From: Dave Johnson [mailto:email@example.com]
Sent: Thursday, 26 July 2001 10:34 AM
Subject: hat matrix
I need some clarification on the 'hat' elements output in the .yht file.
Page 19 of the manual defines them as the diagonal elements of W(W'R
(superscript: -1)W)(superscript: -1)W'. This seems fine for a fixed model
for a more general mixed model should it be WC(superscript: -1)W' where C
the coefficient matrix of the MME ?
Page 127 of the manual suggests using sqrt(1-hat) to alter the residuals,
hat values are much greater than unity ! So it would seem that I need to
my hat values by the residual variance ? But even then I get some values
than 1 which doesn't seem right for a projection matrix. I am also using a
weight variable but after further multiplying by the weights I get values
In summary, I want to know how to get from the hat values to the prediction
error variance [var(ehat) = var (y - yhat)] when using a weight variable.
THE FOLLOWING WARNINGS ATTACH TO THIS COMMUNICATION
This material is intended for the named recipient only and has been
solely for the purpose and function of Livestock Improvement Corporation
and may contain information that is subject to the Privacy Act /
/ Privilege / Copyright. Any person other than the named recipient is
from retaining / disclosing / copying / distributing / using all or part of
Asreml mailinglist archive: http://www.chiswick.anprod.csiro.au/lists/asreml