User Preferences for Shared Autonomous Vehicles Based on Latent-Class Logit Models

Shared autonomous vehicles (SAV), which integrates autonomous driving and shared economy technology, provide people with high-quality travel services. Socio-economic attributes, historical travel characteristics, and behavioral attitude characteristics of the respondents are studied, and a questionnaire of stated preferences for travel mode choice is designed by an orthogonal experiment, then 311 valid data are collected to study their behavior characteristics for choosing SAV. A latent class analysis is carried out to fully consider individual heterogeneity and to explore latent classes of users. Integrating the latent classes as the variables into discrete choice Logit models, latent class-Logit models are formulated to study user's preference for SAV. By combining a multinomial or mixed Logit model with the three latent classes discovered, the significant influencing factors for SAV user preferences are recognized out of 59 variables, including gender, travel mode, SAV user group category, waiting time, etc., calibrated by four reasonable models. Moreover, seven indices of goodness of fit are measured to evaluate the effectiveness of eight models such as multinomial Logit, mixed Logit, and latent class-Logit. The marginal utility analysis is used to investigate the impacts of the attributes of travel mode on SAV preferences. Study results show that the discrete choice Logit models with three latent classes have a higher capacity for explaining the relationship between dependent and independent variables. The three classes can be described as the impulsive and positive innovator, contradictory and conservative innovator, and rational and conservative user respectively. It is also found that the significant influencing factors obviously vary across different latent class groups; the category of SAV users group is a significant factor for all latent class groups, and the significance level of SAV innovators in each model is less than 0.1;the accuracies of the first and second categories predicted by the latent class-Logit model are 5.9%~28.3% and 5.4%~18.5% higher than those predicted by other Logit models respectively.It is also found that waiting time has the greatest impact on travelers' choice of SAV; and when the probability of choosing SAV is close to 0.5, slightly reducing the travel cost of SAV is most effective to attract travelers to use SAV, rather than private cars for travel.