The Zero Inflated Conway-Maxwell-Poisson Generalized Linear Model for Analyzing Motor Vehicle Crash Data

• Activity Number: 220 - Contributed Poster Presentations: Transportation Statistics Interest Group
• Type: Contributed
• Date/Time: Tuesday, August 4, 2020 : 10:00 AM to 2:00 PM
• Sponsor: Transportation Statistics Interest Group
• Author(s): Xiaomeng Li* and Dipak K Dey
• Companies: and University of Connecticut
• Keywords: COM-Poisson distribution; Zero-inflated models; Bayesian inference; model comparison

Abstract

This paper introduces Conway Maxwell Poisson (COM-Poisson) distribution to analyze vehicle crash data. The advantage of COM-Poisson is its ability to handle both under- and over-dispersion through controlling one special parameter in the distribution, which make it more flexible than current frequently used models, i.e., Poisson and Negative Binomial. Considering excess zeros frequently observed in crash data, we apply the zero inflated COM-Poisson distribution to analyze the traffic crashes, and compare the result with zero inflated Poisson and zero inflated negative binomial models. To add more flexibility, we also apply the generalized extreme value (GEV) model, which can capture different skewness with a shape parameter, as the link function for predicting the proportion of excess zeros. All models are fit under Bayesian framework and model comparison are based on mean absolute error, prediction error, Deviance Information Criterion (DIC) and Log-pseudo marginal likelihood (LPML).