Disposição sistemática dos canteiros: sua influência sôbre a estimativa do êrro experimental

[摘要] This paper describes the results of superimposing certain types of 5 x 5 latin squares on a wheat uniformity trial. The purpose was to investigate the bias in the estimate of error when certain systematic squares (knight's move and diagonal) are chosen and to compare the results under conditions in Brazil with those obtained by Tedin in a similar investigation. The following conclusions are drawn : a) when latin squares are chosen by a random process, as recommended by Fisher, the observed distribution of the variance ratio of "treatments" compared with error is i n agreement with that given by theory; b) the estimation of error variance is biased when systematic squares are employed. In agreement with Tedin, we find that the knight's move square furnishes an overestimate, and the diagonal square an underestimate of error variance. The results show that, under the conditions of our trial, the systematic squares suffer from the same disadvantages which have been noted elsewhere.