[R] setting value arg of pdSymm() in nlme

William Valdar valdar at well.ox.ac.uk
Tue May 17 16:22:07 CEST 2005


> I'm afraid that I don't understand what you are trying to do.  With a
> formula of ~ 1 the pdSymm generator creates a 1x1 variance-covariance
> matrix, which you are initializing to a 3x3 matrix.

Oh... I had a feeling I was doing something wrong there.

> What is batch.mat supposed to represent?

I would like to use batch.mat to specify a correlation structure for the 
batches A, B and C. Specifically, I wish to work out the contribution to 
the variance of the batch random effect, given that I know some pairs of 
batches (eg, A and B) are going to be more similar than other pairs (eg, A 
and C) and how similar they are likely to be. My (mis?)reading of P&B p165 
suggested this may be possible turning some of the pdIdents into 

I was hoping to use this test example as a prelude to using genetic 
relationship data to impose a correlation structure on a subject-level or 
family-level effect. I understand from an earlier post (Jarrod Hadfield, 
2003) that lme is not really optimized for this, but I would nontheless 
like to evaluate to what extent it can do it anyway. The example used in 
that post was extreme: each case was effectively a different realization 
of a random effect where the correlation between levels is known. My needs 
are simpler: in the example I gave above, batches A, B and C might 
represent three families related by different degrees.

> Yes, although I think you mean lmer in lme4.  Because the lmer function
> allows multiple nested or non-nested grouping factors, the need for the
> pdMat classes is eliminated (or greatly reduced) and the code can be
> simplified considerably.  There is an article in the 2005/1 issue of R
> News describing the use of lmer.

Thanks for pointing this out as I had missed this article. I know lmer is 
under development and am very interested in what it can do. Having found 
lme/lmer useful for more standard problems I would like to use it for 
genetic-type analysis where possible rather than resort to a different 
language (eg SAS) or specialized proprietary software (eg, ASREML). 
However, I understand that may not be possible.


Dr William Valdar               ++44 (0)1865 287 717
Wellcome Trust Centre           valdar at well.ox.ac.uk
for Human Genetics, Oxford      www.well.ox.ac.uk/~valdar

More information about the R-help mailing list