Skip to content

Bayesian Mixed Effects Models with brms for Linguists

Date: 12. November 2025
Time: 14:00 - 15:30
Speaker: Job Schepens (Project S, CRC 1252)

Workshop Summary

This workshop covers the fundamentals of Bayesian mixed effects modeling using brms, with a focus on two common psycholinguistics experiments:

  1. Reaction Time (RT) data — continuous response times (log-transformed)
  2. Grammaticality Judgments — binary acceptability judgments (logistic regression)

We will learn how to specify domain-specific priors, validate them with prior and posterior predictive checks, and interpret Bayesian mixed effects models. The workshop emphasizes practical implementation in R and understanding the Bayesian workflow.

Learning Objectives

  • Understand how to set domain-specific priors instead of using defaults
  • Validate priors with prior predictive checks
  • Fit mixed effects models with brms
  • Assess model adequacy with posterior predictive checks
  • Interpret posterior distributions and credible intervals
  • Understand hyperpriors and regularization of random effects

Core Topics (60-75 min)

  1. Setting Priors in brms
  2. Flat, weakly informative, and regularizing priors

  3. Hyperprior standard deviations for random effects

  4. Prior Predictive Checks

  5. Validating prior assumptions before model fitting

  6. Posterior Predictive Checks

  7. Assessing model adequacy after fitting

For Later Discussion

  • Sensitivity analysis with different priors
  • Convergence diagnostics (Rhat, ESS, mcmc_plot, mcmc_dens_overlay)
  • Inference with hypothesis() and ROPE
  • Model comparison with Bayes Factors and LOO cross-validation
  • Other dependent variable types (count, ordinal, bounded)

Materials & Resources

Full workshop materials: brms-ws GitHub repository

References: - brms documentation - Stan prior choice recommendations

  • See schedule entry in agenda/winter-2025-26-schedule.md for context.