Made available in DSpace on 2015-04-22T22:16:04Z (GMT). No. of bitstreams: 1 Nelson Lima de Souza Filho.pdf: 1556771 bytes, checksum: 33a38464a9de0ec3dca0da75c9c6b64e (MD5) Previous issue date: 2012-12-06 / CAPES - Coordenação de Aperfeiçoamento de Pessoal de Nível Superior / In regression models, the classical normal assumption for the distribution of the measurement errors is often violated, masking some important features of the variability of the data. Some practical actions to overcome this problem, like transformations of the data, sometimes are not effective. In this work we propose a methodology to overcome this problem, in the context of multivariate linear regression with measurement errors. In these models, the covariate is unobservable and the researcher observes a surrogate variable. These measurements are made with an additive error. We extend the classical normal model, by modeling jointly the covariate and the measurement errors by a finite mixture of densities which are in a general family, accommodating skewness, heavy tails and multi-modality at the same time, allowing a degree of flexibility that can not be met by the normal model. We proceed Bayesian inference through a Gibbs-type algorithm. Some proposed models are compared with existing symmetrical models, using a modified DIC criterion, through the analysis of simulated and real data. / Em modelos de regressão, o pressuposto clássico de normalidade para a distribuição dos erros aleatórios é muitas vezes violado, mascarando algumas características importantes da variabilidade dos dados. Algumas ações práticas para resolver esse problema, como transformações nos dados, revelam-se muitas vezes ineficazes. Neste trabalho apresentamos uma proposta para lidar com esta questão no contexto do modelo de regressão multivariada linear simples, quando a variável resposta e a variável regressora são observadas com erro aditivo o chamado modelo de regressão linear com erros nas variáveis. Em tais modelos, o pesquisador observa uma variável substituta em vez da covariável de interesse. Nós estendemos o modelo clássico normal, modelando a distribuição conjunta da covariável e dos erros aleatórios por uma mistura finita de densidades pertencentes a uma família de distribuições bem geral, acomodando ao mesmo tempo assimetria, caudas pesadas e multimodalidade, permitindo um grau de flexibilidade que não pode ser atingido pelo modelo normal. Para a parte de estimação desenvolvemos um algoritmo do tipo Gibbs para proceder estimação Bayesiana. Alguns modelos propostos foram comparados com modelos simétricos já existentes na literatura, utilizando um critério DIC modificado, através da análise de dados simulados e reais.

Submitted by (ana.araujo@ufrpe.br) on 2016-07-07T12:44:03Z No. of bitstreams: 1 Josimar Mendes de Vasconcelos.pdf: 4725831 bytes, checksum: 716f4b6bc6100003772271db252915b7 (MD5) / Made available in DSpace on 2016-07-07T12:44:03Z (GMT). No. of bitstreams: 1 Josimar Mendes de Vasconcelos.pdf: 4725831 bytes, checksum: 716f4b6bc6100003772271db252915b7 (MD5) Previous issue date: 2011-08-08 / Conselho Nacional de Pesquisa e Desenvolvimento Científico e Tecnológico - CNPq / The last years has increased the quantity of researchers and search scientific in the plantation, production and value of the soybeans in the Brazil, in grain. In front of this, the present dissertation looks for to analyze the data and estimate models that explain, of satisfactory form, the variability observed of the quantity produced and value of the production of soya in grain in the Brazil, in the field of the study. For the development of these analyses is used the classical and Bayesian inference, in the context of simultaneous equations by the tools of indirect square minimum in two practices. In the classical inference uses the estimator of square minima in two practices. In the Bayesian inference worked the method of Mountain Carlo via Chain of Markov with the algorithms of Gibbs and Metropolis-Hastings by means of the technician of simultaneous equations. In the study, consider the variable area harvested, quantity produced, value of the production and gross inner product, in which it adjusted the model with the variable answer quantity produced and afterwards the another variable answer value of the production for finally do the corrections and obtain the final result, in the classical and Bayesian method. Through of the detours normalized, statistics of the proof-t, criteria of information Akaike and Schwarz normalized stands out the good application of the method of Mountain Carlo via Chain of Markov by the algorithm of Gibbs, also is an efficient method in the modelado and of easy implementation in the statistical softwares R & WinBUGS, as they already exist smart libraries to compile the method. Therefore, it suggests work the method of Mountain Carlo via chain of Markov through the method of Gibbs to estimate the production of soya in grain. / Nos últimos anos tem aumentado a quantidade de pesquisadores e pesquisas científicas na plantação, produção e valor de soja no Brasil, em grão. Diante disso, a presente dissertação busca analisar os dados e ajustar modelos que expliquem, de forma satisfatória, a variabilidade observada da quantidade produzida e valor da produção de soja em grão no Brasil, no campo do estudo. Para o desenvolvimento dessas análises é utilizada a inferência clássica e bayesiana, no contexto de equações simultâneas através da ferramenta de mínimos quadrados em dois estágios. Na inferência clássica utiliza-se o estimador de mínimos quadrados em dois estágios. Na inferência bayesiana trabalhou-se o método de Monte Carlo via Cadeia de Markov com os algoritmos de Gibbs e Metropolis-Hastings por meio da técnica de equações simultâneas. No estudo, consideram-se as variáveis área colhida, quantidade produzida, valor da produção e produto interno bruto, no qual ajustou-se o modelo com a variável resposta quantidade produzida e depois a variável resposta valor da produção para finalmente fazer as correções e obter o resultado final, no método clássico e bayesiano. Através, dos desvios padrão, estatística do teste-t, critérios de informação Akaike e Schwarz normalizados destaca-se a boa aplicação do método de Monte Carlo via Cadeia de Markov pelo algoritmo de Gibbs, também é um método eficiente na modelagem e de fácil implementação nos softwares estatísticos R & WinBUGS, pois já existem bibliotecas prontas para compilar o método. Portanto, sugere-se trabalhar o método de Monte Carlo via cadeia de Markov através do método de Gibbs para estimar a produção de soja em grão, no Brasil.

