Endogenous selection models for railroad crash analysis

Railroad crossing crashes have been a covert but demanding safety concern recently due to their frequent occurrence and significant impact on lives and the economy. This research study aims to analyze this issue from the perspective of network science and statistical analysis. Railroad crossings data from Texas was used to conduct the analysis. Three research studies have been developed to address the complex railroad safety issues: a) the first study proposes an econometric model to measure the impact of safety devices and railroad crossing network connectivity factors on the eigencentrality of a crossing; b) the second study proposes spatial models of railroad safety treatments with endogenous regressors; and c) the third study proposes an econometric model of endogenous selection effects of safety treatments. The studies provide a testable empirical basis for new statistical methodologies for the comprehensive assessment of railroad safety outcomes such as number of individuals killed or injured in collisions between trains, vehicles, bicyclists and pedestrians. They show the importance of statistical considerations relating to unobserved heterogeneity, endogeneity and selection bias in railroad safety assessment. Prior safety research in railroad analysis has ignored these considerations, leading to biased and inconsistent results especially pertaining to the safety effectiveness of common crossing treatments such as warning bells, signs, gates and flashing signals. The dissertation shows promise for network level safety policy application through a comprehensive analysis of direct and indirect safety impacts of commonly used crossing treatments.