The zero & $N$-inflated binomial distribution with applications

Abstract: In this article we consider the distribution arising when two zero-inflated Poisson count processes are constrained by their sum total, resulting in a novel zero & $N$-inflated binomial distribution. This result motivates a general class of model for applications in which a sum-constrained count response is subject to multiple sources of heterogeneity, principally an excess of zeroes and $N$'s in the underlying count generating process. Two examples from the ecological regression literature are used to illustrate the wide applicability of the proposed model, and serve to detail its substantial superiority in modelling performance as compared to competing models. We also present an extension to the modelling framework for more complex cases, considering a gender study dataset which is overdispersed relative to the new likelihood, and conclude the article with the description of a general framework for a zero & $N$-inflated multinomial distribution.