The assessment of coefficients of variation (CV) as a measure of precision in experiments has been done for various crops, and some researchers have proposed a methodology to get classification bands of CV, and others defining the classification bands itself; based on the mean and on the standard deviation of the CV, assuming that they have normal distribution, which is not always true. This study had as objective to propose a methodology for CV classification bands definition, independently of their distribution; the new methodology is based on the use of the median and the pseudosigma. Data from 110 experiments of upland rice, designed in randomized complete blocks, generalized incomplete blocks and square lattices were analyzed, selecting the characters related to diseases, lodging, yield and yield components. CV classification bands were constructed for each character studied, in general, and considering the experimental design utilized for experiments with 30 to 100 treatments. The method proposed is efficient for the definition of CV bands, independently of their distribution, and the variation of the values of CV evidences the importance of considering, not only the variable in study, but also the experimental design. Experiments with smaller CV were those that utilized Incomplete Block Design. The variables related to diseases and lodging were those that presented larger coefficient of variation.
Oryza sativa; experimentation; analytical methods
