[R-sig-ME] "parm" parameter in lme4::confint.merMod

[Ben Bolker]

On 14-05-19 02:24 PM, Jake Westfall wrote:
> Thanks Ben, it looks like a slight tweaking of your command should
> work:
> 
> c(names(getME(sig0,c("theta"))), names(fixef(sig0)))
> 
> Although (not sure if this is the right thread for this but here
> goes) I'm having trouble verifying that the order matches because,
> from what I can tell, confint.merMod() does not appear to work for
> profile confidence intervals for models that include covariance
> parameters between random terms. I always get an error about
> "Starting values violate bounds" and additionally some warnings about
> non-monotonic profiles. If I fix the covariance at 0 then
> confint.merMod() works fine for profile confidence intervals. I could
> send more info if you'd like, but it sounds like maybe you are aware
> of this already. Jake


  I'm vaguely aware of it, but it would be great if you were able to add
an issue (with simple reproducible example) at
https://github.com/lme4/lme4/issues ...

  cheers
    Ben


>> Date: Sun, 18 May 2014 10:21:00 -0400 From: bbolker at gmail.com To:
>> henrik.singmann at psychologie.uni-freiburg.de;
>> r-sig-mixed-models at r-project.org Subject: Re: [R-sig-ME] "parm"
>> parameter in lme4::confint.merMod
>> 
>> On 14-05-18 07:34 AM, Henrik Singmann wrote:
>>> Hi Jake,
>>> 
>>> I think fixef() or names(fixef()) gives you what you want:
>> 
>> I don't think it gets you everything.  Your solution works in the
>> method="Wald" case because method="Wald" only returns the fixed
>> effects, so one only has to keep track of those positions.
>> method="profile" and "boot" include both variance-covariance and
>> fixed effect parameters, in that order. Some possibilities are
>> 
>> names(unlist(getME(m1,c("theta","fixef")))) ## this gives names of
>> 'theta' parameters (i.e. Cholesky factors) ## rather than standard
>> deviations/correlations, but I believe ## the order is the same
>> 
>> ## theta parameters only (hidden function) lme4:::tnames(m1)
>> 
>> require(lme4) data(md_16.4, package = "afex") m1 <-  lmer(induct ~
>> cond*cog + (cog|room:cond), md_16.4)
>> 
>> ## I actually had some problems with profiling this example -- ##
>> not quite sure why.  Will have to look into it. pp <- profile(m1)
>> 
>> 
>> confint(m1,parm=3)
>>> fixef(m1) ## (Intercept)       cond1         cog   cond1:cog ##
>>> -6.14153043 10.32001485  0.66722061 -0.02520666
>>> 
>>> confint(m1, parm = c(3,2), method="Wald") ##             2.5 %
>>> 97.5 % ## cog     0.2498764  1.084565 ## cond1 -18.4845630
>>> 39.124593
>>> 
>>> I hope I didn't misunderstand your question, Henrik
>>> 
>>> 
>>> Am 18.05.2014 08:03, schrieb Jake Westfall:
>>>> Hi all,
>>>> 
>>>> According to the documentation, in confint.merMod() you can use
>>>> the "parm" parameter to get confidence intervals for only a
>>>> specified subset of parameters by indicating the integer
>>>> positions of the desired parameters. But how is one supposed to
>>>> know which integer positions correspond with which parameters?
>>>> I poked around a little but found nothing in the methods or
>>>> slots of the merMod object that indicates this, and nothing in
>>>> the documentation detailing this. Have I missed something? I
>>>> guess you can call the function leaving that argument at its
>>>> default, and the order of the parameters in that output will
>>>> correspond with the integer positions, but this approach would
>>>> seem to defeat the point of being able to specify particular 
>>>> subsets of parameters...
>>>> 
>>>> Jake [[alternative HTML version deleted]]
>>>> 
>>> 
>> 
>> _______________________________________________ 
>> R-sig-mixed-models at r-project.org mailing list 
>> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
>  [[alternative HTML version deleted]]
> 
> _______________________________________________ 
> R-sig-mixed-models at r-project.org mailing list 
> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
>