Selection of lines in plant breeding. ll. Discriminative power of several statistical tests.

Abstract

At the initial stages in plant breeding work, due to the high number of items compared, usually it is used a j selection pressure. At the final stages, where a reduced number of lines is tested, they must be compared with a "test cultivar", using designs like complete block or small lattice. Trying to get information about the discriminative power of these tests, some of them were compared. They were: least significance difference, restrict Bonferroni's test, Dunnett's test and Tukey's test. It was evaluated the influence of several characteristics on the discríminative power, such as the number of replications, the coefficient of variation, the magnitude of the real difference between the treatments and the control, use or not of balanced designs and the existence or not of significative F for treatments. 1,400 experiments were simulated using a 5 x 5 lattice design, with treatment means differing from + 25% to - 25% of the control. Considering all cases it was possible to conclude that in the conditions studied the highest discriminative power was obtained with the least significance difference test, followed by Bonferroni's test, and Dunnett's test. Tukey's test was definitively the worst one