[R-sig-ME] R-sig-mixed-models Digest, Vol 85, Issue 6

[Ben Bolker]

On 14-01-08 05:13 AM, Cole, Tim wrote:
> Well spotted! This, it turns out, explains the fitting problems. I feel
> a bit foolish not noticing it myself. 
> 
> I'm not sure if [gn]lmer should flag or ban duplicate measurements, as
> in principle they could be legitimate (though not in my case).
> 
> Many thanks,
> Tim Cole

   In any case it still shouldn't crash with an obscure error message!
   I can imagine this issue *might* have been fixed by some of our
'clamping' fixes, i.e.

https://github.com/lme4/lme4/commit/c51a142be1781520f8bd1d7bb6aee27bf4e96568

  If you have a chance it would be great if you could test with a
development version of lme4 and see if the problem goes away ...

  Ben Bolker

> ---
> Tim.Cole at ucl.ac.uk <mailto:Tim.Cole at ucl.ac.uk> Phone +44(0)20 7905 2666
> Fax +44(0)20 7905 2381
> Centre for Paediatric Epidemiology and Biostatistics
> UCL Institute of Child Health, London WC1N 1EH, UK
> 
> From: Emmanuel Curis <emmanuel.curis at parisdescartes.fr
> <mailto:emmanuel.curis at parisdescartes.fr>>
> Date: Wednesday, 8 January 2014 05:46
> To: Tim Cole <tim.cole at ucl.ac.uk <mailto:tim.cole at ucl.ac.uk>>
> Cc: Ben Bolker <bbolker at gmail.com <mailto:bbolker at gmail.com>>,
> "r-sig-mixed-models at r-project.org
> <mailto:r-sig-mixed-models at r-project.org>"
> <r-sig-mixed-models at r-project.org <mailto:r-sig-mixed-models at r-project.org>>
> Subject: Re: [R-sig-ME] R-sig-mixed-models Digest, Vol 85, Issue 6
> 
> Hi,
> 
> I'm not sure if this may be related or not to the issue, but in the
> subset you sent you seem to have duplicate rows, differing only by
> their row name (see for instance 3090 & 3697, 3156 & 3763). I guess
> this is because some columns were removed, but could it be related to
> the difficulties to fit, and the oddities of error occuring or not
> according to the subset ?
> 
> Best regards,
> 
> On Tue, Jan 07, 2014 at 02:12:13PM +0000, Cole, Tim wrote:
> « The dataset has 5048 records and 607 subjects. I've experimented with
> subsets of the data and it is rather complicated - I'll send you the
> data and code separately. The error message is simple to generate with a
> trivial dataset, restricted to subjects with 2 to 4 records and both 0
> and 1 values for mat. But there is also a problem with subjects that
> have 10 records, though not 9 or 11 ? very odd.
> «
> « mat is a binary mature bone measure, which for each subject consists
> of zero or more 0s followed by zero or more 1s, the 0/1 transition
> providing useful information about the age of maturity.
> 
> « On 14-01-06 04:04 AM, Cole, Tim wrote:
> « The dataset consists of longitudinal measures of bone score during
> « puberty, where 1000 indicates maturity. The aim is to estimate median
> « age at maturity in four groups by sex and ethnicity.
> « This glm code works fine, but ignores the longitudinal element. lm2
> « <- glm(I(RUSBoneScore == 1000) ~ log(DecAge) + Sex * Ethnicity,
> « family=binomial, data=na.omit(bh[, 3:6]))
> « This code, which adds a random subject effect, fails with the
> « unhelpful error message below. lm3 <- glmer(I(RUSBoneScore == 1000) ~
> « log(DecAge) + Sex * Ethnicity + (1 | BHID), family=binomial,
> « data=na.omit(bh[, 2:6]))
> « Error in pwrssUpdate(pp, resp, tolPwrss, GHrule(0L), compDev,
> « verbose) : c++ exception (unknown reason)
> « I've tried various alternatives but they all fail in the same way.
> « Thoughts please.
> «
> «     Hard to say without a reproducible example. My main comment is that
> « this is *not* a problem I have seen anyone report before.  sessionInfo()
> « please?  Can you try with verbose=100? Can you make a reproducible
> « example that is a reasonable size and doesn't have confidentiality
> problems?
> «
> «   Ben Bolker
> «
> « _______________________________________________
> « R-sig-mixed-models at r-project.org
> <mailto:R-sig-mixed-models at r-project.org> mailing list
> « https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
> 
> 
> -- 
>                                 Emmanuel CURIS
>                                 emmanuel.curis at parisdescartes.fr
> <mailto:emmanuel.curis at parisdescartes.fr>
> 
> Page WWW: http://emmanuel.curis.online.fr/index.html
> 
>