R-beta: Stars again

[Prof Brian D Ripley]

On 5 Sep 1998, Peter Dalgaard BSA wrote:

> Jim Lindsey <jlindsey at alpha.luc.ac.be> writes:
> 
> > 3. Then, if we think of interactions, all stars for main effects are
> > meaningless because they cannot be removed without destroying the
> > hierarchy of the model.
> 
> True. Well almost. There are actually a couple of cases where it makes
> sense. (y~C+F:C with F a factor and C continuous gives two lines with
> common intercept, e.g.)

That being a sub-model of y ~ C*F ?  Yes, that needs fiddling to 
appear permissible.
 
> >   Apparently my position is evolving to where I believe that the
> > option should be removed entirely, or implemented as a new function
> > based on changes in deviance.

There is already such a function!  I assume that since you mention deviance
you are talking about glm's (although you do not say so: could you PLEASE
TRY to remember to say such things).  In any event, this is what drop1 is
methods are supposed to do, and drop1.glm does do in my R implementation. 
As the result is of class anova, print.anova will supply stars or not from
the global setting. 

Example:

library(MASS)
data(quine)
library(aov) # for now
quine.glm <- glm(Days ~ .^2, poisson, data=quine)
drop1(quine.glm, test="Chisq")
Single term deletions

Model:
Days ~ Eth + Sex + Age + Lrn + Eth:Sex + Eth:Age + Eth:Lrn + 
            Sex:Age + Sex:Lrn + Age:Lrn
        Df Deviance AIC    Dev. change Pr(Chi)  
<none>     1368.7   1404.7   0.000              
Eth:Sex  1 1391.6   1425.6  22.916     1.692e-06
Eth:Age  3 1497.4   1527.4 128.773     < 2.2e-16
Eth:Lrn  1 1374.1   1408.1   5.459     0.019466 
Sex:Age  3 1518.2   1548.2 149.568     < 2.2e-16
Sex:Lrn  1 1368.8   1402.8   0.091     0.762758 
Age:Lrn  2 1380.3   1412.3  11.611     0.003010 

[I've left the stars off, because, Martin, they are one row out:
xr[!is.na(xr)] is not of the right length when there are NAs.]


> (2) Given that we decide that it would be better to use e.g. a display
> that adds joint test statistics for factors and interactions, how many
> *other* packages will break, because they extract information from
> summary() output? (Perhaps not really that many)

None if you do not remove the information that is currently there: summary
methods are supposed (in S) to calculate information for print.summary.xxx
methods to print, and priting can be selective. 

-- 
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 272860 (secr)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595

-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-
r-help mailing list -- Read http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html
Send "info", "help", or "[un]subscribe"
(in the "body", not the subject !)  To: r-help-request at stat.math.ethz.ch
_._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._