Literature on modelling route choice behaviour of cyclists has been rapidly growing. This can be attributed both to the increasing availability of GPS data, and the widely spread availability of easy-to-use software enabling model estimation using established methods. While this trend can be assessed positively, as it allows modelling cyclists’ behaviour in different regional contexts and without detailed expert knowledge of the estimation procedures, it has also led to certain methodological negligence. One of the procedures where the quality has been compromised is the choice set generation, which constitutes the (immediate) preceding step in the traditional path-based route choice models.

Although cognitive processes leading to the composition of the choice set are well-understood, the knowledge about the decision process in the context of cyclists’ route choice has been both very limited and not fully regarded, with many researchers focusing on the feasibility and performance of the algorithms, rather than their behavioural foundation. The structure of the choice set is known to heavily bias model estimations; however, there is no consensus among researchers on the size and composition of the choice set. Therefore, contradicting approaches are being employed in research and accepted in the published literature by reviewing peers. If we further consider that the results of bicycle route choice models and the related simulations are translated to policy implications, the importance of reflecting real-life behaviour (realistic route alternatives for assessment) is apparent.

Within this conference contribution, we intend to present our solution to the aforementioned challenges. Moreover, we would like to discuss the requirements that a choice set (in particular its size and composition) should fulfil, and a means to achieve it, so that the resulting route choice models can be considered behaviourally realistic.