Categorical regression model for the analysis and interpretation of statistical power

Resumen

Criteria of theoretical-practical value are established in models of analysis of variance of fixed effects (parametric and non-parametric), from an integral analysis of variables related to statistical indicators and experimental design, which includes statistical power as a dependent variable. The analyzed information was selected from independent researches, processed by the Biomathematics department from the Instituto de Ciencia Animal, developed in areas of birds, pigs, grasses and ruminants. The analyzed experiments correspond to completely randomized (CRD), balanced and random block designs (RBD). The results were processed by parametric F Fisher test and were compared with the non-parametric equivalent tests, Kruskal-Wallis and Friedman. A total of 21 experiments were selected, 16 related to the CRD and five to the RBD. For the analysis of data, a data matrix was created with the nine selected variables. It is considered, as the most outstanding result, the strong negative relation that is manifests between the power and the probability
of type I error in the analysis of variance models (parametric and non-parametric). That is, at low values of the probability of type I error, high values of power. It is convenient, in future studies, to deepen in the aspects of sample size, the distribution of the variable under study and the criterion of power-efficiency (Asymptotic Relative Efficiency, ARE), in relation to the probability of type I error and the power.
Key words: Statistical and experimental design indicators, analysis
of variance parametric and non-parametric.