A Comparative Study of Several Models for Zero Inflation Counting Data

Objective To explore the comparison and application of several models to deal with zero inflation counting data.Methods Poisson,negative binomial,zero-inflated Poisson,zero-inflated negative binomial,Poisson hurdle,and negative binomial hurdle models were applied to fit the data in R program.While the likelihood ratio tests,Vuong statistics(for non-nested models),Akaike′s information criterion(AIC)and bayesian information criterion(BIC)were used to evaluate the goodness of fit.Results The zero-inflated negative binomial and the negative binomial hurdle models are better than other models,the negative binomial hurdle model fits the data more accurate.The fitting results showed that the longer stays in the hospital,more chronical conditions,higher education level and have private insurance means physician visit more frequently.Those who with good self-assessment and the male often have fewer visits to physician,which means fewer healthcare demand.Conclusion The zero-inflated negative binomial and negative binomial hurdle models are better in handling over-dispersion and zero-inflation.While in the case of relatively few zero observations,the negative binomial hurdle model may be more suitable.