[R-sig-ME] question about lme4 and nlme

[Niu, Mu]

Dear Prof Bolker

I am a post-doc from school of public health at imperial college london. I would like to use lme4 and nlme package do some statistical gentics work. I got some problem when I run my lme model.

I would like use the fixed effect model with fixed weights on variance. Can I use random effect random= (~1|family) and weights = varIdent(fix=c(het=2), form = ~ 1 | categories)) at the same time?

As a example, I have n patients, their weights  are the dependent variable, my fixed effect model is
    modelN <- try(lme(Weight ~ fixeEffect, random=(~1|family), data=sim, method="REML", na.action=na.omit), silent=T)

because we have the genotype of patients, we categorized the patients into two group (homo , hete) based their gene expression, and we assume the patient in hete group should have twice bigger variance as homo group. we use varIdent function to present our assumption. Is the syntax right?
modelF <- try(lme(Weight ~ fixeEffect, random= (~1|family), data=sim, method="REML", na.action=na.omit, weights = varIdent(fix=c(het=2), form = ~ 1 | categories)), silent=T)

However,if we do not include sibling into our data and only use one member of the family, (the covariance matrix is diagonal), modelF is better modelN. If we include the sibling data and we have two member from each family( the covariance matrix become block diagonal) , modelF is not better than modelN. I do not know why more information about the family make the model fail?
Thank you very much for your help

Regards
Mu Niu




	[[alternative HTML version deleted]]