Unlocking the file drawer of unpublished studies

What is it about?

Publication bias - the preferential publishing of studies that show the predicted effect - leads to an overestimation of effect sizes in research. We developed statistical techniques that correct for this bias, in order to estimate the effect size as it would manifest itself if all experiments were equally likely to be published.

Featured Image

Photo by Maksym Kaharlytskyi on Unsplash

Photo by Maksym Kaharlytskyi on Unsplash

Why is it important?

Several methods already exist to adjust for publication bias. Our method applies all of them at the same time and weights them by how well they describe the data (Bayesian model-averaging). Through simulation studies and applied examples, we show that robust Bayesian meta-analysis outperforms the alternatives in recovering the effect corrected for publication bias.