Simultaneous equation modeling for crash rate of freeway segments.

摘要

In 2010, the total number of reported traffic crashes in the state of Nevada was 51,664 of which 235 (0.6%) resulted in one or more deaths (Nevada Department of Transportation, 2012). The state's "Zero Fatalities" traffic safety campaign aims to reduce the rate even further. Out of the total number of crashes, Clark County (includes Las Vegas) accounted for 78.89%. This study examines safety improvement by developing advanced crash prediction models. The system of crash prediction equations consider geometric conditions and traffic volume using simultaneous equation modeling (SEM). The models are based on geometric characteristics and traffic volume data collected from Las Vegas freeway systems related to crash data provided by Nevada Department of Transportation (NDOT). All data characterizes the year 2010, chosen for the least amount of observed roadway construction zones when compared to other years.;The system of crash rate prediction equations represents connected freeway segment types. The types, defined by entrance (EN) and exit (EX) ramp-pair combinations, are estimated simultaneously instead of developing separate linear regression models. By modeling EX-EN segments connected to EN-EX using SEM, the relationship of crash rate in the EN-EX effects crash rate in EX-EN. Most EN-EX segments are considered weaving sections (lengths shorter than 2,500 feet) contributing to congestion.;The increase of significant model parameters is apparent when comparing SEM to single equation multiple linear regression. The additional information obtained confirms the correlation between crash rate prediction residuals exists between connecting EX-EN and EN-EX segment types and supports the existence of unobserved variables. SEM method of "three-stage least squares" modeling allows for freeway segments containing different characteristics to be modeled together, i.e., presence of auxiliary lane in EN-EX can be modeled with EX-EN segments. Instrumental variables replace the missing auxiliary lane variable in EX-EN segments resulting in a system of regression equations for crash prediction. The models can be used for connecting paired segments of EX-EN and downstream EN-EX or a connecting three-segment semi-corridor of EN-EX, downstream EX-EN and downstream EN-EX.