Orientador: Hildete Prisco Pinheiro / Dissertação (mestrado) - Universidade Estadual de Campinas, Instituto de Matemática, Estatística e Computação / Made available in DSpace on 2018-08-24T09:34:36Z (GMT). No. of bitstreams: 1 Baldoni_PedroLuiz_M.pdf: 4328843 bytes, checksum: 0ab04f375988e62ac31097716ac0eaa5 (MD5) Previous issue date: 2014 / Resumo: Os Modelos Lineares Generalizados Mistos (MLGM) são uma generalização natural dos Modelos Lineares Mistos (MLM) e dos Modelos Lineares Generalizados (MLG). A classe dos MLGM estende a suposição de normalidade dos dados permitindo o uso de várias outras distribuições bem como acomoda a superdispersão frequentemente observada e também a correlação existente entre observações em estudos longitudiais ou com medidas repetidas. Entretanto, a teoria de verossimilhança para MLGM não é imediata uma vez que a função de verossimilhança marginal não possui forma fechada e envolve integrais de alta dimensão. Para solucionar este problema, diversas metodologias foram propostas na literatura, desde técnicas clássicas como quadraturas numéricas, por exemplo, até métodos sofisticados envolvendo algoritmo EM, métodos MCMC e quase-verossimilhança penalizada. Tais metodologias possuem vantagens e desvantagens que devem ser avaliadas em cada tipo de problema. Neste trabalho, o método de quase-verossimilhança penalizada (\cite{breslow1993approximate}) foi utilizado para modelar dados de ocorrência de doença em uma população de vacas leiteiras pois demonstrou ser robusto aos problemas encontrados na teoria de verossimilhança deste conjunto de dados. Além disto, os demais métodos não se mostram calculáveis frente à complexidade dos problemas existentes em genética quantitativa. Adicionalmente, estudos de simulação são apresentados para verificar a robustez de tal metodologia. A estabilidade dos estimadores e a teoria de robustez para este problema não estão completamente desenvolvidos na literatura / Abstract: Generalized Linear Mixed Models (GLMM) are a generalization of Linear Mixed Models (LMM) and of Generalized Linear Models (GLM). The class of models GLMM extends the normality assumption of the data and allows the use of several other probability distributions, for example, accommodating the over dispersion often observed and also the correlation among observations in longitudinal or repeated measures studies. However, the likelihood theory of the GLMM class is not straightforward since its likelihood function has not closed form and involves a high order dimensional integral. In order to solve this problem, several methodologies were proposed in the literature, from classical techniques as numerical quadrature¿s, for example, up to sophisticated methods involving EM algorithm, MCMC methods and penalized quasi-likelihood. These methods have advantages and disadvantages that must be evaluated in each problem. In this work, the penalized quasi-likelihood method (\cite{breslow1993approximate}) was used to model infection data in a population of dairy cattle because demonstrated to be robust in the problems faced in the likelihood theory of this data. Moreover, the other methods do not show to be treatable faced to the complexity existing in quantitative genetics. Additionally, simulation studies are presented in order to verify the robustness of this methodology. The stability of these estimators and the robust theory of this problem are not completely studied in the literature / Mestrado / Estatistica / Mestre em Estatística

abstract: Bivariate responses that comprise mixtures of binary and continuous variables are common in medical, engineering, and other scientific fields. There exist many works concerning the analysis of such mixed data. However, the research on optimal designs for this type of experiments is still scarce. The joint mixed responses model that is considered here involves a mixture of ordinary linear models for the continuous response and a generalized linear model for the binary response. Using the complete class approach, tighter upper bounds on the number of support points required for finding locally optimal designs are derived for the mixed responses models studied in this work. In the first part of this dissertation, a theoretical result was developed to facilitate the search of locally symmetric optimal designs for mixed responses models with one continuous covariate. Then, the study was extended to mixed responses models that include group effects. Two types of mixed responses models with group effects were investigated. The first type includes models having no common parameters across subject group, and the second type of models allows some common parameters (e.g., a common slope) across groups. In addition to complete class results, an efficient algorithm (PSO-FM) was proposed to search for the A- and D-optimal designs. Finally, the first-order mixed responses model is extended to a type of a quadratic mixed responses model with a quadratic polynomial predictor placed in its linear model. / Dissertation/Thesis / Doctoral Dissertation Statistics 2020

In 2018, Raul Matsushita, Donald Pianto, Bernardo B. De Andrade, Andre Can§ado & Sergio Da Silva published a paper titled âTouchard distributionâ, which presented a model that is a two-parameter extension of the Poisson distribution. This model has its normalizing constant related to the Touchard polynomials, hence the name of this model. This diploma thesis is concerned with the properties of the Touchard distribution for which delta is known. Two asymptotic tests based on two different statistics were carried out for comparison in a Touchard model with two independent samples, supported by simulations in R.

