On Wed, 4 Apr 2001 tr217 at cam.ac.uk wrote:
> Hi, can anyone help with this:
>
> I am trying to analyse some data in the form of proportions with the glm
> function in R and S-plus. When comparing different models with an F test,
> I get different results from R and S-plus. Here's an example (there
are
> two factors and an interaction in the full model
> "glm1<-glm(resp~time*set,family=binomial"):
>
> In R, entering "anova(glm1)" I get this:
Don't think so. You specified an inappropriate test: the default is NULL.
See ?anova.glm for what actually happens.
> DF Deviance Resid. DF Resid. Deviance F Pr(>F)
> NULL 51 213.13
> time 1 1.403 50 211.727 1.4032 0.23618
> set 6 98.527 44 113.2 16.4211 <2e-16***
> time:set 6 13.659 38 99.541 2.2765 0.03368*
>
>
> But in S-plus, I get this:
>
> DF Deviance Resid. DF Resid. Deviance F Pr(>F)
> NULL 51 213.13
> time 1 1.403 50 211.727 0.6749 0.4165
> set 6 98.527 44 113.2 7.8986 0.000015***
> time:set 6 13.659 38 99.541 1.0950 0.3831
>
> Everything is the same, except for the F ratios. Does anyone know why (as
> you can see it makes quite a difference to the significance of the
> interaction term)?
Yes. R uses a binomial as you requested. S-PLUS uses an overdispersed
binomial, which you did not request. If that is what you wanted, specify
quasibinomial.
IMO, the author of this in S (and hence S-PLUS) was somewhat confused.
--
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
_._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._