Random Coefficient Models for Work Zone Headway Distribution

Abstract

Vehicle headways in work zones are disaggregated into two types in this study—truck-involved (e.g., car-truck) and nontruck-involved (e.g., car-car) headways. Considering the possible effects of vehicle-following patterns and unobserved heterogeneity, we have developed a multivariate distribution model with random coefficients for each headway type using the work zone headway data from Singapore. Five headway distribution types, including the lognormal, normal, gamma, log-logistic, and Weibull, are considered. The lognormal distribution is found to be the best pattern for both headway types. Results show that, for any given distribution pattern, the distribution model with random coefficients could provide better goodness-of-fit than the model with fixed coefficients. It is further found that there is a bigger effect of work intensity on the truck-involved headways than the nontruck-involved headways. In addition, both types of headways decrease as the traffic flow or truck percentage increases. The likelihood ratio test results confirm the necessity of building a separate distribution model for each headway type in work zones.