Video: “A Bayesian Perspective of the Reproducibility Project: Psychology”

I recently gave a talk at the University of Bristol’s Medical Research Council Integrative Epidemiology Unit, titled, “A Bayesian Perspective on the Reproducibility Project: Psychology,” in which I recount the results from our recently published Bayesian reanalysis of the RPP (you can read it in PLOS ONE). In that paper Joachim Vandekerckhove and I reassessed the evidence from the RPP and found that most of the original and replication studies only managed to obtain weak evidence.

I’m very grateful to Marcus Munafo for inviting me out to give this talk. And I’m also grateful to Jim Lumsden for help organizing. We recorded the talk’s audio and synced it to a screencast of my slides, so if you weren’t there you can still hear about it. 🙂

I’ve posted the slides on slideshare, and you can download a copy of the presentation by clicking here. (It says 83 slides, but the last ~30 slides are a technical appendix prepared for the Q&A)

If you think this is interesting and you’d like to learn more about Bayes, you can check out my Understanding Bayes tutorial series and also our paper, “How to become a Bayesian in eight easy steps.”

2 thoughts on “Video: “A Bayesian Perspective of the Reproducibility Project: Psychology”

  1. This was very clear; this is a great format in which to learn about this stuff.

    I’m curious to learn more about what justifies that H1 ~ N(0,1) assumption, and why it’s better than an alternative (e.g., a bimodal distribution where each component is centered around d = .5 and d = -.5, or something like that) I know the Rouder et al. paper is one source to learn more about this. Any other suggestions— especially ones in this more accessible format?

    • Thanks. The priors always carry a shred of arbitrariness with them. Really there’s no principled reason to use N(0,1) over N(0,.999), etc. But as long as the prior captures some relevant theoretical information then the exact numerical choices don’t usually matter much. The N(0,1) says if any effect we study is non-zero then it is probably not bigger than |d|=1 and almost surely smaller than |d|=2. For most psych studies I think that is reasonable.

      Johnson and Rossell recently proposed using non-local priors like the kind you mention (non-local means the mode isn’t on zero). I think that paper is open access, but unfortunately it isn’t super accessible because it has quite heavy notation.

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s