Public transport route choice modelling: Reducing estimation bias when using smart card data

Abstract

Automated Fare Collection (AFC) data for public transport analyses has received much research interest recently, including its use for the estimation of passenger route choice preferences. However, an important problem persists since AFC data only includes information about the trip within the public transport system, that is stop-to-stop (tap-in to tap-out). Not knowing the full trip from door-to-door might lead to estimation bias, especially when estimating route choice models based on only the chosen stops, which is common practice in current research using AFC data. To avoid this, we propose an improved method for estimating route choice models in public transport using AFC data. The method is based on randomly generating pseudo origin (and destination) points in close vicinity of the actually chosen origin (and destination) stops, thus allowing pseudo access and egress times to be incorporated into the route choice model. The framework is compatible with any probability density function. We suggest using the Beta distribution for generating points when knowledge about access and egress distances are available, whereas the Uniform distribution is suggested when no knowledge is available. The method was applied on replicated AFC data based on traditional travel survey data from the Greater Copenhagen area in Denmark. The results of the model estimations confirm estimation bias in parameter estimates when not correcting for the lack of access/egress information. The proposed method notably improves in-vehicle-time parameter estimates of the route choice model compared to estimation assuming AFC stop-to-stop data, whereas access/egress time and hidden waiting time parameters are still biased, although to a lesser extent than a traditional naïve estimation based on stop-to-stop data.