This repository contains the RMarkdown script to generate slides for an introductory course on Bayesian multilevel modeling. This was presented at the GDR Vision conference 2018, in Paris.
You can visualize the slides online at this link (press left/right arrow keys to change slide).
This repository contains also files with posterior samples of the fitted models (.rds or .RData formats). The code of stan models is indicated with the .stan extension. The file code_GDR2018.R documents all the analyses step by step, and the file GDR2018_slidesML.Rmd is the Rmarkdown script used to generate the slides (ioslides presentation in .html format, GDR2018_slidesML.html).
Tutorial Summary
Multilevel models are ideally suited to the analysis of data that have a hierarchical structure, such as it is the case in psychology and neuroscience, where observations (e.g. trials of an experiment) are grouped according to 'observational clusters' (e.g., the participants of the experiment). In this tutorial I will first introduce multilevel models (also known as hierarchical or mixed-effects) in a frequentist setting, using R (www.r-project.org/) and the lme4 library. Next, I will move toward a Bayesian approach, and illustrate how to fit multilevel models in a Bayesian framework, using Stan (mc-stan.org/) and its R interface (see here for installation instructions: github.com/stan-dev/rstan/wiki/RStan-Getting-Started). One of the advantages of Stan is that it enables users to build the model they want, instead of having to choose from a limited set of default models. Time permitting, we will see through a series of worked examples how this can be particularly useful in experimental psychology, as it allows estimating at the group level more realistic 'process' models.