[R] Two-factorial Huynh-Feldt-Test

Peter Dalgaard p.dalgaard at biostat.ku.dk
Mon Feb 21 10:25:52 CET 2005


Bela Bauer <bela_b at gmx.net> writes:

> Peter Dalgaard wrote:
> 
>  >My suspicion would be that it has something to do with calculating the
>  >correction terms before or after contrast transformations (there must
>  >be a coordinate-free version of the corrections?), but I can't grok
>  >the details that easily.
> 
> I must admit that I'm not too familiar with the way R handles these
> calculations internally, so I don't quite see what you mean. Could you
> somehow clarify this, possibly by pointing me to relevant
> documentation?

I was thinking of the way SAS does it. This is unusually unclear in
the documentation (you can say many things about SAS, but
underdocumentation is not normally one of them). The point is that the
G-G formula involving on- and off-diagonal terms of the empirical
covariance matrix cannot be expected to work if you're looking at
something other than the simple intercolumn contrasts (and you had a
nested two-way structure in the columns). 

I've been reading up on the original papers by G+G and importantly the
two papers by Box (1954) whose results they are using. As it turns
out, the fundamental approximation is in terms of eigenvalues, and
generalizes neatly to arbitrary sets of contrasts. All the unintuitive
stuff is really about fleshing this out in common-use cases in times
where practitioners would not be expected to carry out an eigenvalue
computation. The interesting bit is now whether an implementation of
the eigenvalue formula gives the same results as SAS gets. I'll get
there... 

 
>  >However, if you peek over on the R-devel
>  >list, you will see that I was just about to get serious with
>  >programming some of this stuff as methods for the "mlm" class. I think
>  >you just volunteered to test the code...
> 
> 
> I'd be happy to do that. I've subscribed to R-devel following your
> last email, but I can't seem to be able to find the thread that you're
> referring to. Could you forward the relevent messages or the code to
> me?
 
https://stat.ethz.ch/pipermail/r-devel/2005-February/032240.html

(No code as yet, it is just an outline)

-- 
   O__  ---- Peter Dalgaard             Blegdamsvej 3  
  c/ /'_ --- Dept. of Biostatistics     2200 Cph. N   
 (*) \(*) -- University of Copenhagen   Denmark      Ph: (+45) 35327918
~~~~~~~~~~ - (p.dalgaard at biostat.ku.dk)             FAX: (+45) 35327907




More information about the R-help mailing list