Explore the scholarly record.

Abstract

Safety is considered one of the most important factors for autonomous vehicles (AVs) but surprisingly its role in the adoption of AVs has received relatively less attention. In this paper we investigated the role of perceived safety in the choice of automated taxis (ATs) versus normal taxis (NTs), focusing on both possible vehicle failures (perceived vehicle safety - PVS) and personal harmful/property loss events (perceived personal safety - PPS). We investigated the possible correlation between these safety perceptions and their relationship with trust (T) in their impacts to the choice of ATs. We tested these effects in two different settings, Newcastle, UK and Toronto, Canada to explore which factors are context dependent and which are invariant and transferable. Data were collected in Newcastle and Toronto using the same stated choice experiment customised to account for geographic differences. Hybrid choice models were estimated to capture the effects of these three latent psychological factors in the choices of ATs. We evaluated the models’ statistical fits and their performances in terms of predictive ability using the average over 50 hold-out samples. Results showed that PVS and PPS have a significantly positive effects on T, which in turn has a significant positive effect on the choice of ATs, however the predictive ability of the HCMs is weaker than the statistical results would suggest. A careful analysis suggested that at aggregate level, a more parsimonious mixed logit model could be preferable, but if the focus is on specific groups of the population, then the HCMs with a hierarchical latent variable structure do perform better. Interestingly, while different groups of the population were relevant in Newcastle and in Toronto, the overall model evaluation is consistent between cities.