This paper describes the new commercial vehicle (CV) model recently developed for the Triangle Region (Raleigh-Durham-Chapel Hill) in North Carolina. The new model is mainly based on the commercial vehicle travel survey data collected in the region in spring 2010. While the model overall follows the traditional 3-step process, a few new ideas were attempted in the course of model development and incorporated in the final model. Major differences from CV models used in other MPOs in the US include:

• Most of the major MPO CV models in the US are stratified by vehicle type only. Supported by the survey data, this model is further stratified by trip purpose and the following six sub-models were developed: light-duty vehicle delivering services, delivering goods, or with other purposes; single-unit truck delivering services, or delivering goods; and multi-unit truck delivering goods.

• Most of the major MPO CV models in the US use the gravity model for trip distribution. This model however uses a logit-based destination choice model structure, which allows for explicit inclusion of non-impedance variables, such as socio-economic, geographic, and political-boundary variables, and explicit estimation of their coefficients. Second to travel time, inter-area-type and county-crossing dummy variables were found statistically significant, providing extra strong explanatory power to the model.

This paper also describes an approach used to overcome multicollinearity in regression model estimation for CV trip generation. The approach groups TAZs into districts based on the similarity of socio-economic data in the TAZs, specifically, by ranking percentages of employees of different industry types in each TAZ. It reduced most of the correlation coefficients between any two explanatory variables used in the model below 0.1.

Also described in the paper is a data imputation approach for the CV trips that were reported but not recorded with any details in the survey instrument due to space constraints. Trip purposes, times of day, and trip lengths were imputed for those trips (which account for 20% of total trips surveyed). Imputed trips were then included in estimation and calibration of trip generation and distribution models and derivation of time-of-day factors.