Date of Award


Degree Name

Master of Science in Engineering


Civil and Construction Engineering

First Advisor

Dr. Valerian Kwigizile

Second Advisor

Dr. Jun-Seok Oh

Third Advisor

Dr. Ron Van Houten

Access Setting

Masters Thesis-Open Access


This study aimed at improving the methodology for developing statewide non-motorized safety performance functions (SPFs). Due to lack of pedestrian and bicyclist counts, the methodology proposed a procedure for developing statewide surrogate non-motorized exposure measures using data that are available at statewide level. Eleven years non-motorized crashes at signalized urban intersections joining Arterial and Collector roads in Michigan were used to test the procedure.

The study also explored the use of Bayesian approach for modeling non-motorized crashes as an alternative to traditional classical count data models. Classical count data models that were considered as potential fit to the data include; Poisson Regression (PRM), Negative Binomial Regression (NBRM), Zero-Inflated Poisson (ZIP) and Zero-Inflated Negative Binomial (ZINB). NBRM was selected as the best classical count data model after thorough comparison between competing models using appropriate goodness of fit tests. For Bayesian approach, Poisson likelihood with gamma distribution prior was used for model estimation. The results showed that the Bayesian Poisson-gamma model outperforming classical NBRM model in terms of model estimation and out-of-sample prediction, especially with a relative small sample size.