Hi,
I have fitted quasibinomial GLM [glm(y ~ ..., family = quasibinomial)]
to a binary response variable; quasibinomial, because there were clear
signs of underdispersion in a 'simple' binomial GLM, and so the
dispersion is a free parameter in the model.
My question is now: In a quasi-binomial model with a binary-only
response variable, what are the most appropriate tests to compare
different models? I have studied Faraway's book (Extending the linear
Model with R) and concluded a likelihood-ratio test seems to be
inappropriate, as seems to be the Wald-test. In chapter 7 an F-test is
suggested, but this refers to an example with a beta-distributed response.
Can I conclude that the following code example will be fine in my case:
anova(model1, model2, test= "F") ?
Moreover, the summary of the GLM, including parameters of the
predictors, shall be presented. The summary method however does not
conduct an F test; so in sync with my ideas above, shall I also use
F-tests for the individual predictors (personally, I would, but as I am
not sure I ask here...)?
Thanks a lot,
Thomas
______________________________________________
R-help@r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
and provide commented, minimal, self-contained, reproducible code.
