Thomas Lumley • 1/18/2019

Another way to see why mixed models in survey data are hard:

This technical article examines why fitting linear mixed models (like random intercept models) to survey data with multi-stage sampling (e.g., schools and students) is problematic. It details how standard weighting approaches using inverse sampling probabilities for different model components fail, leading to biased variance estimates, and discusses the conceptual clash between model-based and design-based inference.

0 comments

#Linear Regression #R #Statistical Computing