Simulation on Poisson and negative binomial models of count road accident modeling

Accident count data have often been shown to have overdispersion. On the other hand, the data might contain zero count (excess zeros). The simulation study was conducted to create a scenarios which an accident happen in T-junction with the assumption the dependent variables of generated data follows certain distribution namely Poisson and negative binomial distribution with different sample size of n=30 to n=500. The study objective was accomplished by fitting Poisson regression, negative binomial regression and Hurdle negative binomial model to the simulated data. The model validation was compared and the simulation result shows for each different sample size, not all model fit the data nicely even though the data generated from its own distribution especially when the sample size is larger. Furthermore, the larger sample size indicates that more zeros accident count in the dataset.