Uniformly Valid Inference Based on the Lasso in Linear Mixed Models

Abstract

Linear mixed models (LMMs) are suitable for clustered data and are common in biometrics, medicine, survey statistics and many other fields. In those applications, it is essential to carry out valid inference after selecting a subset of the available variables. We construct confidence sets for the fixed effects in Gaussian LMMs that are based on Lasso-type estimators. Aside from providing confidence regions, this also allows to quantify the joint uncertainty of both variable selection and parameter estimation in the procedure. To show that the resulting confidence sets for the fixed effects are uniformly valid over the parameter spaces of both the regression coefficients and the covariance parameters, we also prove the novel result on uniform Cramer consistency of the restricted maximum likelihood (REML) estimators of the covariance parameters. The superiority of the constructed confidence sets to naive post-selection procedures is validated in simulations and illustrated with a study of the acid neutralization capacity of lakes in the United States.

arXiv comment25 pages, 1 figure

References in corpus (10)

• On asymptotically optimal confidence regions and tests for high-dimensional models
• Valid post-selection inference
• Exact post-selection inference, with application to the lasso
• New Important Developments in Small Area Estimation
• Model Selection in Linear Mixed Models
• Estimation for High-Dimensional Linear Mixed-Effects Models Using ${\ell }_1$-Penalization
• Confidence Sets Based on Penalized Maximum Likelihood Estimators in Gaussian Regression
• Confidence Sets Based on Sparse Estimators Are Necessarily Large
• Uniformly Valid Confidence Sets Based on the Lasso
• On the Distribution of Penalized Maximum Likelihood Estimators: The LASSO, SCAD, and Thresholding