Computing info criteria multilevel models using diff sample sizes
summary
Shows how to extract logLik from nlme output and recompute information criteria so BIC uses clusters instead of observations, using str() to find logLik and multiplying by -2 to get deviance. Uses the model’s parameter count (3 in the example), computes BIC with m (number of groups) and with N (total observations) and verifies that BIC with N matches nlme’s reported value. Code lives on GitHub and the approach works for many models.
AI-written summary · about
related talks
45mThe Statistical Fairness-Accuracy Frontier
17mMaking sense of marginal effects
15mInterpretable Machine Learning with R
24mStochastic Block Models with R: Statistically rigorous clustering
21mSolving the model representation problem with broom
44mA Modern Maximum-Likelihood Theory for High-Dimensional Logistic Regression
31mBias Meets Bayes: A Bayesian Perspective on Improving Model Fairness
43mInformation-Theoretic Bounds and Phase Transitions in Community Detection and High-Dimensional Clust