Last week I attended the 84th Annual Meeting of the Psychometric Society in Santiago, Chili. Together with Maarten Marsman I taught a JASP workshop on Monday, and then gave a keynote on Tuesday. The keynote was called “Bayesian multi-model inference for practical and impractical problems” and you can find the slides here. Due to poor planning on my side, I could only present the practical problems, not the impractical ones.
My main objective was to discuss Bayesian model averaging without any equations, and it seems to have worked well (how could it not, with a cartoon of demons at one’s disposal?). A short summary of the talk:
- Researchers are usually not motivated to emphasize statistical uncertainty;
- The statistical uncertainty we see is just the tip of the iceberg; Many partial solutions exist to reveal the hidden uncertainty, one of them being Bayesian multi-model inference;
- Bayesian multi-model inference was explained with a cartoon, developed by myself and Max Hinne, and drawn by Viktor Beekman. The cartoon is also presented in a preprint — for more information see an earlier post.
- Four example of multi-model inference are provided:
- Effect size estimation (with Geoff Iverson and Michael Lee)
- Linear regression (with Merlise Clyde and Don van den Bergh)
- Meta-analysis (with Quentin Gronau and others)
- Multinomial processing trees (with Quentin Gronau and Dora Matzke)
Just as with any statistical technique, once you are aware of multi-model inference (which need not be Bayesian, see for instance Burnham & Anderson, 2002), it seems that the concept applies to just about every situation!
Burnham, K. P., & Anderson, D. R. (2002). Model selection and multimodel inference: A practical information-theoretic approach (2nd ed.). New York: Springer.
Hinne, M., Gronau, Q. F., van den Bergh, D., & Wagenmakers, E.-J. (2019). A conceptual introduction to Bayesian model averaging. Manuscript posted on PsyArXiv .