This dissertation has multiple aims in studying hierarchical linear models in biomedical data analysis. In Chapter 1, the novel idea of studying the durations of adolescent growth spurts as a predictor of adulthood obesity is defined, established, and illustrated. The concept of contextual effects modeling is introduced in this first section as we study secular trend of adulthood obesity and how this trend is mitigated by the durations of individual adolescent growth spurts and the secular average length of adolescent growth spurts. It is found that individuals with longer periods of fast height growth in adolescence are more prone to having favorable BMI profiles in adulthood. In Chapter 2 we study the estimation of contextual effects in a hierarchical generalized linear model (HGLM). We simulate data and study the effects using the higher level group sample mean as the estimate for the true mean versus using an Empirical Bayes (EB) approach (Shin and Raudenbush 2010). We study this comparison for logistic, probit, log-linear, ordinal and nominal regression models. We find that in general the EB estimate lends a parameter estimate much closer to the true value, except for cases with very small variability in the upper level, where it is a more complicated situation and there is likely no need for contextual effects analysis. In Chapter 3 the HGLM studies are made clearer with large-scale simulations. These large scale simulations are shown for logistic regression and probit regression models for binary outcome data. With repetition we are able to establish coverage percentages of the confidence intervals of the true contextual effect. Coverage percentages show the percentage of simulations that have confidence intervals containing the true parameter values. Results confirm observations from the preliminary simulations in the previous section of this paper, and an accompanying example of adulthood hypertension shows how these results can be used in an application.

In the presented thesis we deal with the generalized linear models framework in a claims reserving problem. Claims reserving in non-life insurance is firstly described and the considered class of models is introduced. Consequently, this branch of stochastic modelling is implemented in the reserving setup. For computation of the risk associated with claims reserving, we need a predictive distribution of future liabilities in order to evaluate risk measures such as Va- lue at Risk and Conditional Value at Risk. Since datasets in non-life insurance commonly consist of a small number of observations and estimation of predictive distributions can be complicated, we adopt a bootstrap method for this purpose. Model fitting, simulations and consequent measuring of the reserving risk are performed within the use of real-life data. Based on this, an analysis of fitted models and their comparison together with graphical outputs is included. 1

1 Abstract: The present work investigates techniques of insurence ratemaking accor- ding to the claims counts of policyholders on the basis of information contained in policies. At the beginning, we provide a closer examination of the theory of genera- lized linear models, which have wide range of applications in the field of actuarial modeling. The second chapter presents the basic Poisson regression model as well as some particular verification methods. Specifically, deviance and Wald test could be found here and furthermore also important results for residuals. The third chapter contains information on alternative approaches to modeling the claim frequencies and at the end the GEE method, that can be applied in case of panel data, is de- scribed. The numerical study based on real insurace data in last part of this diploma thesis illustrate's previously described techniques which were obtained with the help of statistical software SAS.

Recursive partitioning algorithms separate a feature space into a set of disjoint rectangles. Then, usually, a constant in every partition is fitted. While this is a simple and intuitive approach, it may still lack interpretability as to how a specific relationship between dependent and independent variables may look. Or it may be that a certain model is assumed or of interest and there is a number of candidate variables that may non-linearly give rise to different model parameter values. We present an approach that combines generalized linear models with recursive partitioning that offers enhanced interpretability of classical trees as well as providing an explorative way to assess a candidate variable's in uence on a parametric model. This method conducts recursive partitioning of a generalized linear model by (1) fitting the model to the data set, (2) testing for parameter instability over a set of partitioning variables, (3) splitting the data set with respect to the variable associated with the highest instability. The outcome is a tree where each terminal node is associated with a generalized linear model. We will show the method's versatility and suitability to gain additional insight into the relationship of dependent and independent variables by two examples, modelling voting behaviour and a failure model for debt amortization, and compare it to alternative approaches.

