[R] Type IV sum of squares

Bill.Venables@CMIS.CSIRO.AU Bill.Venables at CMIS.CSIRO.AU
Sat Mar 8 06:41:29 CET 2003

I'm really puzzled to know what has caused the sudden rush on these kinds of

>  -----Original Message-----
> From: 	dth2 at u.washington.edu [mailto:dth2 at u.washington.edu] 
> Sent:	Saturday, March 08, 2003 1:27 AM
> To:	r-help at stat.math.ethz.ch
> Subject:	[R] Type IV sum of squares
> To anyone who can help:
> I am trying to do an analysis of variance with a very unbalanced design
> and a few empty cells.  How do I get R to use type III and more
> importantly type IV sums of squares?
	[WNV]  My friendly advice is forget all that malarkey, and even
forget about putting it all in one big analysis of variance table.  Calmly
decide which null hypotheses you want to test within which outer hypotheses,
fit each null and each outer model, separately if necessary, test them using
anova( ) and report your results.

	Unbalanced designs and empty cells do not alter the general
procedure you should adopt one bit.  All they do is change the power of the
test, a fact of which you should be well aware, of course.  (In fact if the
design is really deficient you may find that you have a reduncancy and
cannot perform some of your tests, but that is a side issue and no 'Types'
of sums of squares can ever fix it - you need a better design to make any
progress.  )

	Hard as it may be to accept, that whole business of Type ? sums of
squares is a gigantic red herring and a distinction without a difference.
It would truly have been better if the concepts were never invented and
certainly not inflicted on a naive and unsuspecting whole statistical
generation.  This is one situation where curiously the reality is much
simpler than it might seem from some of the propaganda.

	Trust me: I'm a statistician...

> Thanks
> Deven Hamilton
> Department of Sociology
> University of Washington
> ______________________________________________
> R-help at stat.math.ethz.ch mailing list
> https://www.stat.math.ethz.ch/mailman/listinfo/r-help

More information about the R-help mailing list