Mapas da transmutação : modelagem, propriedades estruturais, estimação e aplicações

Abstract

Initially, we use the quadratic transmutation maps to compose a new probability model: the transmuted log-logistic distribution. Transmutation maps are a convenient way of constructing new distributions, in particular survival ones. It comprises the functional composition of the cumulative distribution function of one distribution with the inverse cumulative distribution (quantil) function of another. Its comprehensive description of properties, such as moments, quantiles, order statistics etc., along with its survival study and the classical and Bayesian estimation methods, are also part of this work. Focusing on analysis of survival, the study included two practical situations commonly found: the presence of regression variables, through the transmuted log-logistic regression model, and the presence of right censorship. In a second moment, searching for a more exible model than the transmuted, we present its generalization, the transmuted distributions of cubic rank. Using the methodology presented in this rst generalization, two models were considered to compose the new cubic transmuted distributions: the log-logistic and Weibull models. Faced with problems presented in the transmutated class of quadratic and cubic orders (such as the restricted parametric space of the transmutation parameter ), we propose in this work, a new family of distribution. This family, which we call e-transmuted or e-extended, is as simple as the transmuted model, because it includes a single parameter to the base model, but more exible than the class of transmuted models, once the transmuted is a particular case of the proposed family. In addition, the nem family presents important properties such as, orthogonality between the baseline model parameters and the e-transmutation parameter, along with unrestricted parametric space for the ! e-transmutation parameter, which is de ned on the real line. Simulation studies and real data applications were performed for all proposed models and generalizations.