Access/egress travel to train stations poses a significant barrier to increasing the number of train travellers. The last mile is challenging for travellers, given the lack of private modes to reach the destination, strongly limiting the egress range from the station. An often-cited solution is shared micromobility (SMM): bicycles, e-bikes, e-scooters and e-mopeds. Through a stated preference survey, we analyse activity-end mode-choice preferences for SMM, walking and public transport (PT) among the Dutch population. Using a latent class choice model, we uncover three user groups: Multimodal SMM enthusiast (58%), who choose based on the trade-offs between various travel characteristics, while not having strong modal preferences. They are the most open, ready and able to use SMM. SMM hesitant cyclists (16%) have a strong preference for cycling and while they are open to using SMM, they may not feel themselves ready, stating that use of SMM can be difficult and dangerous. SMM-averse PT users (27%) are most likely to use PT and avoid SMM as they find it too difficult and dangerous to use. For policymakers, the high preference to walking over short egress distances reaffirms the need for continued focus on transit-oriented development. For longer distances, policymakers should focus on improving PT service in high-density high-demand areas, as high frequencies and dense PT networks can be justified, while stations in low-demand areas are better served by SMM. Policymakers should also prioritise SMM modes that are cheaper and that travellers are familiar and comfortable with, such as bicycles.

This research combines two relatively new additions to the field of discrete choice modelling: sequential best worst discrete choice experiments (SBWDCE) and random regret minimisation (RRM) modelling, with the hope of developing a more behaviourally realistic choice model. SBWDCEs are able to gather a larger number of stated choice observations from fewer respondents, while RRM models challenge the notion of fully compensatory behaviour implied by the traditional RUM model and suggest that consumers choose to minimise regret. According to image theory, best and worst choices are not made with the same kind of decision rule, so accounting for that variability using a RRM model would produce a more realistic model with better model fit. Estimating the combined model proves that people do in fact use a compensatory decision rule when selecting the best alternatives and a semi- to non-compensatory decision rule when selecting the worst. The results also show that the way choice set size variation is accounted for can greatly impact the scale parameters, as these and the choice set size constants are inversely related. Although a better model fit was achieved, using best-worst tasks is contested and researchers also warn against the lower reliability of additional choices in the same choice set. Nevertheless, SBWDCEs provide great benefits in fields with small population sizes and can potentially help in obtaining higher quality prior parameter values for use in efficient experimental design generation.