[pt] Os modelos lineares generalizados auto regressivos com médias móveis (do inglês GARMA), possibilitam a modelagem de séries temporais de dados de contagem com estrutura de correlação similares aos dos modelos ARMA. Neste trabalho é desenvolvida uma extensão multivariada do modelo GARMA, considerando a especificação de um modelo Poisson bivariado a partir da distribuição de Kocherlakota e Kocherlakota (1992), a qual será denominada de modelo Poisson BGARMA. O modelo proposto é adequado para séries de contagens estacionárias, sendo possível, através de funções de ligação apropriadas, introduzir deterministicamente o efeito de sazonalidade e de tendência. A investigação das propriedades usuais dos estimadores de máxima verossimilhança (viés, eficiência e distribuição) foi realizada através de simulações de Monte Carlo. Com o objetivo de comparar o desempenho e a aderência do modelo proposto, este foi aplicado a dois pares de séries reais bivariadas de dados de contagem. O primeiro par de séries apresenta as contagens mensais de óbitos neonatais para duas faixas de dias de vida. O segundo par de séries refere-se a contagens de acidentes de automóveis diários em dois períodos: vespertino e noturno. Os resultados do modelo proposto, quando comparados com aqueles obtidos através do ajuste de um modelo Gaussiano bivariado Vector Autoregressive (VAR), indicam que o modelo Poisson BGARMA é capaz de capturar de forma adequada as variações de pares de séries de dados de contagem e de realizar previsões com erros aceitáveis, além de produzir previsões probabilísticas para as séries. / [en] Generalized autoregressive linear models with moving average (GARMA) allow the modeling of discrete time series with correlation structure similar to those of ARMA’s models. In this work we developed an extension of a univariate Poisson GARMA model by considerating the specification of a bivariate Poisson model through the distribution presented on Kocherlakota and Kocherlakota (1992), which will be called Poisson BGARMA model. The proposed model not only is suitable for stationary discrete series, but also allows us to take into consideration the effect of seasonality and trend. The investigation of the usual properties of the maximum likelihood estimators (bias, efficiency and distribution) was performed using Monte Carlo simulations. Aiming to compare the performance and compliance of the proposed model, it was applied to two pairs of series of bivariate count data. The first pair is the monthly counts of neonatal deaths to two lanes of days. The second pair refers to counts of daily car accidents in two distinct periods: afternoon and evening. The results of our model when compared with those obtained by fitting a bivariate Vector Autoregressive Gaussian model (VAR) indicates that the Poisson BGARMA model is able to proper capture the variability of bivariate vectors of real time series of count data, producing forecasts with acceptable errors and allowing one to obtain probability forecasts.

Os modelos lineares generalizados duplos (MLGD), diferentemente dos modelos lineares generalizados (MLG), permitem o ajuste do parâmetro de dispersão da variável resposta em função de variáveis preditoras, aperfeiçoando a forma de modelar fenômenos. Desse modo, os mesmos são uma possível solução quando a suposição de que o parâmetro de dispersão constante não é razoável e a variável resposta tem distribuição que pertence à família exponencial. Considerando nosso interesse em seleção de variáveis nesta classe de modelos, estudamos o esquema de seleção de variáveis em dois passos proposto por Bayer e Cribari-Neto (2015) e, com base neste método, desenvolvemos um esquema para seleção de variáveis em até k passos. Para verificar a performance do nosso procedimento, realizamos estudos de simulação de Monte Carlo em MLGD. Os resultados obtidos indicam que o nosso procedimento para seleção de variáveis apresenta, em geral, performance semelhante ou superior à das demais metodologias estudadas sem necessitar de um grande custo computacional. Também avaliamos o esquema para seleção de variáveis em até \"k\" passos em um conjunto de dados reais e o comparamos com diferentes métodos de regressão. Os resultados mostraram que o nosso procedimento pode ser também uma boa alternativa quando possui-se interesse em realizar previsões. / The double generalized linear models (DGLM), unlike the generalized linear model (GLM), allow the fit of the dispersion parameter of the response variable as a function of predictor variables, improving the way of modeling phenomena. Thus, they are a possible solution when the assumption that the constant dispersion parameter is unreasonable and the response variable has distribution belonging to the exponential family. Considering our interest in variable selection in this class of models, we studied the two-step variable selection scheme proposed by Bayer and Cribari-Neto (2015) and, based on this method, we developed a scheme to select variables in up to k steps. To check the performance of our procedure, we performed Monte Carlo simulation studies in DGLM. The results indicate that our procedure for variable selection presents, in general, similar or superior performance than the other studied methods without requiring a large computational cost. We also evaluated the scheme to select variables in up to \"k\" steps in a set of real data and compared it with different regression methods. The results showed that our procedure can also be a good alternative when the interest is in making predictions.

