Modeling Crash Rates for a Mountainous Highway by Using Refined-Scale Panel Data

Abstract

Traditional traffic safety analyses on crash frequency or crash rate usually focus on highly aggregated cross-sectional data. The adoption of aggregated data ignores the time-varying nature of some critical factors, and their effects on traffic safety may be masked through data aggregation. For mountainous highways, weather- and environment-related variables become critical as a result of complex time- and spatial-varying characteristics as well as interactions with mountainous terrain. Therefore, refined-scale models are often desired to appropriately model crash safety risks on mountainous highways and disclose the inherent crash mechanism. An advanced random parameter Tobit model with panel data (time series cross-sectional data) in refined temporal scale was developed. This is so far the first reported effort on integrating random parameter Tobit model and refined-scale panel data to develop crash rate models. A random parameter model was adopted not only to handle unobserved heterogeneity explicitly, but also to account for serial correlations across observations in panel data. Refined-scale weather and traffic data in a panel formation were adopted to accommodate the varying nature of complex driving conditions. Interstate 70 (I-70) in Colorado is well known for its typical mountainous terrain, critical role for local and national traffic, and inclement weather. As a demonstration of the modeling techniques, crash rates for a segment of I-70 were investigated in refined temporal scale. Results showed that a random parameter Tobit model outperformed a fixed-parameter Tobit model, and factors related to traffic, weather, and surface conditions were found to play significant roles in an accident rate model.