Using confint.gnm() inside a function

[dwoll]

I believe this is related to issue 1. Consider

library(gnm)
f <- function(d) {
    fit <- gnm(Freq ~ vote, eliminate=class, family=poisson, data=d)
    confint(fit)
}
f(as.data.frame(cautres))

For me, this fails because d could not be found - presumably because some function downstream from update.gnm() (called by profile.gnm()) does not search in the correct environment. I'm sorry I'm not able to locate the exact source of the error, but I don't fully understand what's going on with environments in the following evaluated calls to gnm(), gnmTerms() etc.

Thanks for quickly fixing the previous issue and for your ongoing efforts!

[kauedesousa]

Hi @dwoll

Seems that it works well when you avoid the argument "eliminate". Do you have any especial reason for using this argument?

library(gnm)
f <- function(d) {
fit <- gnm(Freq ~ vote, family = poisson, data = d)
confint(fit)
}
f(as.data.frame(cautres))

[dwoll]

@kauedesousa Yes, I need the conditional Poisson regression because I have to stratify for a lot of factors that are, by themselves, uninteresting. However, even without the eliminate option, I still get an error from terms.formula().

[hturner]

Okay, this is going to be harder to get right. This is a workaround for your example:

library(gnm)
f <- function(d) {
    fit <- do.call("gnm", list(Freq ~ vote, eliminate=quote(class), family=poisson, data=d))
    confint(fit)
}
f(as.data.frame(cautres))

Hopefully you can adapt that for your application. If you need to specify the eliminated variable you can do it as follows:

f <- function(d, x) {
    fit <- do.call("gnm", list(Freq ~ vote, eliminate=substitute(x), family=poisson, data=d))
    confint(fit)
}
f(as.data.frame(cautres), class)

I think it will be a little while before I can implement a proper fix for this.