[R-sig-ME] Error in variance estimation using bivariate animal model

David Duffy David.Duffy at qimr.edu.au
Fri May 18 03:15:57 CEST 2012


On Tue, 15 May 2012, Stephane Chantepie wrote:

> I am using MCMCglmm function to construct bivariate animal models of 
> bustard sperm production according to age. Until six years old, each age 
> are considered as a trait. After 6 years old, age-class "7-8" and "9-11" 
> are computed to have enough information.

> In a second part, I have tried to estimate covariance between these 
> ages. And I have some problems in estimating variance-covariance for the 
> bivariate models which use the age class "7-8" or "9-11". For example, 
> when I compute the model cbind(age_1,age_9_11), the estimation of 
> variance for age_1 and age_9_11 are similar to estimations of the simple 
> animal models. The estimation of covariance also appears consistent with 
> the others covariance results, so it pretty good. But when I do 
> cbind(age_2,age_9_11), the variance estimation of age_9_11 become really 
> high (out of the range I have found in the simple model). So the 
> covariance estimation appears really high too. I do not really 
> understand why it works when I use age_1 against my age-class and not 
> when I use the age 2 3 4 5 6. The chains seem to converge and there is 
> no autocorrelation

I can only suggest checking the results against another package (given you 
are fitting LMMs). Are these large pedigrees?

> I have just added a ID random parameter when I use a age-class as one of
> bivariate trait (because I have repeated measures).

Is this needed, if you already have rcov=~us(trait):units ?

Just 2c, David Duffy.

-- 
| David Duffy (MBBS PhD)                                         ,-_|\
| email: davidD at qimr.edu.au  ph: INT+61+7+3362-0217 fax: -0101  /     *
| Epidemiology Unit, Queensland Institute of Medical Research   \_,-._/
| 300 Herston Rd, Brisbane, Queensland 4029, Australia  GPG 4D0B994A v



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