Determining an efficient and precise choice set for public transport based on tracking data

Abstract

To understand the route choices of public transport users, it is important to know the information available to them, and the context present at that moment. In fact, each choice situation in a transport network has different characteristics and possibilities, also depending on the current status of the transport network. In this regard, travel diaries based on tracking technologies can capture precise observations for a long term. In this work, we exploit a large-scale tracking dataset, collected through a mode detection algorithm, to understand route choices of public transport users. We propose a choice set generation algorithm, able to cover more than 94% of the collected trips without any computational constraint. We compare the users’ paths in the public transport network with different choice sets, under multiple performance indicators, including coverage, size, and fit. This latter is computed by the estimation of a Path Size Logit model. The use of Automatic Vehicle Location (AVL) data allows comparing the available paths in terms of public transport vehicles used. We also consider different information provisions of network conditions and disturbances (full knowledge, no knowledge and current knowledge), and study which information provision best represents the choice set inferred by the observed users’ behaviour. Estimating a Mixed Path Size Logit model, we identified high heterogeneity among the users in only a few aspects. Overall, a condition of no knowledge results as the best fit, i.e. users seem to take into account in a minor way the realized delays in the alternatives considered when deciding their public transport route.