Re: heritability estimation for binary and quantitative traits

Dear Mahood,

In this regard, I have attached two data files,
one for quantitative trait and another is for qualitative trait.
and thereby I have attached my result files also.
Frankly speaking, I am really confused about my next steps:
1. Firstly, I will really be very glad if I am assured that
 my *.as and *.pin files are right on track. i.e. for heritability
  estimation my codes are right in terms of the variance components
  I got in *.asr file and are specified in the *.pin file.

* For body weight, the analysis is fine.
Heritability looks reasonable; coding is fine.
You do not need the * or !M0 on the lines
 id !P
 sire !P dam !P

2. Secondly, I have used those factors and their interactions
 in the model for quantitative trait (body weight) which were
  shown as highly significant ( sex [p<0.0001], birth_year [p<0.0001],
   test_age[p<0.0001], sex*test_age [p<0.01] using SAS proc glm procedure.
- In this case if I use all this factors I get 7 singularities in the design matrix.
- In contrast, if I exclude the interaction (sex*test_age) which is also significant,
 then, I get only 3 singularities cases. So, I am really confused that which model I should choose.
 
 
 The singularities are not a problem. It is just that the model is over parameterised.
  Analysis of Variance NumDF DenDF F_inc Prob
   8 mu 1 108.0 24689.41 <.001

   4 sex 1 849.7 99.17 <.001
   5 birth_year 4 791.2 23.35 <.001
   6 test_age 2 853.9 14.03 <.001
   9 sex.test_age 2 836.1 3.74 0.026

   Sex has 2 levels but 1 DF,
   birthyear has 5 levels but 4 DF
   testage has 3 levels but 2 DF
   sex.testage has 6 levels but 2 DF
   
      Now you report different significance levels from SAS,
      but I do not think it is the same model [here you have the animal model]
      in any case, the first 3 are equivaalent.
      I'm not surprised sex.test_age is slightly different under the animal model.
      
      I expect you will get the same answers from ASReml as from SAS if you fit the same model.
      
      Note there is also an !FCON which will give additional complementary F tests.

[ Actually, how much this differences in singularities may play
 a role in the model output specially when I am getting the convergence
  in every case. I mean, how can I deal the situation like this for
   inclusion or exclusion of factors /interactions?
   In that case the factor was significant
   
   I would leave sex.test in the fitted model though it is now significant at 5% not 1%.
   
     The coding of the binary analysis looks fine too.
     It is harder to formally test fixed effects but I would suggest that only the test_age regression is
     significant.
     
     
     Using an animal model for binary data generally works in the sense that you get an answer.
     Simulation studues though tend to suggest the answer is not what you might predict.
     I'e it tends to underestimate heritabiliy in simulation studies but is very dependent on
     da. If you give a suggestion on my query that will be really appreciable]

3. Thirdly, for the binary analysis , I used !BIN !LOGIT. In this regard, I would like to request you to take a glance on my *.asr, *.pin file for the correctness.
- I used a conversion factor of 3.29 in the pin file for the variance part I got in *.asr file.
-Do you think that this has been done correctly plotted for the right variance ***component ? Moreover, can I draw any valid conclusion on heritability estimates, for a qualitative trait based on binary analysis, specially from my analysis codes?

For

------------------------
Arthur Gilmour

Retired Principal Research Scientist (Biometrics)

-------------------- m2f --------------------

Sent using Mail2Forum (http://www.mail2forum.com).

Read this topic online here:
http://www.vsni.co.uk/forum/viewtopic.php?p=666#666

-------------------- m2f --------------------
Received on Fri Jul 23 2009 - 11:32:00 EST