I am trying to get the input variable names out of the model object returned by the lm() function. I tried to access the attribute 'variables' in under lm_obj$terms. However, the returned object is a 'language' type object rather than a regular vector of names.
For example:
lm_obj = lm(y ~ x + z + z:x, data=df)
attr(lm_obj$terms, 'variables')
# list(x, z)
What is a 'language' type? How to convert this 'language' type object to a regular vector like c('x', 'z')?
You are on the correct track. "terms" object is where you should look at. If you want to omit the response variable, you can use delete.response.
all.vars(delete.response(terms(lm_obj)))
#[1] "x" "z"
I would also like to point you to
labels(terms(lm_obj))
#[1] "x" "z" "x:z"
which is sometimes more useful.
A reproducible example to complement your question
df <- data.frame(y = rnorm(20), x = rnorm(20), z = rnorm(20))
lm_obj <- lm(y ~ x + z + z:x, data = df)
To see why we should look at "terms" than elsewhere, you may try different answers here on the following model:
## thanks to user "WhatIf" for proposing `model = FALSE`
lmfit <- lm(y ~ poly(x) + z + I(z ^ 2) + z:x, data = df,
na.action = na.exclude, model = FALSE)
rm(df) ## do not omit this line! run it before trying other answers
Misc Replies
(1) why does the 'variables' attribute store the 'language' object that spells out a list rather than a regular "list" object?
Because "terms" was created in an early stage of model fitting: formula parsing. Actual variable evaluation happens later.
(2) how to convert a 'language' object to a regular "list" or "vector".
We do evaluation, where the quoted variables can be found:
eval(attr(terms(lm_obj), 'variables'), df)
You may get them out of the call,
fit <- lm(mpg ~ hp, mtcars)
head(all.vars(fit$call), -1)
# [1] "mpg" "hp"
or the names of the model.frame which is probably better.
names(model.frame(fit))
# [1] "mpg" "hp"
"language" is the (storage) mode or typeof of the object just as "double", "integer" or "list" are. See ?mode, for more explanation and nice examples. In the R language definition you find a detailed explanation—anyway a nice reading.
The 'variables' attribute of the terms is such a 'language' object.
(vars <- attr(fit$terms, 'variables'))
# list(mpg, hp, am)
typeof(vars)
# [1] "language"
To make use of it, we may coerce it as.character and remove the first element which is the name of the call, i.e. 'list',
as.character(vars)[-1]
# [1] "mpg" "hp" "am"
or, it can be evaluated, what it might actually intended for. Obviously it only will work, if we state with which data object the information is available.
with(fit$model, eval(vars))
# [[1]]
# [1] 21.0 21.0 22.8 21.4 18.7 18.1 14.3 24.4 22.8 19.2 17.8 16.4 17.3 15.2 10.4
# [16] 10.4 14.7 32.4 30.4 33.9 21.5 15.5 15.2 13.3 19.2 27.3 26.0 30.4 15.8 19.7
# [31] 15.0 21.4
#
# [[2]]
# [1] 110 110 93 110 175 105 245 62 95 123 123 180 180 180 205 215 230 66
# [19] 52 65 97 150 150 245 175 66 91 113 264 175 335 109
#
# [[3]]
# [1] 1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 1 0 0 0 0 0 1 1 1 1 1 1 1
If we ask why it isn't stored as a regular list, I would answer, because the information is already stored in the fit$model. It would be duplicated in a sense, and the size of the object would grow. The data is probably somewhere needed as list, the information itself actually is easier available using fit$model.
In your object m_obj$terms, it is formula and you can access each term of it using [[ extractor operator like
m_obj$terms[[1]]
#> `~` # formula symbol
if you want to get your input variables you can use
strsplit(as.character(lm_obj$terms[[3]])[2] , " \\+ ")[[1]]
#> [1] "x" "z"