The significance test controversy revisited. The fiducial Bayesian alternative (Q744208): Difference between revisions

This work deals with the ``significance test controversy'' and proposes a conceptually sounder alternative to perform the hypothesis testing, namely, the fiducial Bayesian methods. What is meant by this is the objective Bayesian inference, in this case, the Bayesian framework using Jeffreys' prior. The presentation in the book is methodologically oriented. The Bayesian methods developed here are in the analysis of variance framework. It starts with basic notions of frequentist and Bayesian approaches to inference. Then, it discusses three views of statistical testing, the Fisherian, the Neyman-Pearsonian and the Jeffreys' Bayesian. The misuses of the null hypothesis significance tests are exposed next, followed by how the Jeffreys' Bayesian conceptions would reconsider such problems. Next, the book examines what the recommended frequentist ``good statistical procedures'' are in testing effect sizes of the mean and of the univariate and multivariate ANOVA. Examples where such procedures go weird are shown. Finally, routine procedures under fiducial Bayesian perspective are introduced for inference about the mean and ANOVA effect sizes. The authors also indicate the availability of a statistical software, LePAC, which incorporates both frequentist practices and Bayesian methods to help in the computations involved.