Full text
| |
| Author(s): |
João Victor Bastos de Freitas
Total Authors: 1
|
| Document type: | Master's Dissertation |
| Press: | Campinas, SP. |
| Institution: | Universidade Estadual de Campinas (UNICAMP). Instituto de Matemática, Estatística e Computação Científica |
| Defense date: | 2021-03-09 |
| Examining board members: |
Caio Lucidius Naberezny Azevedo;
Alexandre Galvão Patriota;
Filidor Edilfonso Vilca Labra
|
| Advisor: | Juvêncio Santos Nobre; Caio Lucidius Naberezny Azevedo |
| Abstract | |
In this dissertation, two classes of regression models for continuous, skewed and/or heavy tailed data were developed. One for independent data and another for dependent data. We considered a semi-parametric approach using Generalized Additive Partially Linear Models (GAPLM), for independent data, and GAPLM with Generalized Estimation Equations (GEE), for dependent data. In both cases, semi-parametric predictors for response means and scale mixtures of centered skew-normal (SMCSN) distributions for the (marginal) errors were considered. For dependent data, the dependence structures were modelled through GEE. Concerning the SMCSN distributions we considered either usual mixing measures (gamma, beta and binary distributions) as well as never used ones (generalized gamma, Birnbaum-Saunders and beta prime distributions). Estimation methods, goodness of model fit and diagnostic tools for these models, under the frequentist paradigm, were developed. Computational routines were created, to allow for the use of the developed methodologies, as well as simulation studies were performed to study the their performance. Also, the modelling of real problems, through such methodologies, were considered, illustrating the potential of the obtained results (AU) | |
| FAPESP's process: | 18/26780-6 - Longitudinal data modelling based on centred skew scale mixture of distributions: parametric and semi-parametric approaches |
| Grantee: | João Victor Bastos de Freitas |
| Support Opportunities: | Scholarships in Brazil - Master |