to analyse sensitivity to travel time uncertainty. No other RP study has done so in combination with departure time choice. Actual travel times for drivers taking part in the SP survey originated from the dynamic network assignment model CONTRAM. Traffic cameras measured travel time uncertainty.

Run time for the implemented model is 15 minutes for a network consisting of 100000 OD pairs and 14 time departure time periods, assuming 50 random draws and 14 segments in each zone pair.

Publisher

Association for European Transport