Efficiency of nonparametric methods for the analysis of the nodulation capacity of a leguminous crop

Abstract

The validity of the F test and the efficiency of various nonparametric tests for the analysis of the data of nodulation of soybean was studied in the following combinations: F test and that of Kruskal-Wallis (complete randomized design); F test and that of Quade & Friedman (randomized block design); F test and Bennett & Zimmermann (Latin square). All tests were made with four combinations of treatments and replicates. The parameters studied were: number of nodules, dry weight of nodules and dry weight of aerial tissue. The study of the efficiency of each test was performed using 1000 simulations for each case, using data originated from a study of uniformity with 900 plots. The results showed that the parameters associated with nodulation were not normally distributed and their distributions differed from any other standard distribution. The F test was shown to be robust and had a good capability to analyse this type of experiment. The tests of Kruskal-Wallis and Quade & Friedman showed practically the same efficiency as the F test, while that of Bennett & Zimmermann was found to be superior. The test of Friedman was found to be invalid.