[R] Adding correlation and or variance structure in mixed models

[Ben Bolker]

Laz <lmramba <at> ufl.edu> writes:

> 
> Dear R users,
> 
> I am using mixed models to analyze genetic experiments. I have tried to 
> use R packages such as lme4, nlme. I am looking for a model that can 
> allow me to specify my correlation structure and the same time allow to 
> have more than 2 random effects in the model. None of the above packages 
> fully answer  my question and some results obtained have incorrect  
> degrees of freedom at some point.
> 
> For example: Y = XB + Z1g + Z2f + e
> where X is the design matrix of fixed effect, B is the vector of fixed 
> effect, Z1 is a random design matrix of the first random variable,g,  
> and g is the vector of the random effect g, Z2 is random design matrix 
> of the second random variable,  and f is a vector of the random effect 
> f., e is the residual error.
> If I need to specify an autocorrelation structure such as AR1 
> (autocorrelation 1st order) or CORG (general correlation) how would I do 
> that in R? the functions provided do not allow to specify this term 
> except the gls but it does not take random effects.
> 
> I am interested in specifying variance structures such as AR1V 
> (autocorrelation 1st order), diagonal, uniform heterogeneous, uniform 
> correlation, Unstrustured etc in my model that has both fixed and 
> several random effects. Which R packages or functions will help me to 
> accomplish this?
> 
> Thanks.
> Laz

   The lme() function in the nlme package allows you to do some but
not necessarily all of this, and is I think the closest you will get.
If by "more than one random effect" you mean *crossed* (as opposed to
nested) random effects, that will be a little more difficult, but it
is possible in lme: see Pinheiro and Bates 2000 p. 163.  

  For more information/detail you should ask this question at
r-sig-mixed-models at r-project.org , but it would be a good idea to
invest in a copy of Pinheiro and Bates 2000 and find out as much
as you can for yourself first.

  Ben Bolker