function of effect fails with xlevels - r

Question

I am working on developing a function, with the effects package in R. I am constantly running into a problem, which I can't debug.
I run the following code:
rm(list = ls()) # clear working directory
library(effects)
head(mtcars)
mod <- lm(mpg ~ gear*cyl + gear + cyl + carb, data=mtcars)
summary(mod)
eff.dat <- effect("gear*cyl", mod=mod, KR=TRUE, xlevels=list(gear=seq(3,5,1)))
eff.dat <- as.data.frame(eff.dat)
View(eff.dat)
It works like a charm, I get the effects for the interaction term on mpg when gear equals 3, 4, and 5 and the corresponding values of cyl.
However, once I put this into a function like:
proba <- function(term, model, main) {
eff.dat2 <<- effect(term, mod=model, KR=TRUE,
xlevels=list(main=seq(min(mtcars[[main]]),
max(mtcars[[main]]), 1)))
eff.dat2 <<- as.data.frame(eff.dat2)
}
proba("gear*cyl", model=mod, main="gear")
View(eff.dat2)
The xlevels part fails and the interaction term is estimated for the default values of gear, not the ones I specify. Obviously, this is part of a larger function, otherwise I would not bother to write something solely for effect.

Answer 1

First an illustration what happens:
foo <- function(x) {
list(x = x)
}
foo("bar")
#$x
#[1] "bar
Note how the list element is named x. setNames can be used to set names programmatically:
foo <- function(x) {
setNames(list(x), x)
}
foo("bar")
#$bar
#[1] "bar"
Also, you should avoid creating function side effects with <<-. It's very bad practice. Create a proper return value instead:
proba <- function(term, model, main) {
as.data.frame(
effect(term, mod=model, KR=TRUE,
xlevels= setNames(list(seq(min(mtcars[[main]]),
max(mtcars[[main]]), 1)), main))
)
}
eff.dat2 <- proba("gear*cyl", model=mod, main="gear")
all.equal(eff.dat, eff.dat2)
#[1] TRUE