Influência local em modelos geoestatísticos T-Student com aplicações a dados agrícolas

• Autor(a) principal: Assumpção, Rosangela Aparecida Botinha
• Data de Publicação: 2010
• Tipo de documento: Tese
• Idioma: por
• Título da fonte: Biblioteca Digital de Teses e Dissertações do UNIOESTE
• Texto Completo: http://tede.unioeste.br:8080/tede/handle/tede/374
• topic: Geoestatística; Algoritmo EM; Máxima verossimilhança; Geostatistics; EM Algorithm; Maximum Likelihood; CNPQ::CIENCIAS AGRARIAS::ENGENHARIA AGRICOLA

The presence of inconsistent observations make it improper to consider the gaussian process, as it is found in the literature. This process should be replaced by models of the symmetric distribution classes, such as the t-student distribution, which incorporates additional parameters to reduce the influence of inconsistent points. This work has developed the EM algorithm for estimating the structure of the spatial dependence of the parameters and of the spatial linear model, assuming that the process shows t-student n-varied distribution. This distribution has the degree of freedom v as the additional parameter, which has been considered to be fixed in this research. Techniques to diagnose influence are used after the estimation of parameters, in order to assess the quality of the adjustment of the model by the assumptions made and for the robustness of the results of the estimates when there are disturbances in the model or data. In the present work, diagnostic techniques for the assessment of local influence in linear spatial models have been developed, considering the process with t-student n-varied distribution. The usual diagnostic technique evaluates the withdrawing of the likelihood rate by the function of the likelihood logarithm. In this proposal, in addition to considering the usual technique, we use the withdrawing of the likelihood by Q-displacement of the complete likelihood. The application of the usual technique and of the one proposed here are illustrated through the analyses of both simulated and real data, provenient of agricultural experiments.