Name: Boris Fazio
Institution: PUCP, Perú


The endpoint-inflated binomial regression model proposed by Tian, Ma, Zhou and Deng (2015) provides a way of modeling bounded count data with a high proportion of observations at the endpoints. We extended the model by replacing the binomial component of the inflated distribution with a beta-binomial to allow handling of overdispersed data and propose the standard normal cumulative distribution function with fixed cutpoints as a link function for regression on the inflation probabilities.
We tested our model on the same dataset used by Tian et al. We found that our normal cdf link provides a better fit than the softmax and that a normal random intercept on a binomial component had a better fit than our beta-binomial extension.