Multi-State Travel Time Reliability Models with Skewed Component Distributions

• Activity Number: 234
• Type: Topic Contributed
• Date/Time: Monday, July 30, 2012 : 2:00 PM to 3:50 PM
• Sponsor: Section on Government Statistics
• Author(s): Feng Guo*+ and Qing Li and Hesham Rakha
• Companies: Virginia Tech and Virginia Tech and Virginia Tech
• Keywords: Travel time reliability ; Multi-state model ; skewness

Abstract

The multi-state travel time reliability model has demonstrated superior performance, close relationship with underlying traffic conditions, and ease of interpretation for travel time reliability reporting. This paper advances the multi-state model by using skewed component distributions, e.g., the gamma and lognormal distributions, to accommodate non-symmetrically distributed travel times, which are commonly observed in congested states. Six alternative models, the single state normal, gamma, lognormal distributions, and their multi-state versions were fitted to field collected data. The performance of the models was compared using the Akaike's information criterion. The results indicate that the multi-state lognormal model consistently outperforms alternative models. The advantage of the multi-state lognormal model is most substantial during peak hours. The improved fitting of the lognormal model is mainly reflected in the mode and tail portion of the data distribution. During non-peak hours, the single-state model could provide a compatible but parsimonious alternative to multi-state models. We demonstrated the impacts of using the multi-state lognormal model on travel time relia