patch for pedigremm

Hi Roby,

Thank you for the reply. Yes the model is exactly as you describe it (i.e.,
length(unique(dat$id)) == nrow(dat)).

If I install your versions of lme4 and pedigreemm, and in doing so overwrite
the existing versions, can I then overwrite your versions (should I need the
old versions for some reason) by reinstalling lme4 from the cran web site?

If this is the case, could you please send me your pre-patched versions of
lme4 and pedigreemm and some instructions on how to get them going?

Thank again for your help
Regards
David  Aguirre

-----Original Message-----
From: Joehanes, Roby (NIH/NHLBI) [F] [mailto:roby.joehanes at nih.gov]
Sent: Wednesday, 18 July 2012 1:09 AM
To: David Aguirre-Davies
Cc: r-sig-mixed-models at r-project.org
Subject: Re: patch for pedigremm

Hi David:

I apologize for the late reply.

The patch is meant to be implemented at the source code, not at R console.
That is, you need to download the source from the code tracker and retrieve
the correct revision number. Then, you can apply the patch. You will need to
patch both your lme4 and your pedigreemm to make it work. I have a pre-patched
versions, let me know if you want them. Note that if you install my version of
lme4 and pedigreemm, your version of these packages will be overwritten. Also,
I only tested them in Linux. I am not aware of whether they would work on
Windows or Mac.

What the patch is doing is actually to enable 1 observation per subject, which
otherwise not available in the regular pedigreemm. In addition, I incorporated
the latest advances of lme4 as the pedigreemm's back end.

That being said, if the matrix A you mentioned is the relationship matrix, it
should always be square regardless of the length of the IDs (or the
discrepancies thereof). In pedigreemm, the matrix A will be constructed
automatically from the pedigree (i.e., Ped.0) in a function called
"relfactor". This same function will subset the matrix A automatically to
match the IDs that appear in your data. So, there is no need of extra work or
even my patch for this type of problem (i.e., length(dat$id) != length(id)).
The way you specified the ID discrepancies (i.e., length(dat$id) = dim(A)[1] &
length(id) = dim(A)[2]) is a bit unclear to me, and I take it to mean as
(length(dat$id) != length(id)).

Looking at the error message that you presented, however, it seems that there
is one ID per observation (i.e., length(unique(dat$id)) == nrow(dat)). That is
the use case scenario where you would need my patch.

Hope this helps,
Roby

On 6/26/12 11:50 PM, "David Aguirre-Davies" <d.aguirre at uq.edu.au> wrote:


Hi Roby,

Sorry for my incompetence, but running patches on packages is beyond my R
knowledge at this stage. I would like to run the following model using
pedigreemm and I get the error message below,

fm1 <-pedigreemm(value.p ~ gen + (1|id), data=dat, pedigree =
list(id=Ped.0))

Error in function (fr, FL, start, REML, verbose)  :
  Number of levels of a grouping factor for the random effects must be less
than the number of observations

I have followed the clues on the web, and I have attempted to implement the
patch you have provide
(https://r-forge.r-project.org/tracker/index.php?func=detail&aid=1928&group_id
=60&atid=300), but unfortunately I have been unsuccessful.

So my question is, how can I implement the patch you provide to run an animal
model where length(dat$id) = dim(A)[1] & length(id) = dim(A)[2]?

Is there are trick to implementing the patch as cut and paste into the R
console doesn't seem to work for me.

Any assistance would be much appreciated.

Regards
David Aguirre