Date of Award
Doctor of Philosophy (PhD)
Dr. Jennifer Ogle, Committee Chair
Dr. Wayne Sarasua
Dr. Mashrur Chowdhury
Dr. Patrick Gerard
Crash frequency has been identified by many experts as one of the most important safety measures, and the Highway Safety Manual (HSM) encompasses the most commonly accepted predictive models to predict the crash frequency for specific road segments and intersections. The HSM recommends that the models should be calibrated using data from a jurisdiction where the models will be applied. Large amounts of data collection and data analysis are required for this purpose, which presents numerous challenges. The HSM further recommends that, in case of available data and expertise, local agencies may develop jurisdiction-specific models. In this dissertation, the process of HSM calibration and developing state-specific models for the state of South Carolina is described. Also, the contributions to to the established knowledge of highway safety are explained. One of the most common start-up issues with the calibration process is how to estimate the required sample size to achieve a specific level of precision, which can be a function of the variance of the calibration factor. The published research has indicated great variance in sample size requirements, and some of the sample size requirements are so large that they may deter state departments of transportation from conducting calibration studies. In this study, an equation is derived to estimate the sample size based on the coefficient of variation of the calibration factor and the coefficient of variation of the observed crashes. This equation is verified using a regression analysis on a dataset from two recent calibration studies, South Carolina and North Carolina. Also, the bootstrap method is used to derive an unbiased estimate of the variance of the calibration factor in this study. Additionally, different definitions and criteria for the calibration factors are investigated. In addition to the calibration factors in the HSM and previously published definitions, two other calibration factor equations are proposed and compared using multiple goodness of fit measures. Whereas each definition may outperform others in certain measures, in this study, it is recommended to use the definition that maximizes the likelihood between predicted and observed crashes. Furthermore, HSM recommends that for large jurisdictions with a variety of topographical or climate conditions, it may be desirable to develop separate calibration factors for each specific terrain or geographical region. Whereas no further guidance is provided in the HSM, most of the previous research in this field has been focused on comparing a calibration factor that has been developed for a specific predefined region within the jurisdiction of interest with the state-wide calibration factor, to see if the use of that calibration factor is justified. This study aims to provide guidelines on how to define the regions within the jurisdiction of interest that might need separate calibration factors and/or separate crash prediction models. For this purpose, a network level regression is performed, using network level data (i.e. traffic volume and length), and spatial autocorrelation methods are used to find possible clusters of high values of studentized residuals or hot spots. It is argued that any statistically significant hot spot can be an indicator of a region that needs a separate calibration factor and/or prediction model. This method is demonstrated using South Carolina data and the results show that the use of region-specific calibration factors for hotspots identified by this method is more likely to be justified compared to area definitions based on topography or climate.
Rajabi, Mahdi, "Assessing the Predictability of Highway Safety Manual Crash Prediction Models in the State of South Carolina" (2017). All Dissertations. 2051.