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

  ASReml  ~  animal model for survival analysis

emheu
Posted: Tue Jul 21, 2015 2:09 pm Reply with quote
Joined: 25 Mar 2015 Posts: 5
Dear forum members,

I am trying to analyse survival data (coded 1 for stillbirth and 0 for survived) and birth weight with different models (animal and sire model including maternal effects, stillbirth or birth weight as univariate or bivariate stillbirth and birth weight together). With the sire model everything (uni- or bivariate, continuous or binary trait) works very well and as expected concerning LogL convergence and the estimation of variance components. But if I use the animal model with or without maternal effects it unfortunately only works for the univariate analysis of birth weight. When I try to analyse the binary trait it does not converge (not even close) and I get huge variances for animal as random effect and therefore overestimated and very unrealistic heritabilities. I already took care of cleaning up the data set and pedigree file and am currently running out of ideas. Any ideas or expieriences why there could be a problem with the binary variable and the pedigree within the animal model? I have no problem if I include the relationships between the ancestors (within fathers and within mothers) in the model with the help of the pedigree data and the !P qualifier used on animals, fathers and mothers.

Thank you so much in advance for your help!
Kind regards,
emheu
View user's profile Send private message
tbird
Posted: Sun Jul 26, 2015 11:23 pm Reply with quote
Joined: 30 Aug 2013 Posts: 9
hi
I have had very limited experience with binary traits, but you could try a few things that might help pin point the problem, trying simplifying the model including the fixed effects, are you getting boundaries or B codes at all? singularities? have you tried running the univariate with !P just on the animal & not the sire and dam? have you tried analysing it on the underlying scale using the logit function or something similar? not sure if any of this will help Regards tbird
View user's profile Send private message
Arthur
Posted: Mon Aug 10, 2015 1:54 am Reply with quote
Joined: 05 Aug 2008 Posts: 471 Location: Orange, NSW
Dear Hemue,

The problem you have is probably due to the structure of the survival trait in relation to the pedigree. I suppose you have survival to a particular age. If cohorts are affected by a cause of death (e.g. infection, or climatic event, then the animal model will not work because it assumes survival is independent between individuals other than genetic propensity.

Using the underlying (say logit) scale for a binary trait under the animal model will not help as the animal does not work properly for binary data.

If you want me to look at your particular data and attempt a more detailed explanation, contact me directly at arthur@vsni.co.uk

_________________
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