| This study develops the methods of predicting frequency of truck-involved and non-truck-involved crashes on roadway segments and identifies unique characteristics of truck-involved crashes. To capture these nonlinear effects of the variables and temporal correlations among annual crash frequencies, Generalized Estimating Equation (GEE) models with nonlinearizing link functions were developed. Separate GEE models for total, truck-involved and non-truck-involved crashes were developed and compared. The result of the models shows that annual total and non-truck-involved crash frequencies in two successive years at a given location are correlated but the correlation does not exist for truck-involved crashes. The result also shows that nonlinearizing link functions of lane width, truck percentage and speed limit were statistically significant in the truck-involved crash. Thus, the proposed method can capture important nonlinear effects of variables on crash frequencies with temporal correlations, and identify the differences in the factors contributing to crash frequency between truck-involved and non-truck-involved crashes. |
