Effect of Akaike Information Criterion on Model Selection in Analyzing Auto-crash Variables

Count data has become widely available in many disciplines. The mostly used distribution for modeling count data is the Poisson distribution (Horim and Levy; 1981) which assume equidispersion (Variance is equal to the mean). Since observed count data often exhibit over or under dispersion, the Poisson model becomes less ideal for modeling. To deal with a wide range of dispersion levels, Generalized Poisson regression, Poisson regression, and lately Conway-Maxwell-Poisson (COM-Poisson) regression can be used as alternative regression models. We compared the Generalized Poisson regression, Poisson Regression Model and Conway- Maxwell- Poisson. Data on road traffic crashes from the Anambra State Command of the Federal Road Safety Commission (FRSC), Nigeria were analyzed using these three methods, the results from the three methods are compared using the Akaike Information Criterion (AIC) with Poisson showing an AIC value of 2325.8 and GPR having an AIC value of 896.0278 and COM-Poisson showing an AIC value of 951.01. The GPR was considered a better model when analyzing road traffic crashes in Anambra State, Nigeria