[Rd] anova F-tests (PR#8875)

[Thomas Lumley]

On Thu, 18 May 2006, benfitz at utk.edu wrote:

> Full_Name: Ben Fitzpatrick
> Version: 2.3.0
> OS: Windows XP and Mac OSX
> Submission from: (NULL) (160.36.15.155)
>
>
> When you fit a linear model (lm) and use the summary command, the assessment of
> the statistical significance of explanatory variables is very different than if
> you use the anova command. The F statistics calculated by anova are much larger
> than what you get out of SAS, for example. I think R is not calculating F
> statistics correctly.


The anova() function is not supposed to give the same results as 
summary().

You don't say *which* SAS results it is "much larger" than (or even 
provide a single example), but if you mean the most common SAS anova table 
based on "Type III SS" then (a) the anova() results are not supposed to be 
the same and (b) it is not true that they are always "much larger". They 
can be the same or (much) smaller.

If you had read the FAQ you would at least know that the anova() output 
was not supposed to be the same as that from SAS. You could then have 
either asked why it was different or asked how to get the output you 
wanted.

If R were really giving completely wrong answers for a very simple and 
widely used analysis and no-one had noticed, an example would still have 
been needed to make this a useful bug report.


 	-thomas