Using Bayesian statistics to revisit statistical evidence

Large parts of the scientific literature are untrustworthy. A major culprit in this so-called reproducibility crisis is the over-reliance on inappropriate statistical evidence criteria (for a demonstration in the field of biomedicine, see van Ravenzwaaij & Ioannidis, 2016). This project is concerned with reanalyzing existing data using Bayesian statistics. Such a reanalysis allows us to develop a targeted approach for solving the reproducibility crisis by differentiating between compelling evidence for an effect, ambiguous evidence for an effect, and compelling evidence for the absence of an effect. By identifying those studies for which the statistical evidence is either inconclusive or more indicative of the absence of an effect, a targeted replication program can take place.

Researchers and partners

Behavioural and Social Sciences, Psychology

• dr. D. (Don) van Ravenzwaaij, Psychometrics and Statistics

Behavioural and Socials Sciences, outside of Psychology

• dr. R. (Rink) Hoekstra, Educational Sciences

Education

Courses connected to this project

• Transparency in Science

Results

Publications

• van Ravenzwaaij, D. & Ioannidis, J. P. A. (2016). A Simulation Study of the Strength of Evidence in the Endorsement of Medications Based on Two Trials with Statistically Significant Results. Manuscript submitted for publication.
• Hoekstra, R., Monden, R., van Ravenzwaaij, D., & Wagenmakers, E.--J. (2016). Bayesian Reanalysis of Null Results Reported in the New England Journal of Medicine: Strong yet Variable Evidence for the Absence of Treatment Effects. Manuscript submitted for publication.