[R] Mac/PC differences in lmer results

[Bert Gunter]

Unless there us good reason not to, always cc the list. I have done so here.

The R Installation manual has some info on how to use different BLASes I
believe, but someone with expertise (I have none) needs to respond to your
queries.

On Thu, May 30, 2019 at 7:50 AM Nicolas Schuck <nico.schuck using gmail.com>
wrote:

> I know that it is in use on the Mac, see sessionInfo below. I have to
> check on the Win system. Why would that make such a difference and how
> could I make the Win get the same results as the Unix Systems?
>
> R version 3.6.0 (2019-04-26)
>
> Platform: x86_64-apple-darwin15.6.0 (64-bit)
>
> Running under: macOS Mojave 10.14.5
>
> Matrix products: default
>
> BLAS:   /Library/Frameworks/R.framework/Versions/3.6/Resources/lib/libRblas.0.dylib
>
> LAPACK: /Library/Frameworks/R.framework/Versions/3.6/Resources/lib/libRlapack.dylib
>
> Random number generation:
>
>  RNG:     Mersenne-Twister
>
>  Normal:  Inversion Sample:  Rounding
>
> locale:[1] en_US.UTF-8/en_US.UTF-8/en_US.UTF-8/C/en_US.UTF-8/en_US.UTF-8
>
> attached base packages:[1] stats     graphics  grDevices utils     datasets  methods   base
>
> Thanks, Nico
>
> On 30. May 2019, at 16:34, Bert Gunter <bgunter.4567 using gmail.com> wrote:
>
> The BLAS in use on each?
>
> Bert Gunter
>
> "The trouble with having an open mind is that people keep coming along and
> sticking things into it."
> -- Opus (aka Berkeley Breathed in his "Bloom County" comic strip )
>
>
> On Thu, May 30, 2019 at 5:27 AM Nicolas Schuck <nico.schuck using gmail.com>
> wrote:
>
>> Dear fellow R coders,
>>
>> I am observing differences in results obtained using glmer when using a
>> Mac or Linux computer versus a PC. Specifically, I am talking about a
>> relatively complex glmer model with a nested random effects structure. The
>> model is set up in the following way:
>> gcctrl = glmerControl(optimizer=c('nloptwrap'), optCtrl = list(maxfun =
>> 500000), calc.derivs = FALSE)
>>
>> glmer_pre_instr1 = glmer(
>>       formula = cbind(FREQ, NSAMP-FREQ) ~ FDIST_minz + poly(RFREQ,2) +
>> ROI + (1 + FDIST_minz + RFREQ + ROI|ID/COL),
>>       data = cdf_pre_instr,
>>       family = binomial,
>>       control = gcctrl)
>>
>> Code and data of an example for which I find reproducible, non-negligible
>> differences between Mac/Win can be found here:
>> https://gitlab.com/nschuck/glmer_sandbox/tree/master <
>> https://gitlab.com/nschuck/glmer_sandbox/tree/master>
>> The differences between the fitted models seem to be most pronounced
>> regarding the estimated correlation structure of the random effects terms.
>> Mac and Linux yield very similar results, but Windows deviates quite a bit
>> in some cases. This has a large impact on p values obtained when performing
>> model comparisons. I have tried this on Mac OS 10.14, Windows 10 and Ubuntu
>> and Debian. All systems I have tried are using lme 1.1.21 and R 3.5+.
>>
>> Does anyone have an idea what the underlying cause might be?
>>
>> Thanks,
>> Nico
>>
>>
>>
>>
>>         [[alternative HTML version deleted]]
>>
>> ______________________________________________
>> R-help using r-project.org mailing list -- To UNSUBSCRIBE and more, see
>> https://stat.ethz.ch/mailman/listinfo/r-help
>> PLEASE do read the posting guide
>> http://www.R-project.org/posting-guide.html
>> and provide commented, minimal, self-contained, reproducible code.
>>
>

	[[alternative HTML version deleted]]