[R-sig-ME] Incorrect std errors from nlmer?
Douglas Bates
bates at stat.wisc.edu
Tue Mar 31 23:59:45 CEST 2009
On Tue, Mar 31, 2009 at 4:47 PM, Russell Millar <r.millar at auckland.ac.nz> wrote:
> Thanks for that, but I do notice that the point estimates have now changed.
> The laplace method (ADMB), quadrature (SAS), and simulated likelihood
> (Millar, 2004, Aust&NZ J. Stat)
> all produce the same ML point estimates (to at least 4 sig places) that the
> current nlmer is finding.
> Is the development version doing something different?
Yes. It is doing the optimization of the fixed-effects parameters in
a different part of the algorithm. Looks like I need to think that
through more carefully. Rats - it sure seemed like a good idea.
>
> Regards,
>
> Russell Millar
>>
>> On Tue, Mar 31, 2009 at 3:11 AM, <r.millar at auckland.ac.nz> wrote:
>>
>>>
>>> Hi All,
>>>
>>
>>
>>>
>>> I just copied and paste the example nlmer code from the lmer help file
>>> (see below), which fits the logistic to the orange tree data.
>>> The point estimates of the fixed effects look correct, but the s.e.'s
>>> are not (I've fitted the model using three other software). The s.e. of
>>> Asym is way out (it should be about 15), as is the tree effect variance
>>> (should be about 1000)??????
>>>
>>
>> You're right. I was kind of hoping that people wouldn't notice that :-)
>>
>> This is corrected in the development version, which uses a slightly
>> different syntax in the model formula (all fixed-effects terms must be
>> explicitly included). I won't have the development version ready for
>> release soon so I should go back and fix that in the released version.
>> Unfortunately, it will probably be a week before I can look at it.
>> If someone can produce a patch I would greatly appreciate it.
>>
>> Here is what the development version produces.
>>
>>
>>>
>>> (nm1 <- nlmer(circumference ~ SSlogis(age, Asym, xmid, scal) ~ Asym +
>>> xmid + scal + (Asym|Tree), Orange, start = c(Asym = 200, xmid = 725, scal =
>>> 350)))
>>>
>>
>> Nonlinear mixed model fit by the Laplace approximation
>> Formula: circumference ~ SSlogis(age, Asym, xmid, scal) ~ Asym + xmid
>> + scal + (Asym | Tree)
>> Data: Orange
>> AIC BIC logLik deviance
>> 273.2 280.9 -131.6 263.2
>> Random effects:
>> Groups Name Variance Std.Dev.
>> Tree Asym 1000.911 31.637
>> Residual 61.466 7.840
>> Number of obs: 35, groups: Tree, 5
>>
>> Fixed effects:
>> Estimate Std. Error t value
>> Asym 191.06 15.51 12.32
>> xmid 722.61 33.59 21.51
>> scal 344.20 25.94 13.27
>>
>> Correlation of Fixed Effects:
>> Asym xmid
>> xmid 0.373
>> scal 0.353 0.755
>>
>>
>>
>>>
>>> Regards,
>>>
>>> Russell Millar
>>> U. Auckland
>>>
>>>
>>>>
>>>> (nm1 <- nlmer(circumference ~ SSlogis(age, Asym, xmid, scal) ~
>>>> Asym|Tree,
>>>>
>>>
>>> + Orange, start = c(Asym = 200, xmid = 725, scal = 350)))
>>> Nonlinear mixed model fit by the Laplace approximation
>>> Formula: circumference ~ SSlogis(age, Asym, xmid, scal) ~ Asym | Tree
>>> Data: Orange
>>> AIC BIC logLik deviance
>>> 1901 1908 -945.3 1891
>>> Random effects:
>>> Groups Name Variance Std.Dev.
>>> Tree Asym 53985.920 232.349
>>> Residual 52.868 7.271
>>> Number of obs: 35, groups: Tree, 5
>>>
>>> Fixed effects:
>>> Estimate Std. Error t value
>>> Asym 192.04 104.09 1.845
>>> xmid 727.89 31.97 22.771
>>> scal 347.97 24.42 14.252
>>>
>>> Correlation of Fixed Effects:
>>> Asym xmid
>>> xmid 0.053
>>> scal 0.050 0.763
>>>
>>>
>>>
>>>>
>>>> sessionInfo()
>>>>
>>>
>>> R version 2.8.1 (2008-12-22)
>>> i386-pc-mingw32
>>>
>>> locale:
>>> LC_COLLATE=English_New Zealand.1252;LC_CTYPE=English_New
>>> Zealand.1252;LC_MONETARY=English_New
>>> Zealand.1252;LC_NUMERIC=C;LC_TIME=English_New Zealand.1252
>>>
>>> attached base packages:
>>> [1] stats graphics grDevices utils datasets methods base
>>>
>>> other attached packages:
>>> [1] lme4_0.999375-28 Matrix_0.999375-22 lattice_0.17-20
>>>
>>> loaded via a namespace (and not attached):
>>> [1] grid_2.8.1 tools_2.8.1
>>>
>>> _______________________________________________
>>> R-sig-mixed-models at r-project.org mailing list
>>> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
>>>
>>>
>
>
More information about the R-sig-mixed-models
mailing list