[R] set type of SS in anova()

Prof Brian Ripley ripley at stats.ox.ac.uk
Sat Jul 5 09:04:00 CEST 2008

drop1() in R meets both your objection and that to 'Type III'.

If you really want the so-called Type III, look at function Anova() in 
package car.  But also consider

library(fortunes); fortune(54); fortune((55); fortune(56)

On Fri, 4 Jul 2008, jeroenooms wrote:

> When you use the 'general linear model' analysis in SPSS, the first result is
> a table with all terms with F-tests and significance values for all IV's. It
> uses  http://joyx.joensuu.fi/~ek/anova/sstypes.txt SS Type III , which has
> the advantage that the order in which the variables are added to the model
> does not matter, and therefore it is relatively objective.

No, as the terms may well have a natural hierachy in which case the order 
does matter.

> I would like to reproduce this output in R. However, when using
> anova(glm.object, test="F"), the F test shows "Terms added sequentially
> (first to last)". Because of this, the F values and p values of the terms
> depend on the sequence in which they were added. Eg: an anova() of
> glm(a~b+c) will give other results than glm(a~c+b).

Hmm: glm() is used for generalized linear models, and there sums of 
squares are only appropriate for those models where lm() is more 

> How can I specify the SS type of an anova in R, so that i can reproduce the
> exact results as that i got in SPSS?

Brian D. Ripley,                  ripley at stats.ox.ac.uk
Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
University of Oxford,             Tel:  +44 1865 272861 (self)
1 South Parks Road,                     +44 1865 272866 (PA)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595

More information about the R-help mailing list