Accident analysis with the random parameters negative binomial panel count data model

We analyzed factors affecting the frequency of accident counts in 81 cities over a three-year period (2008-2010) with monthly data using random-parameters negative binomial panel count data models. Among models considered, the random parameters model with the correlated coefficients outperformed the other models and was found to fit the data best with almost a perfect prediction of the conditional mean level. Most of variables used to control for the variation in the frequency of accident counts play crucial role with higher significance levels. Also, marginal effects and elasticity estimates were derived to get more insight into the effects of percent and unit changes in the dependent variable in response to changes in the exogenous variables.