[R-sig-ME] estimating variance components for arbitrarily defined var/covar matrices
Jarrod Hadfield
j.hadfield at ed.ac.uk
Fri Feb 27 07:34:34 CET 2015
Hi Matthew,
It depends a bit on the pedigree (is it a pedigree or derived from
markers?) 10K - 100K would take a few seconds to a few minutes in
asreml, but it is often not very good with non-Guassian data. MCMCglmm
(being MCMC based) would probably take half an hour to a half a day
for pedigrees of this size. In both cases it depends on pedigree
structure - big marriage chains can slow it down dramatically. WOMBAT
(which I've never used) is probably comparable to asreml in speed. If
G is derived from markers something like GCTA will perform better, but
this is slow because G is completely dense.
Cheers,
Jarrod
Quoting Matthew Keller <mckellercran at gmail.com> on Thu, 26 Feb 2015
15:05:34 -0700:
> Hi all,
>
> This has been wonderful to follow, thank you very much to all who have
> contributed!!
>
> Quick clarification:
> Z*Z' is fixed/known. VG is unknown and would be estimated from the data.
>
> Another issue:
> The number of individuals fit in these models is often very large (e.g.,
> 10K - 100K) because the variance of the off-diagonals of Z*Z' is tiny. Of
> the above approaches suggested, are any able to work with datasets of this
> size in a 'reasonable' amount of time? E.g., < 1 day?
>
> Best,
>
> Matt
>
> On Thu, Feb 26, 2015 at 2:47 PM, Rolf Turner <r.turner at auckland.ac.nz>
> wrote:
>
>> On 26/02/15 16:54, Ben Bolker wrote:
>>
>>> -----BEGIN PGP SIGNED MESSAGE-----
>>> Hash: SHA1
>>>
>>> I thought we were assuming a fixed var-cov matrix
>>>
>>
>> So Z*Z'*VG is fixed/known, rather than being estimated from the data.
>>
>> That's what I didn't properly apprehend.
>>
>> PLUS an error
>>> variance, i.e. Sigma + s^2*I (increasing the variance and decreasing
>>> the correlation).
>>>
>>> But I could be wrong about what model is intended.
>>>
>>
>> No, I think that the misunderstanding was entirely mine.
>>
>> Sorry for the noise.
>>
>> cheers,
>>
>> Rolf
>>
>> --
>> Rolf Turner
>> Technical Editor ANZJS
>> Department of Statistics
>> University of Auckland
>> Phone: +64-9-373-7599 ext. 88276
>> Home phone: +64-9-480-4619
>>
>> _______________________________________________
>> R-sig-mixed-models at r-project.org mailing list
>> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
>>
>
>
>
> --
> Matthew C Keller
> Asst. Professor of Psychology
> University of Colorado at Boulder
> www.matthewckeller.com
>
> [[alternative HTML version deleted]]
>
> _______________________________________________
> R-sig-mixed-models at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
>
>
--
The University of Edinburgh is a charitable body, registered in
Scotland, with registration number SC005336.
More information about the R-sig-mixed-models
mailing list