Acessibilidade / Reportar erro

Discriminative power of the combining probabilities test in analysis series of experiments

The classical analysis of series of experiments presents some levels of complications due, among several reasons, to the possibility of existence of heterogeneity of variances from place to place, lack of homogeneity of errors from one season to another or different number of treatments or replications in each place. Fisher developed a method, to analysis series of experiments, that combine the probability of the means tests results or tests of contrasts, independently of the test used, the number of replications, the structure of the experiments or the presence of interactions. In order to evaluate the influence of the experimental precision, the magnitude of the general mean of the experiment and the magnitude of the contrasts between the control and another treatment mean on the test proposed by Fisher, using the 't' test of Student and that of Bonferroni, it was simulated 2,160 experiments in completely randomized block design with six treatments and three different number of replications, four, eight and twelve. The combined analysis was made considering three strata defined by the precision and level of yield production: good (B), median (M) and low (R). The experiments were grouped within each strata, for example, three years (BBB), (MMM) or (RRR) and another group considering different strata (BMR). The results showed that the magnitude of the coefficient of variation and the magnitude of the contrasts were very important on the discriminative power of the tests used; when the groups were formed by experiments from three different levels of precision and yield, using at least eight replications and considering the highest value of the contrast between means, the use of the test of combining probabilities showed that the discriminative power of the Student "t" test was greater than 90% and that of Bonferroni's was greater than 80%; when it was combined four or five experiments including more than one experiment of median precision the discriminative power of the "t" test was greater than 80% when the number of replication used was eight, but it was needed twelve replications for Bonferroni's test.

combining probabilities tests; "t" test; Bonferroni's test; analysis of series of experiments


Instituto Agronômico de Campinas Avenida Barão de Itapura, 1481, 13020-902, Tel.: +55 19 2137-0653, Fax: +55 19 2137-0666 - Campinas - SP - Brazil
E-mail: bragantia@iac.sp.gov.br