This function makes classical (rather than Bayesian) Normal distribution-theory calculations of the form proposed in Obenchain(1977). Instead of providing "new" confidence regions for estimable linear functions, Generalized Ridge Regression (GRR) can focus interest on estimates that are within traditional confidence intervals and regions but which deviate reasonably from the centroid of that interval or region.
unr.aug(urobj)
An output object of class "unr.ridge".
An output list object of class "unr.aug"...
Number of regression predictor variables.
The lm() output object for the model fitted using unr.ridge().
The p by p+2 matrix of shrunken GRR coefficients. The first p correspond to "knots" on piecewise linear splines, and the last column contains minimum MSE risk estimates.
p+2 increasing measures of shrinkage "Extent". The first is 0 for the OLS (BLUE) estimate, the next to last denotes the "Extent" most likely to yield minimum MSE risk, and the last is p [this shrinkage terminus is frequently outside of the unr.biv() plot frame].
Names of variables used in the GRR model.
Obenchain RL. (1977) Classical F-Tests and Confidence Regions for Ridge Regression. Technometrics 19, 429-439. http://doi.org/10.1080/00401706.1977.10489582
Obenchain RL. (2020) Ridge TRACE Diagnostics. https://arxiv.org/abs/2005.14291
Obenchain RL. (2020) RXshrink_in_R.PDF RXshrink package vignette-like file. http://localcontrolstatistics.org