[R-sig-ME] cox reid likelihood in R

[Ben Bolker]

Akshita Chawla <chawlaak at ...> writes:

> 
> 
> Dear all
> 
> I am trying to compute the Cox reid adjusted likelihood ratio
> test for mixed models in r. The adjustment
> needs the information matrix. Is there a package/function in R
> that runs the test or is there a way to
> extract the information matrix from an lme object?
> 
> Thanks
> Akshita
> 
> 


  Which information matrix?

  Within a merMod object, the "RX" component (getME(model,"RX"))
gives the (unscaled) Cholesky factor of the fixed-effects information
matrix (there's also an "RXi" component which has the inverse,
i.e. the Cholesky factor of the fixed-effects variance-covariance
matrix).  You can take the cross-product of this element to get
the information matrix, scaling by sigma^2 if necessary.  (See
https://github.com/lme4/lme4/issues/47 for some related discussion.)

  However, this is the information matrix *conditional on the
theta (random-effects) parameters*.  A full information matrix including
the uncertainty in the theta parameters is trickier ...

  Ben Bolker