I want to automatically determine which coefficients in a lm belong to a factor. So assume that I have the following models:
d <- data.frame(a = gl(4, 2, 16), b = gl(2, 1, 16),
x = runif(16), y = runif(16), Y = runif(16))
l1 <- lm(Y ~ a + b + x + y, data = d)
l2 <- lm(Y ~ x + y, data = d)
Then the names of the coefficients of the first model are as follows:
names(coef(l1))
# [1] "(Intercept)" "a2" "a3" "a4" "b2"
# [6] "x" "y"
Now ideally I would like a function which tells me that a2, a3, a4and b2 are coefficients of dummy-coded factors.
For a model which does not contain any factors (like l2), the output should be NULL.
I had a look at str(l1) and I saw that there is (in case of the presence of factors in the model) a slot xlevels. I could use names(l1$xlevels) to get a list of all factors in the model and then use grep on the names of the coefficients:
names(coef(l1))[unlist(sapply(names(l1$xlevels), function(.) grep(., names(coef(l1)))))]
# [1] "a2" "a3" "a4" "b2"
But this seems to me like a very dirty work-around and won't work as soon as I have similar names in my model:
d$a4 <- runif(16)
l3 <- update(l1, . ~ . + a4, data = d)
names(coef(l3))[unlist(sapply(names(l3$xlevels), function(.) grep(., names(coef(l3)))))]
# [1] "a2" "a3" "a4" "a4" "b2"
Also, a change of the default contrasts will change the name of the dummy coefficients in my models, so even the most elaborate strategy working on the names of the coefficients probably won't work.
Long story short: how do I get a list of all coefficients which belong to a factor?