[R] Kenward-Roger correction in lme

Peter Dalgaard p.dalgaard at biostat.ku.dk
Mon Feb 25 20:46:10 CET 2008


Ben Bolker wrote:
> stian <at> mail.rockefeller.edu <stian <at> mail.rockefeller.edu> writes:
>   
>> Hi,I am wondering how to conduct Kenward-Roger correction in 
>> the linear mixed model using R. Any idea?
>>
>> Thanks a lot,
>>
>> Suyan
>>     
>
>   Not really possible, I'm afraid.  Having already invested a huge
> amount of time in making lme4 available (and this is the cutting-edge
> version of linear mixed models for R), Doug Bates has declined to
> spend effort implementing K-R because (1) he's not convinced of the
> appropriateness of adjusting F-distribution degrees of freedom in
> this way, (2) he doesn't think that the K-R algorithm will be
> feasible for the sorts of large-data problems he's interested in,
> (3) [can't find the reference] he finds the correspondence between
> K-R's notation and his difficult.
>
>   
The last one could be due to me rather than Doug, and certainly, it is 
just an obstacle rather than anything prohibitive in itself. I tend to 
think that (2) is "nontrivial" rather than "infeasible", but it is true 
that the K-R formulas depend on items like the covariance matrix (say, 
Sigma) for the entire data set, which is at best implicitly available in 
large-data problems. Whether or not it would be possible to work with 
implicit representations of Sigma is what is difficult to see because of 
the notational differences (it's been a while, but my recollection is 
that Doug uses Sigma for a different purpose, and there are a couple of 
similarly confusing notational switches).

The whole K-R theory (and others of its kind) has the same weakness as 
the t-distribution: It relies on 3rd and 4th moments of the data 
distribution, so it is going to be wrong if data are not Gaussian. 
However, it is still useful to know whether the correction is small or 
large under idealized conditions.

>   This should probably go on an FAQ list somewhere:
>
> By the way,  RSiteSearch("Kenward-Roger") would have
> yielded some information ...
>
> http://finzi.psych.upenn.edu/R/Rhelp02a/archive/81969.html
> http://finzi.psych.upenn.edu/R/Rhelp02a/archive/49633.html
>
>   For further information, please see the r-sig-mixed archives.
>
>   cheers
>     Ben Bolker
>
> ______________________________________________
> R-help at r-project.org mailing list
> 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.
>   


-- 
   O__  ---- Peter Dalgaard             Øster Farimagsgade 5, Entr.B
  c/ /'_ --- Dept. of Biostatistics     PO Box 2099, 1014 Cph. K
 (*) \(*) -- University of Copenhagen   Denmark          Ph:  (+45) 35327918
~~~~~~~~~~ - (p.dalgaard at biostat.ku.dk)                  FAX: (+45) 35327907



More information about the R-help mailing list