lme4Eigen's bootMer & installing latest svn

Ben Bolker <bbolker at ...> writes:

Mike Lawrence <Mike.Lawrence at ...> writes:

I'm playing with lme4Eigen (version 0.9996875-8, running on Mac OS
10.7.3 using R 2.14.2) and am quite excited by the new bootMer()
function. However, when I try to run it, regardless of what fit I
provide for the argument "x" or function I provide for the argument
"FUN" (including running the examples), I get the error:
Error in envRefInferField(x, what, getClass(class(x)), selfEnv) :
  "resp" is not a valid field or method name for reference class ?lmerResp?
I presume that this is why the bootMer documentation example section
says "## Not run: %%--- fails for now --- FIXME"? I just thought I'd
double-check.

  More recently (version 12) this should work ...

Also, I thought I'd make sure the devs know that the latest svn
version doesn't build on mac; when I try to do so, I get the error:
glmFamily.cpp: In member function ?virtual const Eigen::ArrayXd
glm::negativeBinomialDist::variance(const Eigen::ArrayXd&) const?:
glmFamily.cpp:228: error: ?Rcout? is not a member of ?Rcpp?

 Thanks for the heads-up.  I will check into it and try to see about
getting new binary versions of RcppEigen and lme4Eigen up on the
repository -- although possibly not before Monday.
  cheers
    Ben

I was able to get lme4Eigen_0.9996875-13  to compile this (Monday)
morning on my Mac from source after first compiling version 0.2.0
of RcppEigen, which is available on CRAN but I don't see it on Rforge.
(R version 2.14.2 Patched (2012-02-29 r58552)
Platform: x86_64-apple-darwin9.8.0/x86_64 (64-bit)
Lion 10.7.3).

On an issue from a while back (and going off subject), 
I'm pleased to say that the links from
the psyphy package that allow non-zero lower asymptotes for binomial
families seem to  work with lme4Eigen, at least for binomial aggregated
data.  So far, I get errors for binary responses.  For example to use the
mafc.probit link for a 4-alternative forced-choice experiment, where
one might want to limit the lower asymptote of the link function to 0.25,
I first do the following:

Bi4 <- glmFamily$new(family = binomial(mafc.probit( 4 )))

and then use the argument 

family = Bi4$family

in the arguments to glmer.

It seems to produce promising results in simulated data when I aggregate the
binary responses but when I try it with a binary response variable, I get:

Error in FUN(1:3[[1L]], ...) : Downdated VtV is not positive definite

I would be happy to share the simulation script, if anyone is interested.

Thanks.

Ken Knoblauch
Inserm U846
Stem-cell and Brain Research Institute
Department of Integrative Neurosciences
18 avenue du Doyen L?pine
69500 Bron
France
tel: +33 (0)4 72 91 34 77
fax: +33 (0)4 72 91 34 61
portable: +33 (0)6 84 10 64 10
http://www.sbri.fr/members/kenneth-knoblauch.html