La modélisation par équations structurelles est très répandue dans des domaines très variés et nous l'appliquons pour la première fois en métrologie dans le traitement de données de comparaisons interlaboratoires. Les modèles à équations structurelles à variables latentes sont des modèles multivariés utilisés pour modéliser des relations de causalité entre des variables observées (les données). Le modèle s'applique dans le cas où les données peuvent être regroupées dans des blocs disjoints où chaque bloc définit un concept modélisé par une variable latente. La structure de corrélation des variables observées est ainsi résumée dans la structure de corrélation des variables latentes. Nous proposons une approche bayésienne des modèles à équations structurelles centrée sur l'analyse de la matrice de corrélation des variables latentes. Nous appliquons une expansion paramétrique à la matrice de corrélation des variables latentes afin de surmonter l'indétermination de l'échelle des variables latentes et d'améliorer la convergence de l'algorithme de Gibbs utilisé. La puissance de l'approche structurelle nous permet de proposer une modélisation riche et flexible des biais de mesure qui vient enrichir le calcul de la valeur de consensus et de son incertitude associée dans un cadre entièrement bayésien. Sous certaines hypothèses l'approche permet de manière innovante de calculer les contributions des variables de biais au biais des laboratoires. Plus généralement nous proposons un cadre bayésien pour l'amélioration de la qualité des mesures. Nous illustrons et montrons l'intérêt d'une modélisation structurelle des biais de mesure sur des comparaisons interlaboratoires en environnement. / Structural equation modelling is a widespread approach in a variety of domains and is first applied here to interlaboratory comparisons in metrology. Structural Equation Models with latent variables (SEM) are multivariate models used to model causality relationships in observed variables (the data). It is assumed that data can be grouped into separate blocks each describing a latent concept modelled by a latent variable. The correlation structure of the observed variables is transferred into the correlation structure of the latent variables. A Bayesian approach of SEM is proposed based on the analysis of the correlation matrix of latent variables using parameter expansion to overcome identifiability issues and improving the convergence of the Gibbs sampler. SEM is used as a powerful and flexible tool to model measurement bias with the aim of improving the reliability of the consensus value and its associated uncertainty in a fully Bayesian framework. The approach also allows to compute the contributions of the observed variables to the bias of the laboratories, under additional hypotheses. More generally a global Bayesian framework is proposed to improve the quality of measurements. The approach is illustrated on the structural equation modelling of measurement bias in interlaboratory comparisons in environment.

La modélisation par équations structurelles est très répandue dans des domaines très variés et nous l'appliquons pour la première fois en métrologie dans le traitement de données de comparaisons interlaboratoires. Les modèles à équations structurelles à variables latentes sont des modèles multivariés utilisés pour modéliser des relations de causalité entre des variables observées (les données). Le modèle s'applique dans le cas où les données peuvent être regroupées dans des blocs disjoints où chaque bloc définit un concept modélisé par une variable latente. La structure de corrélation des variables observées est ainsi résumée dans la structure de corrélation des variables latentes. Nous proposons une approche bayésienne des modèles à équations structurelles centrée sur l'analyse de la matrice de corrélation des variables latentes. Nous appliquons une expansion paramétrique à la matrice de corrélation des variables latentes afin de surmonter l'indétermination de l'échelle des variables latentes et d'améliorer la convergence de l'algorithme de Gibbs utilisé. La puissance de l'approche structurelle nous permet de proposer une modélisation riche et flexible des biais de mesure qui vient enrichir le calcul de la valeur de consensus et de son incertitude associée dans un cadre entièrement bayésien. Sous certaines hypothèses l'approche permet de manière innovante de calculer les contributions des variables de biais au biais des laboratoires. Plus généralement nous proposons un cadre bayésien pour l'amélioration de la qualité des mesures. Nous illustrons et montrons l'intérêt d'une modélisation structurelle des biais de mesure sur des comparaisons interlaboratoires en environnement. / Structural equation modelling is a widespread approach in a variety of domains and is first applied here to interlaboratory comparisons in metrology. Structural Equation Models with latent variables (SEM) are multivariate models used to model causality relationships in observed variables (the data). It is assumed that data can be grouped into separate blocks each describing a latent concept modelled by a latent variable. The correlation structure of the observed variables is transferred into the correlation structure of the latent variables. A Bayesian approach of SEM is proposed based on the analysis of the correlation matrix of latent variables using parameter expansion to overcome identifiability issues and improving the convergence of the Gibbs sampler. SEM is used as a powerful and flexible tool to model measurement bias with the aim of improving the reliability of the consensus value and its associated uncertainty in a fully Bayesian framework. The approach also allows to compute the contributions of the observed variables to the bias of the laboratories, under additional hypotheses. More generally a global Bayesian framework is proposed to improve the quality of measurements. The approach is illustrated on the structural equation modelling of measurement bias in interlaboratory comparisons in environment.