Statistical selection for estimating the accuracy in experimental corn trials.
The objective of this study was to select the most optimal statistic to estimate the experimental accuracy and to assess the Coefficient of Variation often used to estimate the validity of experiments. For the analysis, 406 trials were included; from research conducted at El Ejido Experimental Station and at cornfields of contributing producers of Azuero Region, Panama, during 2000-2013. To each trial it was calculated the various components of variance or Mean Square (MS), Coefficient of Variation (CV), Repeatability, Rank, Least Significant Differences (LSD), Standard Error (SE), Coefficient of determination (R2) of treatment and residual. We found that the CV is more related than repeatability with the overall mean of the experiment (correlation index of 0.57 vs. 0.24), thus being less robust to indicate experimental accuracy. Increasing the number of repetitions of experiments with the same MS Error reduced the SE and increased accuracy. The repeatability of the experiment was highly related to both treatments R2 (0.93) and the unexplained fraction model (0.87), in contrast with the CV (0.23 and 0.23, respectively). When repeatability and CV were linked to the DMS/Range ratio, it was found that repeatability is highly correlated (R2 = 0.76) with this ratio, while the CV provided an R2 of 0.18. This result suggests that the repeatability is a better indicator than the CV for good experimental accuracy.
Keywords: repeatability, coefficient of variation, mean square error, standard error of difference