An analysis of carsharing vehicle choice and utilization patterns using multiple discrete-continuous extreme value (MDCEV) models

• Published in: Transportation research. Part A, Policy and practice Policy and practice, 2017-09, Vol.103, p.362-376
• Main Authors: Jian, Sisi, Rashidi, Taha Hossein, Dixit, Vinayak
• Format: Article
• Language: English
• Subjects: Carsharing; Forecast; MDCEV; Vehicle choice and utilization
• ISSN: 0965-8564; 1879-2375
• DOI: 10.1016/j.tra.2017.06.012

•MDCEV models to allocate continuous budget to multiple carsharing vehicle types.•Demographics attributes influenced users’ vehicle utilization patterns.•Travel time, mileage and expenditure affect utilization in the same way.•The method can be used to decide most attractive vehicle fleet in carsharing systems. Facing the growing demand for carsharing services, it is critical for operators to accurately predict users’ preferences on different vehicle types and their vehicle usage. This vehicle choice behavior involves choosing multiple vehicle types simultaneously and allocating continuous amounts of budget to the chosen vehicles. The recent developed multiple discrete-continuous extreme value (MDCEV) modeling framework provides a suitable platform for allocation of continuous amounts of a consumer good (expenditure) to different discrete outcomes (different vehicle types). In this study, we develop three MDCEV models considering travel time, mileage, and monetary expenditure as the continuous consumption constraints. The three models estimate the impacts of a set of socio-demographic attributes on user’s vehicle choice and capture the satiation effect with increasing the consumption for each vehicle type. The study also employs an efficient simulation procedure to obtain the simulated results of the three models, and compare the results to the observed data using normalized RMSE and correct ratio to determine the best-fitted model. The estimation results suggest that user age, income level, driving license country, insurance plan, membership plan, and origin location have impacts on users’ vehicle utilization patterns. The comparison results indicate that travel time, mileage and expenditure affect users’ vehicle utilization patterns in the same way, and all three models can provide accurate predictions for the vehicle choice behavior. These findings can be referred to by operators when determining the most efficient allocation of resources within carsharing systems.