[pt] Melhorias da mortalidade vêm sendo observadas em praticamente todo o mundo desde o início do século XX e impactam diretamente o resultado dos cálculos atuariais. A incorporação das tendências futuras da mortalidade no cálculo atuarial é possível através do uso de tábuas de mortalidade geracionais, que fornecem probabilidades de morte baseadas não só na idade x do indivíduo, como também no tempo t. O estudo aborda técnicas para projeção da mortalidade e consequente determinação dos fatores de improvement, utilizados para tornar uma tábua de mortalidade na forma geracional. As metodologias Lee-Carter e modelos lineares generalizados são utilizadas para construir previsões de mortalidade com base na experiência de mortalidade da população da Inglaterra e País de Gales da última metado do século passado. / [en] By the beginning of the 20th century, improvement on mortality started rising in many countries and this has a direct impact on the results of actuarial calculus. The trend of mortality can be incorporated into actuarial calculus through the use of generation mortality tables, that consider not only the age x of the individual but also the time t. This study explores techniques to project the mortality and the improvement factors used to turn a mortality table into a generational one. The methodologies of Lee-Carter and generalized linear models were used to forecast mortality by using the England and Wales mortality experience of the past half century.

[pt] O tema central deste trabalho é apresentar modelos de persistência. As probabilidades de persistência na carteira de um produto de determinada empresa de seguros e previdência serão estudadas de forma agregada, de tal forma que se torna possível a elaboração de outros estudos, como por exemplo, de análise de lucratividade, mesmo com poucos dados, o que inviabiliza a elaboração de tábuas de múltiplos decrementos. Serão avaliadas as possíveis causas de saídas de acordo com as características do plano. O desenvolvimento dos modelos tomam por base dados em forma de triângulo, técnica normalmente utilizada para cálculo de provisões de seguros. / [en] The objective of this work is to present persistency models. The probabilities of remaining in a Insurance and Pension company portfolio will be studied in a aggregate way, in this way it is possible to develop another results such as profitability, even though, there is poor data, what turns impossible to build multiple decrement tables. The possible lapses causes will be evaluated according to the plan. The models development is based on triangular data, this technique is usual on claims reserving.

