[R-sig-ME] What does "number of groups < 50"

[Maria Paola Bissiri]

Dear Dr. Curis,
thank you very much. I will run the bootstrap with a higher number of  
simulations.

Unfortunately there is another major problem now. When I try to  
generate the glmer model on Linux again, it does not converge anymore.
I get this error message, which I did not get before:
In (function (fn, par, lower = rep.int(-Inf, n), upper = rep.int(Inf,  :
   failure to converge in 10000 evaluations
The other two models, for which I need to run the bootstrap on Linux,  
too, do not converge on Linux either. All three models still converge  
on Windows.

This happened after our system administrator installed R on two more  
Linux machines upon my request, although I don't see any connection  
with this, the glmer model does not converge on any Linux machine. On  
all four Linux machines and on Windows the versions of R, lme4, Matrix  
and lattice have the same number.
I would be thankful for any suggestion to solve the problem.
I copy sessionInfo() information below.
Kind regards,
Maria Paola

Windows:

R version 3.0.2 (2013-09-25)
Platform: i386-w64-mingw32/i386 (32-bit)

locale:
[1] LC_COLLATE=German_Germany.1252  LC_CTYPE=German_Germany.1252
[3] LC_MONETARY=German_Germany.1252 LC_NUMERIC=C
[5] LC_TIME=German_Germany.1252

attached base packages:
[1] splines   stats     graphics  grDevices utils     datasets  methods
[8] base

other attached packages:
[1] phia_0.1-5      car_2.0-19      lme4_1.0-4      Matrix_1.0-14
[5] lattice_0.20-23 Hmisc_3.12-2    Formula_1.1-1   survival_2.37-4

loaded via a namespace (and not attached):
[1] cluster_1.14.4 grid_3.0.2     MASS_7.3-29    minqa_1.2.1    nlme_3.1-111
[6] nnet_7.3-7     rpart_4.1-3


Linux:

R version 3.0.2 (2013-09-25)
Platform: x86_64-unknown-linux-gnu (64-bit)

locale:
  [1] LC_CTYPE=de_DE.UTF-8       LC_NUMERIC=C
  [3] LC_TIME=de_DE.UTF-8        LC_COLLATE=de_DE.UTF-8
  [5] LC_MONETARY=de_DE.UTF-8    LC_MESSAGES=de_DE.UTF-8
  [7] LC_PAPER=de_DE.UTF-8       LC_NAME=C
  [9] LC_ADDRESS=C               LC_TELEPHONE=C
[11] LC_MEASUREMENT=de_DE.UTF-8 LC_IDENTIFICATION=C

attached base packages:
[1] stats     graphics  grDevices utils     datasets  methods   base

other attached packages:
[1] lme4_1.0-4      Matrix_1.0-14   lattice_0.20-23

loaded via a namespace (and not attached):
  [1] afex_0.6-82       car_2.0-19        coin_1.0-23       grid_3.0.2
  [5] MASS_7.3-29       minqa_1.2.1       modeltools_0.2-21 mvtnorm_0.9-9996
  [9] nlme_3.1-111      nnet_7.3-7        pbkrtest_0.3-7    plyr_1.8
[13] reshape2_1.2.2    splines_3.0.2     stats4_3.0.2      stringr_0.6.2
[17] survival_2.37-4




Zitat von Emmanuel Curis <emmanuel.curis at parisdescartes.fr>:

> Hello,
>
> I'm not a specialist of bootstrap, hence will not give advice about
> using normal approximation for bootstrap confidence interval; however,
> I guess your warnings come from the fact that trying to obtain the
> bilateral 99.9 % confidence interval using percentiles of a n = 1000
> sample (as done with your 1000 bootstrap "iterations") roughly means
> leave one value out (either min or max) and use all remaining values
> as the confidence interval. Hence, either min or max (depending on the
> one that was rejected) is used, leading to using the extreme values as
> quantiles...
>
> In other words, it simply suggests that 1000 iterations gives a too
> small sample to have a good approximation of the 99.9 % confidence
> interval, you should use more bootstrap replications. However, it
> should be ok for the 95 % one, unless may be you have a lot of ties.
>
> Hope this helps,
>
> Emmanuel Curis
>
> On Mon, Oct 28, 2013 at 06:43:58PM +0000, Maria Paola Bissiri wrote:
> « My bootMer script completed the 1000 iterations calculating
> « confidence intervals at the levels 0.95, 0.99 and 0.999 using the
> « methods "norm", "basic" and "perc", but gave the following warnings
> « for each fixed effect:
> « Warning : Basic Intervals used Extreme Quantiles
> « Some basic intervals may be unstable
> « Warning : Percentile Intervals used Extreme Quantiles
> « Some percentile intervals may be unstable
>
> --
>                                 Emmanuel CURIS
>                                 emmanuel.curis at parisdescartes.fr
>
> Page WWW: http://emmanuel.curis.online.fr/index.html

-- 
Dr. Maria Paola Bissiri

TU Dresden
Fakultät Elektrotechnik und Informationstechnik
Institut für Akustik und Sprachkommunikation
01062 Dresden

Barkhausen-Bau, Raum S54
Helmholtzstraße 18

Tel: +49 (0)351 463-34283
Fax: +49 (0)351 463-37781
E-Mail: Maria_Paola.Bissiri at tu-dresden.de
http://wwwpub.zih.tu-dresden.de/~bissiri/index.htm