On performance of parametric and distribution-free models for zero-inflated and over-dispersed count responses.

摘要: Zero-inflated Poisson (ZIP) and negative binomial (ZINB) models are widely used to model zero-inflated count responses. These extend the (NB) address excessive zeros in response. By adding a degenerate distribution centered at 0 interpreting it as describing non-risk group population, ZIP two-component population mixture. As applications of NB, key difference between ZINB is allowance for overdispersion by its NB component modeling response at-risk group. Overdispersion arising practice too often does not follow such data yield invalid inference. If sources known, other parametric may be directly overdispersion. Such subject assumed distributions. Further, this approach applicable if information about unavailable. In paper, we propose distribution-free alternative compare performance with these popular well moment-based proposed Yu et al. [Statistics Medicine 2013; 32: 2390-2405]. Like generalized estimating equations, requires no elaborate assumptions. Compared al., more robust overdispersed We illustrate our both simulated real study data.