[vsni.co.uk] Contact us
Author Message
Post new topic

  ASReml  ~  Bug in ASREML3 pedigree analysis if genetic groups aliased?

Dorre
Posted: Mon Dec 26, 2016 12:58 am Reply with quote
Joined: 21 May 2010 Posts: 7
In ASREML-R, I fitted the model:
asreml(trait~covariates,~Anim+ped(AnID)+Batch+Chamb+date, ginverse=list(AnID=ginv),data=Dat)

The above was fitted 1) with and 2) without Anim (the permanent environment effect of animal). I was surprised to find that the likelihood from the model without Anim was higher than when it was included in the model.

I therefore used initial values from model 2, plus a small positive value for Anim, only to see that likelihood for these values was lower than expected from the analysis of model 2.

I then tried it in the stand alone version, obtaining similar results in ASREML3, but a much more plausible result in ASREML4 that indicated a significant permanent environmental effect of Anim.

This suggests that ASREML3 has a bug that has now been fixed. The dataset in question is somewhat unusual in that there is some aliasing between the effect of test batch and the genetic groups. Is this the cause of the problem? If so, is there any way of fixing or getting round it in ASREML-R?

Thanks!
View user's profile Send private message
Dorre
Posted: Tue Dec 27, 2016 10:06 pm Reply with quote
Joined: 21 May 2010 Posts: 7
On more detailed checking, it became apparent that this problem also affects ASREML4.

To obtain likelihoods for the ratio test, I comparing converged likelihoods of the full model with those from the first iteration with initial values from the analysis without Anim plus a very small positive value for Anim.

Wishing all ASREML forum users a happy, healthy and productive New Year!
View user's profile Send private message
Arthur
Posted: Wed Dec 28, 2016 4:12 am Reply with quote
Joined: 05 Aug 2008 Posts: 450 Location: Orange, NSW
Dear Dorothy,

The problem is that you also need to specify !LAST Animal 14
to make sure that any singularities in the Animal term (genetic groups) actually occur in the group equations. The ordering means that non group equations in the linearly dependent set are being declared singular. The difference between ASReml 3 and 4 is that they use different ordering and so end up declaring different animals singular.

The !LAST qualifier is available in ASReml 3 and 4 and should resolve the problems.

_________________
Arthur Gilmour

Retired Principal Research Scientist (Biometrics)
View user's profile Send private message Send e-mail Visit poster's website

Display posts from previous:  

All times are GMT
Page 1 of 1
Post new topic

Jump to:  

You cannot post new topics in this forum
You cannot reply to topics in this forum
You cannot edit your posts in this forum
You cannot delete your posts in this forum
You cannot vote in polls in this forum
You can attach files in this forum
You can download files in this forum