RXshrink: Maximum Likelihood Shrinkage using Generalized Ridge or Least Angle Regression Methods

Functions are provided to calculate and display ridge TRACE diagnostics for a variety of shrinkage Paths. TRACEs identify the m-Extent of shrinkage most likely, under Normal-theory, to produce optimally biased estimates of beta-coefficients with minimum MSE Risk. The unr.ridge() function implements the "Unrestricted Path" introduced in Obenchain (2020) <arXiv:2005.14291>. This Shrinkage-Path is more efficient than the Paths used by the qm.ridge(), aug.lars() and uc.lars() functions. Optimally biased predictions can be made using RXpredict() for all six types of RXshrink linear model estimation methods. Functions MLboot(), MLcalc(), MLhist() and MLtrue() provide insights into the true bias and MSE risk characteristics of non-linear Shrinkage estimators. Functions unr.aug() and unr.biv() augment the calculations made by unr.ridge() to provide plots of the bivariate confidence ellipses corresponding to any of the p*(p-1) possible pairs of shrunken regression coefficients. The correct.signs() function provides estimates with "correct" numerical signs when ill-conditioned (nearly multicollinear) models yield OLS estimates that disagree with the signs of the observed correlations between the y-outcome and the selected x-predictor variables.

Version: 1.4.3
Depends: R (≥ 3.5.0), lars, ellipse
Published: 2020-11-01
Author: Bob Obenchain
Maintainer: Bob Obenchain <wizbob at att.net>
License: GPL-2
URL: https://www.R-project.org , http://localcontrolstatistics.org
NeedsCompilation: no
In views: MachineLearning
CRAN checks: RXshrink results

Downloads:

Reference manual: RXshrink.pdf
Package source: RXshrink_1.4.3.tar.gz
Windows binaries: r-devel: RXshrink_1.4.3.zip, r-release: RXshrink_1.4.3.zip, r-oldrel: RXshrink_1.4.3.zip
macOS binaries: r-release: RXshrink_1.4.3.tgz, r-oldrel: RXshrink_1.4.3.tgz
Old sources: RXshrink archive

Linking:

Please use the canonical form https://CRAN.R-project.org/package=RXshrink to link to this page.