make.varcov: Create moment matrices

In bqtl: Bayesian QTL Mapping Toolkit

Create moment matrices

Description

Create a moment matrix of the marker variables and of the regressors by the phenotype variable. For use in regression modelling on the markers.

Usage

make.varcov(regressor.matrix, y, subset=is.finite(y), casewt=NULL)

Arguments

regressor.matrix	The object produced by make.regressor.matrix
y	A vector of phenotype information with the same number of elements as there are rows in regressor.matrix
subset	Logical vector with the same number of elements as there are rows in regressor.matrix to indicate which rows to keep.
casewt	Optional vector of case weights.

Value

A list with components

var.x	Moment matrix of the marker regressor variables
cov.xy	Moment matrix of the marker regressor variables versus the phenotype variable
var.y	The Second central moment of the phenotype variable
df	sum(subset==TRUE) - 1

Note

It is generally NOT a good idea to do regressions on ill-conditioned designs using the moment matrices like this. The excuse for doing so here is twofold. First, calculations using this method are used to perform importance sampling, so minor numerical inaccuracies in computing the probabilites used in sampling get straightened out by the importance weights. Second, it will typically be the case that a prior is set on the regression coefficients and this results in a positive constant (aka a 'ridge' parameter) being added to diagonal of varcov$var.x and this reduces the ill-conditioning. Of course the rational for using the method is to speed the sampling, and it is very effective at doing so.

Author(s)

Charles C. Berry [e-mail]

bqtl documentation built on Sept. 25, 2024, 1:08 a.m.