CODEWITHSUNDEEP
r
Michael
Michael

Reputation: 5898

Extracting the name of an interaction variable given a full and reduced model

Given a glm model reduced containing variable of interest varofint and an arbitrary number of additional variables v1 ... v[n] of class numeric, integer, or factor (some of which may be interactions with each other), and a glm model full that contains all terms in reduced as well as an additional term which is the interaction of varofint and exactly one of the additional variables v[i] (this interaction specified using a : or * in the formula for full), how can I extract the name of v[i]? I'd prefer solutions that don't rely on regular expressions but rather use the model structure itself.

In the example code, v[i] is named "interaction_var" and that character vector ("interaction_var") is the return value that I want.

x <- data.frame(outcome=rbinom(100,1,.3),varofint=rnorm(100), interaction_var=sample(letters[1:3],100,replace=TRUE))

reduced <- glm(outcome~varofint+interaction_var,data=x)
full <- glm(outcome~varofint*interaction_var,data=x)

Here's the start of a solution using sub which I can't figure out how to finish and makes some assumptions that I don't like: 

addl_terms <- setdiff(names(coef(full)),names(coef(reduced)))
varname_pluslevel <- sub(".+?:",replacement="", addl_terms[1])

Upvotes: 1

Views: 73

Answers (1)

josliber
josliber

Reputation: 44320

It seems like the term.labels attribute of the terms field in full provides more straightforward access to the interacted variable name:

term.labels.reduced = attr(reduced$terms, "term.labels")
term.labels.full = attr(full$terms, "term.labels")
interaction = setdiff(term.labels.full, term.labels.reduced)
strsplit(interaction, ":")[[1]][2]
# [1] "interaction_var"

Upvotes: 1

Related Questions