Aspectos computacionais para inferência na distribuição gama generalizada

• Ano de defesa: 2014
• Autor(a) principal: Ramos, Pedro Luiz [UNESP]
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Estadual Paulista (Unesp)
• Palavras-chave em Português: Computação - Matematica; Funções gama; Teoria bayesiana de decisão estatistica; Computer science - Mathematics
• Link de acesso: http://hdl.handle.net/11449/108639

The most important goals of this research project are related to the study of computational aspects to get inferences for the parameters of the Generalized Gamma distribution. The Generalized Gamma distribution is a good option to get better inferences for lifetime data, but usually in applications, the maximum likelihood estimators (MLE) of the parameters of the distribution could be very unstable depending on appropriated initial values in the numerical iterative procedure, especially for small sample sizes. Under a Bayesian approach, we observe that the posterior distributions could depend heavily on the choice of a prior distribution. In this work, using the classical inference will be shown a way to simplify the likelihood equations, minimizing the problem of instability of the method. Under a Bayesian approach, we will study the effect of different empirical and non-informative prior distributions on the posterior estimates, especially for small sample sizes. Numerical methods are also proposed in this work, in order to get good initial values used to improve the obtained inferences, considering censored data (random censoring) under the Classical and Bayesian approach. Simulated and real data sets will be used to illustrate the proposed methodology.