Unobserved Heterogeneity and the Effects of Age And Gender on Highway Accident Injury-Severities: A Dynamic Correlated Random Parameters Ordered Probit Analysis
MetadataShow full item record
Shah, Aafiya. M.S. in Transportation Systems Engineering, University at Buffalo, May 2017. Unobserved Heterogeneity and the Effects of Age and Gender on Highway Accident Injury-Severities: A Dynamic Correlated Random Parameters Ordered Probit Analysis. Major Professor: Dr. Panagiotis Ch. Anastasopoulos. Traditional accident injury-severity analysis has demonstrated the benefits of random parameters modeling in ordered probability modeling settings, in terms of accounting for unobserved heterogeneity. A common limitation of past work is that the random parameters are restricted to be uncorrelated from each other, suggesting that the unobserved factors captured by a random parameter are uncorrelated from those captured by other random parameters. This Thesis extends the existing body of literature by exploring the correlation of unobserved factors (varying systematically across the observations) in the statistical modeling mechanism of accident injury-severities with the correlated random parameters ordered probit model. Because driving behavior that can lead to a non-injury, injury, serious injury or fatal injury accident is a complex process that can be affected by a number of factors such as gender and age, this Thesis explores the differences in accident-injury severity among male and female drivers, and among different age groups. In addition, this Thesis seeks to explore time-varying (dynamic) factors that can affect accident injury-severities, the impact of which has not been thoroughly investigated in the literature. Using highway accident data collected between 2011 and 2013 in the State of Washington, fixed parameters and correlated and uncorrelated random parameters ordered probit models of accident injury-severities are estimated by gender and age groups-including stationary (non-time-varying) and dynamic (time-varying) explanatory variables-with the results suggesting the significant potential in accounting for the dynamic elements and the correlation in unobserved heterogeneity, in terms of statistical fit.