[R-sig-ME] lme4 at flexLambda Error

[Steve Walker]

On 2015-02-02 2:20 PM, Ben Bolker wrote:
> Steve Walker <steve.walker at ...> writes:
>
>>
>> Sorry, it's going to be pretty hard to figure this out without a windows
>> machine.  I tried building you a binary with winbuilder, but gave up
>> after it didn't Just Work.  Given that flexLambda is still considered
>> experimental, getting it to work on windows is still pretty far down on
>> the list.
>>
>> The only other thing I can think of is that your R version is pretty
>> old, so maybe updating would help?
>>
>> Good luck!
>>
>> Steve
>
>    For the record, this also appears at
>
> https://github.com/lme4/lme4/issues/230
>
>     We may have a look, but we can't make any promises about when
> we're going to get to it ...

I just had a look.  I'm able to reproduce Bonitta's error message in a 
different context with:

L <- load(system.file("testdata","polytomous_vcov_ex.RData",
                       package="lme4"))
vcov(polytomous_vcov_ex)
Error in .Call(merPredDCreate, as(X, "matrix"), Lambdat, LamtUt, Lind,  :
   Incorrect number of arguments (17), expecting 15 for 'merPredDCreate'

However, the only reason for this failure is that `polytomous_vcov_ex` 
is a glmerMod object fitted with lme4 version < 2.0 (i.e. not 
flexLambda).  In those pre-2.0 versions, merPredDCreate took 17 
arguments, whereas in 2.0 it takes 15.  In particular, the `Lind` and 
`theta` arguments are no longer used in flexLambda.

The reason for this change relates to the whole point of flexLambda, 
which is to allow more flexible parameterizations of the relative 
covariance factor, Lambda.  In pre-2.0 versions (e.g. the current 
version on CRAN), `theta` contains the parameters of `Lambda` and `Lind` 
controls the mapping from `theta` to `Lambda`.  In flexLambda (i.e. 
v2.0), the parameterization is handled on the R-side to make it more 
flexible for users and those interested in extending lme4.

I still don't know exactly how your error arose, but I would need more 
information for that.  In general however, I would just recommend 
keeping your lme4_1.8 and lme4_2.0 sessions very clearly separated.

Steve

> _______________________________________________
> R-sig-mixed-models at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
>
>