Métodos estatísticos em pesquisa com plantas daninhas: escolhendo adequadamente

摘要

Statistical concepts and methods play an important role in the society, and statistical data analysis require considerable human labor and knowledge. From one side, computers and statistical softwares allow almost anyone to run free on statistical methods, but on the other side any researcher, professor, student or professional, even lacking on basic statistical knowledge to test their data, may use these softwares, often producing biased statistical analyses. The objective of this review is to demonstrate how the choice for statistical methods in weed science may create a bias in the interpretation of herbicide efficiency, and impact herbicide recommendations. We propose minor changes to the ordinary approach to help avoiding data misinterpretation and unintentional erroneous herbicide recommendations. The problems discussed throughout the review are illustrated with real field experimental data. Great part of the results of studies involving herbicide efficacy seems to be based on underpowered experiments and prone to output distorted information. Flawed choices of statistical methods, specially the p-value based statistics (ANOVA and post-hoc tests), can pave the way for mistaken conclusions even in properly conducted experiments in weed research. It is proposed the use of confidence intervals for both qualitative and quantitative data analysis, coupled to an appropriate number of samplings (“n”).