Intro to Bayes at STEP 2023
The Spring Training in Experimental Psycholinguistics is a mostly annual methods school targeted at doctoral students and post-docs looking to increase their knowledge of state-of-the-art research methods. I was invited to give a 10-hour introductory course on Bayesian modelling, and it was some of the most fun I’ve ever had teaching. I initially planned on covering hierarchical models but realized after the first two sessions (of six) that it should be scrapped so that we could focus on the fundamentals of choosing priors and fitting models, with a special focus on prior predictive simulation to understand priors in generalized linear models.
I learned a lot by doing this, and I do have a list of what I would change in future iterations. However, I was very impressed with how much attendees learned in such a short time frame! Of course, they were a room full of intelligent and self-motivated researchers, so I can only take small fraction of the credit.
The general format was that there were two “lessons” in each session, each followed by an “exercise”. The lessons were the part where mostly I talked, and the exercises were where I asked the attendees to open up R, load some data, and do some stuff. They alternated, with the . So S1L1 is “Session 1 Lesson 1” and S2E2 is “Session 2 Exercise 2”.
Here I share my slides from the material that we covered. You can also download
S1L1: Setting expectations and meeting our data
S1L2: Getting started with Bayesian inference
S2L2: A conceptual introduction to Hamiltonian Monte Carlo
S3L1: Getting categorical variables into our models: an intro to contrast coding
S3L2: Prior predictive checks in brms
S4L1: Lognormal model of reaction times
S4L2: Comparing Bayesian models
S5L1: Logistic regression in brms
Oh, the Places You’ll Go: flexible modelling options available with brms