Statistical model assumptions achieved by linear models: classics and generalized mixed

• Autor(a) principal: Melo,Rita Carolina de
• Data de Publicação: 2020
• Outros Autores: Trevisani,Nicole, Santos,Marcio dos, Guidolin,Altamir Frederico, Coimbra,Jefferson Luís Meirelles
• Tipo de documento: Artigo
• Idioma: eng
• Título da fonte: Revista ciência agronômica (Online)
• Texto Completo: http://old.scielo.br/scielo.php?script=sci_arttext&pid=S1806-66902020000100415
• topic: Analysis of variance; Homogeneity of variance; Normality of errors; Crop breeding; Generalized linear mixed models

ABSTRACT When an agricultural experiment is completed and the data about the response variable is available, it is necessary to perform an analysis of variance. However, the hypothesis testing of this analysis shows validity only if the assumptions of the statistical model are ensured. When such assumptions are violated, procedures must be applied to remedy the problem. The present study aimed to compare and investigate how the assumptions of the statistical model can be achieved by classical linear model and generalized linear mixed model, as well as their impact on the hypothesis test of the analysis of variance. The data used in this study was obtained from a genetic breeding program on the cooking time of segregating populations. The following solutions were proposed: i) Classical linear model with data transformation and ii) Generalized linear mixed models. The assumptions of normality and homogeneity were tested by Shapiro-Wilk and Levene, respectively. Both models were able to achieve the assumptions of the statistical model with direct impact on the hypothesis testing. The data transformations were effective in stabilizing the variance. However, several inappropriate transformations can be misapplied and meet the assumptions, which would distort the hypothesis test. The generalized linear mixed models may require more knowledge about the identification of lines of programming, compared to the classical method. However, besides the separation of fixed from random effects, they allow for the specification of the type of distribution of the response variable and the structuring of the residues.