Re: Followed by a B

From: Bruce Southey <southey_at_UIUC.EDU>
Date: Wed, 6 Apr 2005 10:17:32 -0500

>I am dealing with a very stubborn group of small datasets for calculating
>heritabilities in dairy cattle. Sometimes I got from univariate animal
>models variance components followed by a B which means "fixed to a boundary"
>┐?. In all these analyses convergence is OK.

Not really...
ASREML is stopping the variance component from going out of the parameter space
so the likelihood is being maximized in the restricted parameter space where
that parameter is zero. Essentially this means that the variance component was
not fitted in the model.

>┐Should I believe or not these variances? all of them are close to zero. I
>have reasons to think they are really not zero from other sub-analyses.
>Thank you,

The problem is that you lack the statistical information to estimate these
parameters given the data and the model. Just by fitting the random animal
effect and the intercept only should provide some indication of whether it is a
data or model issue.

One reason for the problem with the full model, if the sub-analyses actually
are reasonable, is that two or more terms in the model for the full data are
partially or completely confounded. If it is not a term in the model, then
there is some unobservable stratification and/or variation in the data that you
have to address.

Received on Tue Apr 06 2005 - 10:17:32 EST

This webpage is part of the ASReml-l discussion list archives 2004-2010. More information on ASReml can be found at the VSN website. This discussion list is now deprecated - please use the VSN forum for discussion on ASReml. (These online archives were generated using the hypermail package.)