Study On Seneral Issues In Zero-Inflated Model

In the relevant field of applied science research,discrete data sometimes shows a bigger variability. When the variance of variable is far larger than its mean value, the over-dispersion phenomenon exists, especially when there is a great deal of zeros in the sample. If neglecting this kind of phenomenon, it will cause the mistake of parameter estimate to make a false reasoning and forecast.As is stated above, the foreign scholars proposed several models to deal with the problem, for example Zero-Inflated Poisson model, Zero-Inflated Negative Binomial model and Zero-Inflated General Poisson model. But in the practice, there is a certain deviation on parameter estimate when modeling with small samples using these models. Based on the current situation of traditional models, this paper investigates Zero-Inflated model with small samples which are from Beta-Binomial distribution and contrast the new model with the above three models under the simulation experiment. The results show the new model has better estimations on parameters and zero probability than other three models under the situation of small samples and lower over-dispersion. At the same time, considered the complexity of some practical problems, time-space effect is imported into the modeling of Zero-Inflated model. The Zero-Inflated model with time-space effect and the detailed MCMC implementation procedure are described in detail in the paper. And the analysis of the example shows the Zero-Inflated poisson model with time-space effect has more accurate estimation than Zero-Inflated poisson model on the forcast of zero probability which illuminates the feasibility and validity of the Zero-Inflated model with time-space effect.