Análise estatística de ensaios agronômicos não balanceados

Abstract

The present work is a contribution to the demonstration of methods that minimize distortion in the analysis of data from agronomic trials. It was specifically intended: > review some unbalanced data analysis methods; > demonstrate the application of these methods in parameter estimation; > Compare balanced data analysis results using different methods. To illustrate this contribution, data from trials with peanut crops carried out in the field of the faculty of agronomy and forestry engineering in the 1984/1985 and 1990/1991 agricultural seasons and also in marracuene in the 1983/1984 agricultural season were used. The parameters estimation methods described were: least squares and maximum likelihood. Unbalanced data were subjected to analysis of variance with missing data and regression analysis with indicator variables. The first method consists of estimating the missing values ​​using the iterative process and following with normal analysis of variance, while the second analyzes the available information with the aid of the sum of extra squares principle. Using the two approximations, the same conclusions are reached despite the difference observed in the variance components. However, it is noted that regression is a viable alternative for analyzing this type of data, since it is not necessary to estimate missing data. The balanced data were also submitted to traditional regression and variance analyses, having arrived at the same results both in numerical terms and in terms of inferences with the application of the two methods. This proves the truth of the regression method