[R-sig-ME] Questions about porting pedigreemm to the new lme4

[Ben Bolker]

Andrzej T Galecki 1 <agalecki at ...> writes:

> Ben prepared getME(} function to extract various components of the model
> fit
> obtained using lme4.0 and new lme4.
> 
> Try 
> getME(lmf,"Zt")
> getME(lmf,"A")
> getME(lmf,"flist")
> 
> Andrzej
> 
> On Mon, 2 Apr 2012 14:27:20 -0400, "Joehanes, Roby (NIH/NHLBI) [F]"
> <roby.joehanes at ...> wrote:
> > Hi:
> > 
> > I am trying to port pedigreemm to use the new lme4. However, the new lme4
> > is somewhat opaque to me and hence I need help. Specifically, how can I
> > extract the following terms:
> > lmf$FL$trms[[tn]]$Zt
> > lmf$FL$trms[[tn]]$A
> > 
> > The information lies within lmf <at> pp object, but it is too opaque to me to
> > fish anything out.
> > 
> > I understand that lmf$FL$fl is lmf <at> flist and that lmer_finalize calls
> > should be replaced with optwrap and mkMerMod. Nevertheless, I think that
> > the lmer or glmer could use some refactoring to ease the calls.
> > 
> > I appreciate any help on these.
> > 

   A couple of questions:

 1. how broadly are you using A?  Off-list, Doug Bates has commented to
me that defining A is relatively straightforward for LMMs but not necessarily
as simple (nor uniquely defined!) for [GN]LMMs -- for the time being, I
am quite likely to add code that *disallows* getME(.,"A") except for LMMs --
I can imagine that the bulk of pedigreemm use is for LMMs, but I suppose
that people are using it for [GN]LMMs as welll ...

  2. I can't quite tell from your e-mail -- would it be useful to have
the stuff between the first optwrap() and the last mkMerMod abstracted
into a single (exposed) function?  For [GN]LMMs it should probably
also allow control of whether the preliminary "nAGQ=0" optimization
is done or not ...

  (All comments are mine alone, not necessarily speaking for all lme4
authors ...)

  Ben Bolker