Estimação do tamanho populacional a partir de um modelo de captura-recaptura com heterogeneidade

• Ano de defesa: 2014
• Autor(a) principal: Pezzott, George Lucas Moraes
• Orientador(a): Salasar, Luis Ernesto Bueno
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de São Carlos
• Programa de Pós-Graduação: Programa de Pós-Graduação em Estatística - PPGEs
• País: BR
• Palavras-chave em Português: Inferência (Estatística); Processo sequencial de captura-recaptura; Modelo de mistura; Em-algoritmo; Gibbs sampling; Erros de software
• Palavras-chave em Inglês: Capture-recapture model; Mixture model; Software errors; EM algorithm; Gibbs sampling algorithm
• Área do conhecimento CNPq: CIENCIAS EXATAS E DA TERRA::PROBABILIDADE E ESTATISTICA::ESTATISTICA
• Link de acesso: https://repositorio.ufscar.br/handle/20.500.14289/4585

In this work, we consider the estimation of the number of errors in a software from a closed population. The process of estimating the population size is based on the capture-recapture method which consists of examining the software, in parallel, by a number of reviewers. The probabilistic model adopted accommodates situations in which reviewers are independent and homogeneous (equally efficient), and each error is an element that is part of a disjoint partition in relation to its detection probability. We propose an iterative process to obtain maximum likelihood estimates in which the EM algorithm is used to the nuisance parameters estimation. The estimates of population parameters were also obtained under the Bayesian approach, in which Monte Carlo on Markov Chains (MCMC) simulations through Gibbs sampling algorithm with insertion of latent variables were used on the conditional posterior distributions. The two approaches were applied to simulated data and in two real data sets from the literature.