The Application of Count Regression Models on Traffic Accidents in Case of Addis Ababa, Ethiopia
Abstract
Road traffic accident is the major phenomena of the world as well as our country, Ethiopia which is from the low-income countries. Statistical modeling for count response variables is a primary interest in, insurance, and other areas. The main objective of this study is used to identify the most appropriate count regression model to fit the number of human deaths per road traffic accident (RTAs). The data for this study get from Addis Ababa Traffic Control, and Investigation Department (AATCID), daily basis recorded from July 30, 2013 to July 29, 2014. The difficultyassumption of Poisson regression model shifts to look for extended models like the negative binomial model, zero inflatedPoisson, and zero-Inflated Negative binomial regression models. Specifically, traffic accidents generate count response variables with an invalid assumption of Poisson distribution such thatthe variance and mean of human death per road traffic accident are (0.58) and (0.36), and the overdispersion parameter under the negative binomial was detected to indicate the existence of over-dispersion implies ZIP (or ZINB) model is favored over the Poisson (or NB) model, respectively, by using Vuong test. By using the goodness of fit model criteria like LRT, AIC, and BIC zero-inflated Poisson (ZIP) is the most fitted model for road traffic accident dataset.Therefore, quarter of year, slope of road, age of driver, vehicle type and ownership, time of accident, and type of accident are found statistically significant factors at α = 0.05fornumber of human death per road accidents.
Keywords:
Negative binomial regression, Traffic accidents, Zero inflated poisson regression, Zero inflated negative regressionDownloads
Published
How to Cite
Issue
Section
License
Copyright (c) 2020 AJST
This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.