| Texto completo | |
| Autor(es): |
Galindo, Aurea F. L.
;
Azevedo, Caio L. N.
;
Nobre, Juvencio S.
Número total de Autores: 3
|
| Tipo de documento: | Artigo Científico |
| Fonte: | COMMUNICATIONS IN STATISTICS-SIMULATION AND COMPUTATION; v. N/A, p. 21-pg., 2025-09-19. |
| Resumo | |
It is common to find studies where the aim is to model proportion or ratio data in terms of a set of covariates, where for some of them the relationship with the response variable is not clear and nonlinear. In these situations, it is suitable to consider extensions of the Generalized Additive Partial Linear Models (GAPLM) that deal with data on the unit interval. In this work, we developed a new class of GAPLM based on beta, beta rectangular (BR), and simplex distributions. We considered a combination of parametric and nonparametric predictors for the response mean and parametric predictor for the precision/dispersion parameter. Under the frequentist paradigm, we developed parameter estimation, we adapted some Information Criteria and proposed quantile residuals. We discussed some inferential tools as standard errors and hypothesis test based on Wald-type statistic. Also, we developed measures of global influence, based on Cook's distance and generalized leverage, and local influence. Then, our approach offers a flexible way to analyze data in (0,1) interval in several ways: response variable, regression structure, and model fit assessment. Simulation studies and real data analysis were conducted to illustrate the potential of the developed methodologies, showing their advantage over some usual methodologies. (AU) | |
| Processo FAPESP: | 20/16713-0 - Modelos de regressão paramétrico e semi-paramétrico sob a classe de distribuições misturas de escala skew-normal |
| Beneficiário: | Caio Lucidius Naberezny Azevedo |
| Modalidade de apoio: | Auxílio à Pesquisa - Regular |
| Processo FAPESP: | 18/26784-1 - Modelos paramétricos e semi-paramétricos para dados longitudinais com resposta limitada |
| Beneficiário: | Áurea Fonseca Lopes Galindo |
| Modalidade de apoio: | Bolsas no Brasil - Mestrado |