Pressupostos multivariados e efeito dos parâmetros do modelo em análises multivariadas para ensaios com a aveia

Abstract

Oat is one of the main winter cereals grown in the world, used in human food and animal feed, ground cover, straw production, and crop rotation in the no-tillage system. In order to enhance the oat production systems, statistical techniques have been used to study the linear relationships between characters, in order to identify characters that directly or indirectly favor the selection of superior genotypes, among these techniques the linear correlation stands out, and path analysis. When performing multivariate analyses such as path analysis, some statistical assumptions must be met to avoid obtaining biased results. Furthermore, when working with this technique, the parameters of the mathematical model referring to the experimental design and treatments are disregarded, using only average observations, without stratifying the possible effects. Therefore, this study was developed with the aim of analyzing the implications of removing the parameters from the mathematical model on the results of Pearson correlation analysis and path analysis, in field trials with the oat crop, cultivated in different years and stratifying agricultural scenarios (with and without the use of fungicide). The experiments were conducted from 2015 to 2019, in the municipality of Augusto Pestana, Rio Grande do Sul, Brazil. The experimental design used was complete randomized blocks, with treatments characterized by oat cultivars and fungicide applications, with three replications. For each year, scenario, and data group, a multicollinearity diagnosis was performed, Pearson's correlation coefficients were calculated, and a path analysis was performed. The occurrence of multicollinearity generates biased path coefficients without biological interpretation, regardless of the environment and data group analyzed. Removing parameters from the mathematical model changes the explanatory capacity of characters in relation to yield variance, for all environments, scenarios, and types of path analysis performed. Removing the effects of model parameters results in changes in direction and magnitude (>50%) in the path coefficients regardless of the environment, scenario, and type of path analysis performed.