Distribuição slash multivariada aplicada a dados agrícolas

• Ano de defesa: 2017
• Autor(a) principal: Fagundes, Regiane Slongo
• Orientador(a): Opazo, Miguel Angel Uribe
• Banca de defesa: Johann, Jerry Adriani , Assumpção, Rosangela Aparecida Botinha , De Bastiani, Fernanda
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Estadual do Oeste do Paraná; Cascavel
• Programa de Pós-Graduação: Programa de Pós-Graduação em Engenharia Agrícola
• Departamento: Centro de Ciências Exatas e Tecnológicas
• País: Brasil
• Palavras-chave em Português: Algoritmo EM; Diagnóstico de influência global e local; Modelagem espacial linear slash; Testes de hipóteses
• Palavras-chave em Inglês: EM algorithm; Global and local influence; Slash linear spatial model; Hypothesis tests
• Área do conhecimento CNPq: CIENCIAS AGRARIAS::ENGENHARIA AGRICOLA
• Link de acesso: http://tede.unioeste.br/handle/tede/3087

This study aimed at a discussing problems of multivariate statistical inference and linear spatial modeling when observations are from a continuous, symmetric population, with multivariate slash distribution. Firstly, a reparametrization of slash distribution was performed, assuming the existence of the finite second moment. Thus, some iterant properties were shown. Analytical expressions were tested for the score function and Fisher information matrix of reparameterized distribution. An approach to estimate some parameters by maximum likelihood was considered based at the EM (Expectation-Maximization) algorithm. Linear hypothesis tests have been described regarding the means vector and the covariance matrix using statistics such as C(α), likelihood ratio, Wald, and score. Studies of simulation were carried out to evaluate the efficiency of the statistical tests and EM algorithm. Data related to the agricultural area illustrated the methodology developed, and the hypothesis tests for equality of means, sphericity and equicorrelation were also applied. A slash linear spatial model, with and without the use of covariates, was proposed. Were Discussed the global and local influence diagnostic analysis in order to evaluate the influence of observations on the process of parameters’estimation. The curvatures required for the local influence procedure and based on the slash model were derived, in which the perturbation scheme has been chosen properly and related to the different perturbation schemes. Spatial variability maps of chemical attributes of soil and yield were generated by kriging with external drift. Finally results of simulations and applications indicated that the slash distribution is a robust alternative when the data present high kurtosis.