No description available.

Frequentist model averaging has started to grow in popularity, and it is considered a good alternative to model selection. It has recently been applied favourably to gen- eralized linear models, where it has mainly been purposed to aid the prediction of probabilities. The performance of averaging estimators has largely been compared to that of models selected using AIC or BIC, without much discussion of model screening. In this paper, we study the performance of model averaging in classification problems, and evaluate performances with reference to a single prediction model tuned using cross-validation. We discuss the concept of model screening and suggest two methods of constructing a candidate model set; averaging over the models that make up the LASSO regularization path, and the so called LASSO-GLM hybrid. By means of a Monte Carlo simulation study, we conclude that model averaging does not necessarily offer any improvement in classification rates. In terms of risk, however, we see that both methods of model screening are efficient, and their errors are more stable than those achieved by the cross-validated model of comparison.

Several models for predicting future customer profitability early into customer life-cycles in the property and casualty business are constructed and studied. The objective is to model risk at a customer level with input data available early into a private consumer’s lifespan. Two retained models, one using Generalized Linear Model another using a multilayer perceptron, a special form of Artificial Neural Network are evaluated using actual data. Numerical results show that differentiation on estimated future risk is most effective for customers with highest claim frequencies.

No description available.

No description available.

Graduate students in the Brigham Young University Chemistry Department are working to develop a filtering device that can be used to separate substances into their constituent parts. The device consists of a monomer and water mixture that is polymerized into a monolith inside of a capillary. The ideal monolith is completely solid with interconnected pores that are small enough to cause the constituent parts to pass through the capillary at different rates, effectively separating the substance. Although the end objective is to minimize pore sizes, it is necessary to first identify an experimental region where any combination of input variables will consistently yield homogeneous monoliths capable of flow. To accomplish this task, an experimental mixture design is used to model the relationship between the variables related to the creation of the monolith and the probability of creating an acceptable polymer. The results of the mixture design suggest that, inside of the constrained experimental region, mixtures with higher proportions of monomer and surfactant, low amounts of initiator and salt, and DEGDA as the monomer have the highest probability of producing a workable monolith. Confirmatory experiments are needed before future experimentation to minimize pore sizes is performed using the refined constrained experimental region determined by the results of this analysis.

The Design of an experiment involves selection of levels of one or more factor in order to optimize one or more criteria such as prediction variance or parameter variance criteria. Good experimental designs will have several desirable properties. Typically, one can not achieve all the ideal properties in a single design. Therefore, there are frequently several good designs and choosing among them involves tradeoffs. This dissertation contains three different components centered around the area of optimal design: developing a new graphical evaluation technique, discussing designs for non-regular regions for first order models with interaction for the two- and three-factor case, and using the standard designs in the case of generalized linear models (GLM). The Fraction of Design Space (FDS) technique is proposed as a new graphical evaluation technique that addresses good prediction. The new technique is comprised of two tools that give the researcher more detailed information by quantifying the fraction of design space where the scaled predicted variance is less than or equal to any pre-specified value. The FDS technique complements Variance Dispersion Graphs (VDGs) to give the researcher more insight about the design prediction capability. Several standard designs are studied with both methods: VDG and FDS. Many Standard designs are constructed for a factor space that is either a p-dimensional hypercube or hypersphere and any point inside or on the boundary of the shape is a candidate design point. However, some economic, or practical constraints may occur that restrict factor settings and result in an irregular experimental region. For the two- and three-factor case with one corner of the cuboidal design space excluded, three sensible alternative designs are proposed and compared. Properties of these designs and relative tradeoffs are discussed. Optimum experimental designs for GLM depend on the values of the unknown parameters. Several solutions to the dependency of the parameters of the optimality function were suggested in the literature. However, they are often unrealistic in practice. The behavior of the factorial designs, the well-known standard designs of the linear case, is studied for the GLM case. Conditions under which these designs have high G-efficiency are formulated. / Ph. D.