**Bayesian mixed-effects models for psychophysical data: A tutorial with R and brms**
**Thomas Wallis**<br/>
*Werner Reichardt Center for Integrative Neuroscience, Eberhard Karls Universität Tübingen, and the Bernstein Center for Computational Neuroscience, Tübingen, Germany*
Psychophysical experiments produce hierarchical data. We sample observers from a population of possible observers, and often we sample stimuli (e.g. natural images) from a population of possible stimuli. Ideally we would like to estimate model parameters (such as thresholds and slopes in a psychometric function) in a way that accounts for the hierarchical nature of our data. Mixed-effects models, in which both population- and sample-level parameters are estimated, allow one to do this – but these models can be difficult to fit using standard maximum-likelihood estimation. In this tutorial I will introduce the R package “brms” (Bayesian Regression Models in Stan; Bürkner 2016), which allows one to estimate even complex nonlinear mixed-effects models in an understandable high-level formula language. I will show examples and provide code for typical psychophysical models, and discuss GLMs, contrast coding, parameter interpretation and model comparison. Attendees will work with examples on their own laptops (prior setup required). Prior familiarity with Bayesian methods, GLMs, and R (particularly the tidyverse packages) will be helpful.
