A modified Binomial likelihood model for zero/n-inflated count data
James S. Sweeney, John Haslett · arXiv (Cornell University) · 2014
A statistical inconsistency of a zero-inflated Binomial likelihood model for count data is identified. This issue occurs when the response, $y$, is equal to the sum constraint $n$, and results in statistically inconsistent and erroneous parameter inferences being drawn from the data. The zero-modified likelihood is amended to address this issue of \textit{n-inflation}, resulting in a fully symmetric Binomial likelihood model for both zero and n-inflated counts. An ecological regression problem is presented which details the superiority, through consistency of inference, of the new likelihood model.