Aiming at studying the criteria of selection ranking in the intermediate cycles of plant breeding, 900 experiments simulated in randomized complete block designs considering a coefficient of variation of 10%, and differences of 17% to -17%, between the mean of the lines and the mean of the control. Three cycles of selection and three line groups within each cycle were considered the first with 36, the second with 143 and the third with 134 selections, in the first cycle. The percentage of selection at the end of the 1st cycle, was 31.4, 10.6 and 11.3% for each group, and, at the end of the 2nd cycle, 45.4% for the 1st group and 40% for the other two. In the 3rd cycle, the discriminative power of `t' the tests Student, Duncan, Dunnett (unilateral), SNK and Tukey were also studied . The results indicated, that for one hundred or more selections in the firsts cycle, they must be allocated in 3 or 4 small experiments including the control, using a 40% intensity of selection. In the final stage of selection differences of 15, 12.5, 10, 7.5, 5 and 2.5% were considered. The discriminative power was evaluated using the same statistical tests of the 3rd cicle. Coefficients of variation of 10, 15 and 20% for 2,700 simulated individual experiments considering three types of years (good, normal and bad). Groups of nine experiments were analyzed and the discriminative power of the tests evaluated: in the individual experiments it was variable, with differences due to the type of year and coefficient of variation. The discriminative power of the different tests increased in the groups of experiments up to or above 78%, when the C.V. was 10% and differences were equal to or greater than 7.5%.
plant breeding; percentage of selection; ranking; statistical tests
