Halldórsdóttir, Katrín3; Rieser-Schüssler, Nadine6; W. Axhausen, Kay6; Prato, Carlo Giacomo1; Nielsen, Otto Anker5
1 Department of Transport, Technical University of Denmark2 Traffic modelling and planning, Department of Transport, Technical University of Denmark3 Traffic Modelling, Department of Transport, Technical University of Denmark4 Swiss Federal Institute of Technology5 Department of Management Engineering, Technical University of Denmark6 Swiss Federal Institute of Technology
With a growing interest in sustainable transport systems, the interest has increased on encouraging more cycling. To encourage cycling, it is important to identify which network attributes influence cyclists route choice and evaluate the trade-offs among these attributes. To analyse travel behaviour, observed choices and alternatives composing the choice set of each cyclist are necessary. However, generating the alternative choice sets can prove challenging. This paper analyses the efficiency of various choice set generation methods for bicycle routes in order to contribute to our understanding of choice generation for highly detailed networks. There is a substantial amount of literature that studies cyclists’ route choices. Most studies have been based on stated preference (SP) data (see, e.g., [1,2]). Although SP data have a lot of benefits there are some disadvantages, e.g. the challenge to, without bias, predefine what cyclists consider when choosing a route. There have been few revealed preference (RP) studies reported in the literature (see, e.g., [1,2]). One disadvantage with RP data is that generating alternative routes can prove difficult. The benefit of collecting travelling information with GPS loggers, compared to self-reported RP data, is more accurate geographic locations and routes. Also, the GPS traces give more reliable information on times and prevent trip underreporting, and it is possible to collect information on many trips by the same person without reporting fatigue. GPS data require nevertheless extensive post-processing and in some cases mode imputation. They also require a very detailed digital network to map the routes accurately, which can lead to high computation times during choice set generation, as well as issues with behavioural realism that might produce inconsistent estimates. There have been some studies on bicycle route choice set generation reported in the literature, whereof few studies focussed on route choice models for bicyclist estimated from GPS observations. Menghini et al.  successfully applied a Breadth First Search on Link Elimination (BFS-LE) approach. Broach et al.  tested three different choice set generation methods, i.e. K-shortest paths, simulated shortest paths, and route labelling. None of these methods proved to be satisfactory and a modified route labelling method was proposed instead.
Main Research Area:
Latsis Symposium 2012 - 1st European Symposium on Quantitative Methods in Transportation Systems