509 – The Lasso and L1 Regularization Methods
LASSO for Clustered Data
Rosanna Overholser
University of California at San Diego
Ronghui Xu
University of California at San Diego
The LASSO was introduced by Tibshirani for the purposes of estimation and variable selection in linear regression. Most work on the LASSO has included the assumption of independent observations. Several papers have recently extended the LASSO to linear mixed models for clustered data. We will examine through simulations a further extension of the LASSO to general linear mixed models for clustered data that contain within-cluster correlation. Regression splines for correlated data can be formulated as a general linear mixed model so the problem of knot selection for splines is equivalent to variable selection of fixed effects. We can therefore use the LASSO to simultaneously select knots and estimate variance parameters. We apply our methods to functional MRI time courses from several subjects. This work is joint with Ronghui Xu.