As distribuições de probabilidade gama, normal inversa, Weibull, log-normal e exponencial são uma boa alternativa para modelar observações associadas ao tempo, pois, em geral, a variável tempo possui assimetria à esquerda ou à direita, o que caracteriza as distribuições citadas anteriormente. O objetivo deste trabalho constitui-se em avaliar o comportamento dos predadores, Euborellia annulipes (\"Tesourinha\") e Harmonia axyridis (\"Joaninha\"), em relação à praga conhecida como Aphis gossypii (\"Pulgão\"). Outra pretensão deste trabalho é a aplicação da modelagem estatística, dando ênfase as técnicas dos modelos lineares generalizados e análise de sobrevivência, as quais foram aplicadas aos dados provenientes de um experimento, instalado no Laboratório de Ecologia de Insetos da Escola Superior de Agricultura \"Luiz de Queiroz\" (ESALQ). O experimento foi composto por 21 repetições, sendo cada repetição efetuada por meio de uma placa de Petri medido 60 X 15 mm. Em cada placa foi liberado um pulgão adulto áptero na parte central, tendo três pesquisadores responsáveis por observar a varável definida como tempo de ataque. Inicialmente, foram ajustados os modelos com distribuição gama e diferentes funções de ligação, e o modelo com a distribuição normal inversa com função de ligação canônica. Esses modelos foram ajustados aos dados desconsiderando as censuras, em que por meio do gráfico half-normal plot e testes de hipóteses, verificou que o modelo com a distribuição normal inversa com função de ligação canônica, apresentou o melhor ajuste. Posteriormente, foram ajustados os modelos exponencial, Weibull e log-normal para os dados considerando as censuras, os quais foram avaliados mediante o teste de razão de verossimilhança, sendo o modelo log-normal mais apropriado aos dados. / The probability density function of gamma, inverse normal, Weibull, log-normal and exponential distributions are good alternatives for modelling observations related with time, since, in general, the time variable has left or right asymmetry, which characterizes the distributions previously mentioned . The aim of this work is the application of statistical modeling, emphasizing the techniques of generalized linear models and survival analysis, which were applied to data from an experiment, installed in the Laboratory of Insect Ecology of the \"Luiz de Queiroz\" College of Agriculture (ESALQ), in which the goal of this experiment was to evaluate the behavior of predators, Euborellia annulipes (\"ring-legged earwig\") and Harmonia axyridis (\"Ladybird\"), in relation to the pest known as Aphis. The experiment was composed of 21 replicates, each replicate being done by means of a petri dish measured 60 X 15 mm. On each plate an adult aphid was released in the central part, with three researchers responsible. The model with distribution was used to determine the variance, which was defined as the attack time. Normal distribution with canonical link function. These models were adjusted to the data disregarding censorship, in which through the half-normal plot and hypothesis tests, verified that the model with the normal inverse distribution with canonical link function, presented the best fit. Subsequently, the exponential, Weibull and log-normal models were adjusted for the data considering the censorship, which were evaluated by the likelihood ratio test, the log-normal model being more appropriate to the data.

Neste trabalho, trˆes modelos foram ajustados a um conjunto de dados obtido de um ensaio de controle biol´ogico para Diatraea saccharalis, uma praga comum em planta¸c˜oes de cana-de-a¸c´ucar. Usando a distribui¸c˜ao binomial como modelo de probabilidade, um ajuste adequado n˜ao pode ser obtido, devido `a superdispers˜ao gerada pela variabililidade dos dados e pelo excesso de zeros. Nesse caso, o modelo binomial inflacionado de zeros (ZIB) superdisperso ´e mais flex´ývel e eficiente para a modelagem desse tipo de dados. Entretanto, quando o interesse maior est´a sobre os valores positivos das propor¸c˜oes, pode-se utilizar o modelo binomial truncado superdisperso. Uma abordagem alternativa eficiente que foi utilizada para a modelagem desse tipo de dados foi a Bayesiana, sendo o ajuste do modelo realizado usando as t´ecnicas de simula¸c˜ao Monte Carlo em Cadeias de Markov, atrav´es do algoritmo Metropolis-Hastings e a sele¸c˜ao dos modelos foi feita usando o DIC (Deviance Information Criterion) e o fator de Bayes. Os modelos foram implementados no procedimento IML (Iteractive Matrix Linear) do programa SAS (Statistical Analysis System) e no programa WinBUGS e a convergˆencia das estimativas foi verificada atrav´es da an´alise gr´afica dos valores gerados e usando os diagn´osticos de Raftery & Lewis e de Heidelberger & Welch, implementado no m´odulo CODA do programa R. / In general the standard binomial regression models do not fit well to proportion data from biological control assays, manly when there is excess of zeros and overdispersion. In this work a zero-inflated binomial model is applied to a data set obtained from a biological control assay for Diatraea saccharalis, a commom pest in sugar cane. A parasite (Trichogramma galloi) was put to parasitize 128 eggs of the Anagasta kuehniella, an economically suitable alternative host (Parra, 1997), with a variable number of female parasites (2, 4, 8,..., 128), each with 10 replicates in a completely randomized experiment. When interest is only in the positive proportion data, a model can be based on the truncated binomial distribution. A Bayesian procedure was formulated using a simulation technique (Metropolis Hastings) for estimation of the posterior parameters of interest. The convergence of the Markov Chain generated was monitored by visualization of the trace plot and using Raftery & Lewis and Heidelberg & Welch diagnostics presented in the module CODA of the software R.