[R-SIG-Mac] nls bug on Intel Mac (or is it me?)

[Joerg van den Hoff]

Karl W Broman wrote:
> I'm finding some odd behavior in nls.summary() on my MacBookPro.
> Consider the following code:
> 
> thedata <- data.frame(x=c( 4,  6,  8, 10, 12,
>                            14, 16, 18, 20, 22),
>                        y=c(21.0, 21.7, 21.7, 22.0, 23.1,
>                            22.8, 23.2, 23.1, 23.7, 24.3))
> 
> summary(nls(y ~ a + b * (1 - exp(d*x)),
>              start = list(a=20, b=60, d=-0.003),
>              data = thedata))
> 
> In the R GUI on the Mac, I get the following results:
> 
>    Estimate Std. Error t value Pr(>|t|)
> a 20.31317    0.56571  35.907 3.37e-09
> b 11.20556   18.93357   0.592    0.573
> d -0.01851    0.04064  -0.455    0.663
> 
> If I run R from the command line, the SEs for a and b give relatively
> random results; if I run the code twice in a row fast, I get the same
> results (though they differ from the above).  If I wait a bit and run
> it again, I get different results.  Here are two examples:
> 
>    Estimate Std. Error t value Pr(>|t|)
> a 20.31317   10.37548   1.958   0.0911
> b 11.20556    7.82498   1.432   0.1952
> d -0.01851    0.04064  -0.455   0.6625
> 
>    Estimate Std. Error t value  Pr(>|t|)
> a 20.31317    4.99515   4.067   0.00477
> b 11.20556    1.33990   8.363 0.0000686
> d -0.01851    0.04064  -0.455   0.66254
> 
> Note that with the same version of R in Linux, I get the (apparently
> correct) result that the R GUI gives.
> 
> Does anyone have any idea what might be happening?
> 
> My version of R is the following (and I'm running MacOS ver 10.4.6):
>                 _
> platform       i386-apple-darwin8.6.1
> arch           i386
> os             darwin8.6.1
> system         i386, darwin8.6.1
> status
> major          2
> minor          3.1
> year           2006
> month          06
> day            01
> svn rev        38247
> language       R
> version.string Version 2.3.1 (2006-06-01)
> 
> karl
> ----------
> Karl W. Broman                  phone   410-614-9408
> Department of Biostatistics     fax     410-955-0958
> Johns Hopkins University        email   kbroman at jhsph.edu
> http://www.biostat.jhsph.edu/~kbroman
> http://www.gnu.org/philosophy/no-word-attachments.html
> 
> _______________________________________________
> R-SIG-Mac mailing list
> R-SIG-Mac at stat.math.ethz.ch
> https://stat.ethz.ch/mailman/listinfo/r-sig-mac


on a G5 this behaviour is _not_ found (using 2.2.0., I admit...):

I copied your code from the mail window into a Terminal and ran it a few 
times. I got always your "GUI" results. just as a info.

by-the-by: the model is no good for the data: the exponential term is 
completely unidentifiable in the data  which essentially follow a 
straight line (the result for `d' of 400% std. dev. shows this as well).