Analysis of panel count data model for Malaysian road accidents / Prof. Madya Dr Mohd Alias Lazim, Prof. Madya Dr Yap Bee Wah and Wan Fairos Wan Yaacob

The most common probability models for modeling count data are those based on the traditional Poisson and Negative Binomial assumption usually developed on individual basis. On the other hand, this research focused on modeling procedure using panel data analysis approach. The fixed-effects Poisson and Negative Binomial (FENB), random effects negative binomial (RENB) model and the cross-sectional negative binomial (NB) model were examined in order to take into account for heterogeneity in the accident data on a panel of 14 states in Malaysia covering the period of 1996 to 2007. We examined various factors associated with road accidents occurrence. It is hypothesized that the factors considered to affect road accidents are the monthly registered vehicle within the state, the amount of rainfall, the number of rainy day, time trend and the monthly seasonal effect. Various model specifications were estimated including the pooled Poisson, Fixed and Random Effects Poisson as well as Fixed and Random Effects Negative Binomial model. The results showed that road accident occurrences are positively associated with the increase in the number of registered vehicle, increase in the amount of rain and time of the occurrence. The effect of seasonality also indicates that accident occurrence is higher in the month of October, November and December. The models developed confirmed the factors identified that have effect on the number of road accidents in Malaysia. The specification comparisons also indicate the benefits gained from using the NB model with spatial and temporal effects. The RENB model was found to be more superior when incorporated temporal and cross sectional variations which offers advantages in model flexibility.