This study aimed to quantify the proportion of experiments with statistical difference for blocks, cultivar and the F test assumption (additivity, normal distribution, random error, and error variance homogeneity among cultivars) observance. Also, the effects on the least significant difference (LSD) based on Tukey test (P=0.05) were quantified when some assumption was not observed. The statistical analysis was based upon 307 sets of experimental cultivar competition data carried out from 1993 to 1996 in the State of Rio Grande do Sul, Brazil. Among the experiments with significant differences in treatments (cultivars), 85.67% had an average LSD equal to 30.28%. Low frequency of non assumption observance was verified and increased the LSD. Lack of additivity on the model occured in 8.75% the trials and not modify LSD of data transformation. No random error effect was verified in 9.88% of the experiments due to variability among plots at the same block caused by inadequate block placement in the experimental area.
precision; assumptions; residual analysis
