[R-sig-ME] call for lme4 with given Z?

[Ben Bolker]

Ryan King <c.ryan.king at ...> writes:

> I see that a similar question has been asked in the past, but I am
> not totally clear with the in-development status of lme4 if things
> have changed.
 
> Is there a way to call lmer (and glmer) where I have pre-computed
> random effect design matrices and residual covariances?  My design
> matrices are crossed and don't correspond to any of the easy formula
> interfaces that I know about.  I see in the belly of the lmer code
> where Zt's are stored, but am hesitant to jump in and change them;
> I'm hoping someone familiar with the package has a wrapper for this
> use-case.  I know how to do this in the "regress" package for LMMs,
> but would like to use lme4's superior fast methods for sparse
> matrices and have something reasonable for (binary) GLMMs.  I am
> aware of how to do this in MCMCglmm, but for a particularly
> large-scale project have found it unacceptably slow.

  This still hasn't happened, that I know of: Joehanes Roeby had
a patch that modularized some of this (for use with pedigreemm), at
https://stat.ethz.ch/pipermail/r-sig-mixed-models/2012q2/018199.html

  I am hoping to make a branch like this on github sometime Very Soon
(but past experience suggests you shouldn't hold your breath ...)

  Ben Bolker