Precision gain and sample size: use of precision statistics for soybean early-cycle traits

Abstract Empirical decisions on the number of samples to be collected per experimental unit may cause experiments to obtain misleading results. Once little attention has been given to soybean early-cycle traits, the aim of this study was to find the optimal sample size to estimate the overall experimental mean, coefficient of variation, and least significant difference for soybean early-cycle traits; and to assess precision gain as a function of sample size. Two experiments were performed in Erval Seco, Rio Grande do Sul, Brazil, with 30 soybean cultivars in a randomized block design with three repetitions. Thirty-nine days after emergence, 20 plants per experimental unit were measured, totaling 3,600 sampled plants. Then, sampling scenarios of 1, 2, ..., 100 plants were considered and the statistics overall experimental mean, coefficient of experimental variation, and least significant difference at 5% error probability were estimated in each scenario. In general, more precise estimates were made for the three precision statistics as a higher number of plants per experimental unit was sampled. However, precision gain tends to stabilize at ~ 70%. Therefore, a sample size of 17 plants per experimental unit was sufficient to estimate the experimental precision statistics reliably for soybean early-cycle traits.