Limited dependent variable and structural equations models: empirical applications to traffic operations and safety

Abstract

This dissertation presents empirical applications of econometric and statistical methodologies to traffic operations and safety. The motivation of this research was to develop methodological frameworks to investigate factors affecting rural freeway safety and operations in an intelligent transportation system (ITS) setting. The research effort focuses on methodological frameworks for evaluating pre-ITS conditions but is applicable to post-ITS settings as well.Methodologies used in this dissertation include limited dependent variable models such as Poisson and negative binomial regressions as well as nested logit structures, and structural models involving simultaneous equations. The Poisson and negative binomial models investigate accident likelihoods on roadway sections, while the nested logit structure examines the conditional likelihood of accident severity. Simultaneous equations models examine factors affecting the cross-sectional endogeneity between lane mean speeds and lane-speed deviations. Accounting for cross-sectional endogeneity captures traffic flow dynamics which critically affect safety. These techniques are applied to an empirical setting where ITS infrastructure is being installed by the Washington State Department of Transportation (WSDOT) on a 61-kilometer section of rural Interstate 90 (I-90) located some 50 kilometers east of Seattle. Variable message signs at critical roadway locations coupled with in-vehicle signing in a selected number of vehicles will be provided to inform travelers of adverse driving conditions. The study area includes the Snoqualmie Pass summit, and experiences significant climatic interactions coupled with challenging roadway geometrics.Findings from this dissertation provide significant insights into the complex interactions and contemporaneous effects of spatial, temporal, environmental, geometric and traffic flow factors affecting accident causality and cross-sectional traffic flow dynamics. The approaches embodied in this dissertation, while being local in model specification, have broader implications beyond ITS, encompassing critical regional and national infrastructure design, programming and investment issues relating to traffic safety and operations. They afford greater flexibility in decision making through enhancement of the design strategy identification and definition process.