pca_2step: PCA when first 'vr' rows have a variance multiplicatively...

In dcgerard/vicar: Various Ideas for Confounder Adjustment in Regression

Description

Modified truncated SVD. The variance estimates are just the column-wise mean-squares of the last n - vr rows. This form of factor analysis is mostly for variance inflation with RUV2.

Usage

pca_2step(Y, r, vr, limmashrink = TRUE, likelihood = c("t", "normal"))

Arguments

Y	A matrix of numerics. The data.
r	the rank.
vr	The number of the first few rows whose variances differ by a multiplicative factor.
limmashrink	A logical. Should we shrink the variance estimates prior to performing gls to get Z1 (TRUE) or not (FALSE)?
likelihood	What should the likelihood be when we estimate Z1? Options are "t" and "normal".

Details

This doesn't work too well. I think the Z's and sigmas need to be estimated jointly.

Author(s)

David Gerard

dcgerard/vicar documentation built on July 7, 2021, 1:08 p.m.