Skip to content

lme4 Model Criticism

Workshop 27 - Winter Semester 2026-27

Field Details
Date 30 September 2026
Time 14:00 - 15:30
Location House of Prominence, Attic (Top floor), Luxemburger Str. 299, Cologne
Speaker Job Schepens
Prerequisites Basic knowledge of R and linear mixed-effects models (lme4)
Materials status Workshop materials available
Slides / recording Session 29: lme4 Model Criticism & Advanced Mixed-Effects Diagnostics (HTML)

Overview

Diagnostics, assumptions, model comparison, and robust interpretation for lme4 models.

Linear mixed-effects models fitted with lme4::lmer() or lme4::glmer() are standard tools across linguistics and cognitive science. However, fitting a model is only the first step. This workshop focuses on model criticism: the essential diagnostics, checks, and decisions needed after fitting mixed-effects models to ensure that conclusions are robust, well-specified, and justifiable.

Topics Covered

  • Residual Diagnostics & Model Assumptions: Checking linearity, normality of residuals, and homoscedasticity (performance, DHARMa, residual plots)
  • Random Effects Structure & Singular Fit Warnings: Diagnosing boundary (singular) fits, correlation parameters near $\pm 1.0$, and strategies for simplifying or regularizing random effect specifications (trouBBlme4SolveR::dwmw())
  • Contrast Coding vs. Post-Hoc Estimation: Why contrast coding (sum coding) is essential before fitting to stabilize random slopes and avoid singular fits, and why emmeans is essential post-fitting for balanced marginal means, interaction slicing, and scale back-transformation
  • Ordinal & Likert Data: Diagnosing the failure of naive Gaussian models and fitting Cumulative Link Mixed Models (ordinal::clmm)
  • Influential Observations & Outliers: Identifying cluster-level leverage points with influence.ME (Subject and Word Cook's distance)
  • Model Comparison & Selection: Likelihood ratio tests (anova()), REML vs. ML rules, AIC/BIC, and Nakagawa's $R^2$
  • Convergence Issues & Optimizers: Troubleshooting convergence warnings with optimizer benchmarking (allFit())
  • Reporting Mixed-Effects Models: Best practices and open science reporting checklists (Meteyard & Davies, 2020)

Session Details

  • Date: 30 September 2026
  • Time: 14:00 - 15:30
  • Location: House of Prominence, Attic (Top floor), Luxemburger Str. 299, Cologne
  • Speaker: Job Schepens

← Back to Workshop Overview