[R-sig-ME] Bivariate animal models with both "ill-conditioned G/R structure" and "Mixed model equations singular" errors

Stephane Chantepie chantepie at mnhn.fr
Thu Mar 15 17:22:28 CET 2012


Dear all,

I have tried the different solutions you proposed.

I have tried to scale fixed parameters and the response variables between 0 
and 1 but the problem remained

However, using ASREML, it seems that the problem comes from the residual 
covariance matrix. As supposed by Jarrod the residual covariance matrix is 
singular. I am pretty sure that it is due the structure of my data but I 
really don't know how I can fix this problem.


I use several age classes (which represent my traits) and I have intra-annual 
repeated measurements for several years and for each individual. 
The fixed parameters I use are: 
-tse : time since the last ejaculation collect: it is used to take into 
account the pressure due to the repeated collect.
-joe : day of ejaculation : represents the time since the first ejaculation of 
the year. I use this parameter to take into account the seasonal variation of 
sperm production.
The model is AgeClass1 AgeClass2 ~ at.level(trait,1): tse+ at.level(trait,2): 
tse+ at.level(trait,1): joe + at.level(trait,2): joe, 
random=~us(trait):animal+us(trait):ID+us(trait):Year, rcov = ~us(trait):units


With the data structure I have, it is impossible to have measurements of the 
same individual on the same line (snapshot to help comprehension: 
http://ubuntuone.com/1W19vErC5jUtHqMn1dtmPg ), likely making it impossible to 
estimate a residual covariance. One of the issues is that I can not see really 
how to change the structure of the data so it’s estimable. 
Is there a solution? I’m afraid if I fix the residuals covariance matrix to 0, 
I will inflate Va.

Do you have an idea?

many thanks for your help

stephane




More information about the R-sig-mixed-models mailing list