# [R] define F-ratio computations with aov

Michael Rennie mdrennie at gmail.com
Wed Mar 17 13:32:18 CET 2010

```Howdy,

In the past, I've just run the ANOVA as normal, and then just grabbed
the appropriate MS for the estimation of F ratios. Eg, this will get you
the MS in your anova object:

summary(obj.aov)[[1]][3]

or

summary(obj.aov)\$Mean

And if you want a specific MS,

summary(obj.aov)[[1]][[1,3]]

or

summary(obj.aov)[[1]]\$Mean[1]

Then you can just put whichever MS over whichever other MS, estimate

Ffact<- summary(obj.aov)[[1]]\$Mean[1]/summary(obj.aov)[[1]]\$Mean[3]

estimate the p-values with:

pFfact<-1-pf(Ffact, summary(obj.aov)[[1]]\$Df[1],
summary(obj.aov)[[1]]\$Df[3])

and you're off to the races.

You can also specify error strata in the aov() model, but then all you
get is the MS and you have to estimate your F-ratios anyway (though the
indexing is a little different). E.g., if you had a nested anova, you
could specify it as:

ex.aov<-aov(Fixed ~ Nested + Error(Nested/Fixed))

At least this way, the summary() doesn't give you the wrong F-ratios, so
you aren't temped to interpret them incorrectly (as you would in the
previous example).

HTH,

Mike

Galanidis Alexandros wrote:
> Greetings to all,
>
> This is my model: aov.fit<-aov(Y~A+B+C+D+E+A:C+A:E)
>
> In summary(aov.fit) all F values are comptuted by eg MS(A)/MS(Residuals). This is not correct (or what I want), except for F(B) and F(A:E). I suppose P values are not correct either.
>
> Is it possible with aov to define the way F computations will be done? I 'd like them to be like this: F(A)=MS(A)/MS(E), F(C)=MS(C)/MS(E), F(D)=MS(D)/MS(E), F(E)=MS(E)/MS(A:E), F(A:C)=MS(A:C)/MS(A:E)
>
> thanks
> ______________________________________________
> R-help at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help