R help - Apr 2006 - recommendation for post-hoc tests after anova

Greetings all,

I've done some ANOVAs and have found significant effects across my
groups, so now I have to do some post-hoc tests to ascertain which of
the groups is driving the significant effect. Usually one would do
something like a Newman-Keuls or Scheffe test to get at this but based
on googling and browsing through the r-help archives R doesn't support
these and they seem to be badly received by the R community generally
(for reasons related to alpha not being properly controlled, if I
understand correctly).

I found the TukeyHSD function but on reading the help page, I'm unsure
whether it's safe for me to use it since the number of subjects in my
groups is not balanced (ctrl=6, short=9, long=9). Would you reckon
that this is "mildly unbalanced" in the spirit of the help page or
that is my caution warranted?

In the absence of NK or Scheefe tests could someone recommend an
appropriate test for me to conduct in this instance? From what I've
read in Explaining Psychological Statistics, the REGWQ test seems to
be the way to go. Can R perform this test?

Thanks for your time,

Regards,

--
Dr Colm G. Connolly
School of Psychology and Institute of Neuroscience
The Lloyd Building
University of Dublin
Trinity College, Dublin 2, ?ire

You might want to take a look at the multcomp package, which does it in more
modern fashion.  The "tukey" or "dunnett" options there
refer to the types
of comparisons ("Tukey" means all pairwise, "Dunnett" means
all vs. control)
rather than the named procedures.  The package has a vignette that ought to
be helpful to you.

Andy

From: Colm Connolly
> 
> Greetings all,
> 
> I've done some ANOVAs and have found significant effects 
> across my groups, so now I have to do some post-hoc tests to 
> ascertain which of the groups is driving the significant 
> effect. Usually one would do something like a Newman-Keuls or 
> Scheffe test to get at this but based on googling and 
> browsing through the r-help archives R doesn't support these 
> and they seem to be badly received by the R community 
> generally (for reasons related to alpha not being properly 
> controlled, if I understand correctly).
> 
> I found the TukeyHSD function but on reading the help page, 
> I'm unsure whether it's safe for me to use it since the 
> number of subjects in my groups is not balanced (ctrl=6, 
> short=9, long=9). Would you reckon that this is "mildly 
> unbalanced" in the spirit of the help page or that is my 
> caution warranted?
> 
> In the absence of NK or Scheefe tests could someone recommend 
> an appropriate test for me to conduct in this instance? From 
> what I've read in Explaining Psychological Statistics, the 
> REGWQ test seems to be the way to go. Can R perform this test?
> 
> Thanks for your time,
> 
> Regards,
> 
> --
> Dr Colm G. Connolly
> School of Psychology and Institute of Neuroscience
> The Lloyd Building
> University of Dublin
> Trinity College, Dublin 2, ?ire
> 
> ______________________________________________
> R-help at stat.math.ethz.ch mailing list 
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide! 
> http://www.R-project.org/posting-guide.html
> 
>