Scientific Self-Correction: The Bayesian Way

The enduring replication crisis in many scientific disciplines casts doubt on the ability of science to self-correct and to produce reliable knowledge. There are different approaches for addressing this challenge. Social reformists hypothesize that the social structure of science such as the credit reward scheme must be changed. Methodological reformists suggest changes to the way data are gathered, analyzed and shared, such as compulsory pre-registration or multi-site experiments with different research teams. Statistical reformists argue more specifically that science would be more reliable and self-corrective if null hypothesis significance tests (NHST) were replaced by a different inference framework, such as Bayesian statistics. On the basis of a simulation study for meta-analytic aggregation of effect sizes, we articulate a middle ground between the different reform proposals: statistical reform alone won't suffice, but moving to Bayesian statistics eliminates important sources of overestimating effect sizes.