Abstract
It is expected that vehicle holding durations differ substantially across households. Some households may keep their vehicles as long as possible, while others may replace some of their vehicles after holding several years. Even though existing studies have incorporated such heterogeneity using duration models, one-peak distribution is usually assumed. In reality, this assumption can be easily violated. In this sense, the heterogeneity of duration distributions has not been satisfactorily represented in existing literature. To represent the heterogeneity in the household vehicle holding duration, the paper proposes to apply a latent class modeling approach to simultaneously integrate different duration distributions. Such modeling approach could deal with multi-peak distribution of vehicle holding duration. Using a data collected in several Japanese cities in 2006, this paper confirms the model effectiveness from both model performance and applicability. It is found that household characteristics are important factors to explain the latent classes.