Avaliação da normalidade, validade dos testes de médias e opções não-paramétricas: contribuições para um debate necessário.

Experimentation is an important methodological basis for innovations in the agricultural sector. Nevertheless, several aspects can be improved in the classical statistical analysis used in agricultural research. The objective of this review was to discuss a few conceptual elements and research results about the validity of statistical tests usually applied in experimentation and present some recommendations that can improve the quality of the analyzes commonly used in the scope of fixed models. Useful elements for the discussion of tests of means, assessment of the condition of normality, and non-parametric analysis options are presented. Understanding the statistical hypotheses and Type I error subtypes, for example, can help in better result interpretation and choice of means test. Some doubts about the evaluation of the normality requirement of the residues explored here can also help researchers better use parametric statistical tools. Finally, we present a general decision flowchart and a brief exemplified discussion of some non-parametric analysis options with emphasis on the differences between classical methods and methods based on generalized models. [ABSTRACT FROM AUTHOR]