In this paper we consider power distributions for modeling proportions or rates with zero and/or one inflation as an alternative to beta regression. The model considered is a mixture between a Bernoulli type process for explaining the zero and/or one excess and a truncated power-normal model for explaining the continuous response. The maximum likelihood approach is considered for parameter estimation. Observed and expected information matrices are derived, illustrating interesting aspects of the likelihood approach. Given the flexibility of the power-normal distribution, we are able to show, in a practical scenario, the better performance of our proposal over the beta regression model.
Digital Object Identifier (DOI)
Mart´ınez-Fl´orez, Guillermo; Bolfarine, Heleno; and W. G´omez, H´ector
"Power-Models for Proportions with Zero/One Excess,"
Applied Mathematics & Information Sciences: Vol. 12
, Article 3.
Available at: https://dc.naturalspublishing.com/amis/vol12/iss2/3