Development of Accident Modification Factors for Rural Frontage Road Segments in Texas Using Generalized Additive Models

Abstract

The objective of this study consists of assessing the application of generalized additive models (GAMs) for estimating accident modification factors (AMFs). GAMs are a new type of models that have been recently introduced by the statistical community for modeling observed data. These models offer more flexible functional forms than traditional generalized linear models and allow for more adaptable variable interactions. As recently documented in the literature, variable interactions should be included in the development of AMFs. To accomplish the study objective, AMFs were derived from GAMs using data collected on rural frontage roads in Texas. The AMFs were then compared to the AMFs produced from a previous study using the same data set. The results of the study show that AMFs produced from GAMs are more flexible to characterize the safety effect of simultaneous changes in geometric and operational features (or variable interactions) than when independent AMFs are applied together. The results also show that GAMs indicated a nonlinear relationship between crash risk and changes in lane and shoulder widths for frontage roads in Texas.