pecme - Penalized ECME Estimation for Censored Linear Mixed Models

Fits Gaussian linear mixed models with a random intercept when the response is subject to left, right, and/or interval censoring, using the Expectation/Conditional Maximization Either (ECME) algorithm of Liu and Rubin (1994) in the spirit of the fast censored-response mixed-model algorithm of Vaida and Liu (2009). Simultaneous estimation and variable selection is supported through coordinate-descent penalized maximization with Lasso, Adaptive Lasso, SCAD, MCP, Elastic Net, and Ridge penalties (no penalty is also supported). The random intercept is integrated out by Gauss-Hermite quadrature at every iteration, and the two ECME conditional-maximization steps respectively maximize the expected penalized complete-data objective (for the regression coefficients) and the actual observed-data marginal likelihood (for the variance components), which is the defining feature of ECME relative to plain ECM/EM. The package provides a single-fit engine, a sequential/parallel penalty-parameter grid search with information-criterion or cross-validated selection, data-dependent or user-supplied lambda grids, and an Expectation-Maximization based treatment of a completely missing (at random) response, sharing the same truncated-normal machinery used for censoring. References: Liu and Rubin (1994) "The ECME algorithm: A simple extension of EM and ECM with faster monotone convergence" <doi:10.1093/biomet/81.4.633>; Vaida and Liu (2009) "Fast Implementation for Normal Mixed Effects Models With Censored Response" <doi:10.1198/jcgs.2009.07130>.