The Significance Test Controversy Revisited

1. Front Matter

   Pages i-xi

   PDF

2. Introduction

   • Bruno Lecoutre, Jacques Poitevineau

   Pages 1-7

3. Preamble—Frequentist and Bayesian Inference

   • Bruno Lecoutre, Jacques Poitevineau

   Pages 9-20

4. The Fisher, Neyman–Pearson and Jeffreys Views of Statistical Tests

   • Bruno Lecoutre, Jacques Poitevineau

   Pages 21-38

5. GHOST: An Officially Recommended Practice

   • Bruno Lecoutre, Jacques Poitevineau

   Pages 39-47

6. The Significance Test Controversy Revisited

   • Bruno Lecoutre, Jacques Poitevineau

   Pages 49-62

7. Reporting Effect Sizes: The New Star System

   • Bruno Lecoutre, Jacques Poitevineau

   Pages 63-75

8. Reporting Confidence Intervals: A Paradoxical Situation

   • Bruno Lecoutre, Jacques Poitevineau

   Pages 77-90

9. Basic Fiducial Bayesian Procedures for Inference About Means

   • Bruno Lecoutre, Jacques Poitevineau

   Pages 91-103

10. Generalizations and Methodological Considerations for ANOVA

    • Bruno Lecoutre, Jacques Poitevineau

    Pages 105-126

11. Conclusion

    • Bruno Lecoutre, Jacques Poitevineau

    Pages 127-129

12. Back Matter

    Pages 131-134

    PDF

The purpose of this book is not only to revisit the “significance test controversy,”but also to provide a conceptually sounder alternative. As such, it presents a Bayesian framework for a new approach to analyzing and interpreting experimental data. It also prepares students and researchers for reporting on experimental results. Normative aspects: The main views of statistical tests are revisited and the philosophies of Fisher, Neyman-Pearson and Jeffrey are discussed in detail. Descriptive aspects: The misuses of Null Hypothesis Significance Tests are reconsidered in light of Jeffreys’ Bayesian conceptions concerning the role of statistical inference in experimental investigations. Prescriptive aspects: The current effect size and confidence interval reporting practices are presented and seriously questioned. Methodological aspects are carefully discussed and fiducial Bayesian methods are proposed as a more suitable alternative for reporting on experimental results. In closing, basic routine procedures regarding the means and their generalization to the most common ANOVA applications are presented and illustrated. All the calculations discussed can be easily carried out using the freeware LePAC package.  

• Bayesian ANOVA
• Null Hypothesis Significance Tests
• experimental data report
• statistical inference

• ERIS, Laboratoire de Mathématiques Raphaël Salem, Université de Rouen UMR 6085 CNRS, St Et. du Rouvray CEDEX, France

  Bruno Lecoutre

• ERIS, IJLRA UMR-7190 case 162, Université Pierre et Marie Curie CNRS, Paris Cedex 05, France

  Jacques Poitevineau

Policies and ethics