Abstract by Timo Pew
Comparing Zero-Inflated Poisson and Negative Binomial-Lindley models to model intersection-related crashes
Roadway intersections are one of the most hazardous areas for drivers due to the many decisions, potential distractions, and maneuvers that are made in a relatively short amount of time. As such, the Utah Department of Transportation (UDOT) has attempted to model intersection-related crash counts to identify dangerous intersections in their efforts to reduce the number of crashes. Overdispersion and a high proportion of observations with zero crashes are two challenging data characteristics that often arise when modeling crash counts. There have been several methods developed in the recent years to model crash counts with these two data characteristics. We will compare the effectiveness of a Zero-Inflated Poisson model and Negative Binomial-Lindley model in a Bayesian framework using intersection-related crashes throughout the state of Utah from 2012-2016. We will examine how well each model fits the data using a Bayesian chi-square test for goodness-of-fit and the out of sample predictive power for the respective methods.