Preferences for automated taxis. A comparison between immersive virtual reality and screen-based stated choice experiments

Automated Taxi (AT) services, as a promising business model for Autonomous Vehicles (AVs), have paved the way for novel mobility options. Understanding users’ preferences for this innovative alternative is crucial for its success but proves to be challenging, as the lack of a real market mostly requires setting hypothetical situations to elicit respondents’ preferences. Recent applications have seen an increasing use of Virtual Reality (VR), as a way to control the context, or to provide visual representation of some attributes. Research is however still in its infancy, in particular in the transport context, and results often are not comparable or show opposite effects. In this research, we aim to contribute to this limited research by studying the impact of the immersive VR environment in the preferences for AT elicited with Stated Choice (SC) experiments. Differently from previous studies, our experiment is built to ensure that the SC experiment in the VR-based environment is perfectly comparable with the standard SC screen-based, for internal validity. A control group is also used to test the order effect in the presentation of the two surveys, and to control for the carryover effect. Using data collected from a quite large sample, compared to the existing studies, joint mixed logit models are estimated allowing to assess the impact of the immersive VR environment on the choice bias, heterogeneity in the preferences for specific attributes, as well as panel effect, and order effect. Our results show that the immersive VR-based experience has no impact on the preferences for level-of-service attributes, travel time and cost, unless respondents can, to some extent, experience them, like waiting time in our study. On the other hand, the hints and cues provided by the immersive VR environment seem to affect the evaluation of the social aspects, descriptive norms and customers’ reviews. This highlights a significantly high heterogeneity in the preferences, not revealed in the screen-based SC experiment. Finally, differently from previous literature, our results show no difference in the choice bias between VR- and screen-based environment, but only if differences in the preferences for specific attributes between the two environments are properly taken into account.