update regression objects by passing an expression to "subset" - r

Question

I have many regression objects created by lm(). Each one has been built from a different data
frame, and these different data frames have different dimensions. But each data frame contains the logical variables x, y, and z. In some cases, I want to update each regression object so that the subset argument is x. In other cases, I want to update each regression object so that the subset argument is y. And in still other cases, I want to update each regression object so that the subset argument is z. What is an efficient way to do this?
This is the inefficient way:
# Set only one of these three variables to be TRUE
subsetX <- TRUE
subsetY <- FALSE
subsetZ <- FALSE
# Now update the regressions.
if (subsetX) {
update(lm1, subset = x)
update(lm2, subset = x)
[...]
} else if (subsetY) {
update(lm1, subset = y)
update(lm2, subset = y)
[...]
} else if (subsetZ) {
update(lm1, subset = z)
update(lm2, subset = z)
[...]
}
This approach is inefficient because there is a lot of duplication across the three code blocks that update the regressions. I would rather do something like
subsetVar <- dplyr::case_when(
subsetX ~ expression(x),
subsetY ~ expression(y),
subsetZ ~ expression(z))
update(lm1, subset = substitute(subsetVar))
update(lm2, subset = substitute(subsetVar))
[...]
That is, I would like to write at most one update() command for each
regression object, while still varying the subset argument on the basis of
logical (boolean) variables like subsetX and subsetY. Is this possible?
The code above doesn't work; when I try it, I get an Error in xj[i] : invalid subscript type 'symbol' error message.
I've searched other Stack Overflow questions, but I haven't found anything that speaks directly to this problem.

Answer 1

Here's a way to make your workflow a little easier with a simple function and using purrr::map2 to feed it the list of models and subsets you want
library(purrr)
set.seed(2020)
mtcars$x <- sample(c(TRUE, FALSE), 32, replace = TRUE)
mtcars$y <- sample(c(TRUE, FALSE), 32, replace = TRUE)
mtcars$z <- sample(c(TRUE, FALSE), 32, replace = TRUE)
lm1 <- lm(mpg ~ hp, mtcars)
subset_lm_by <- function(model, subset = NULL) {
if (subset == "x") {
update(model, subset = x)
} else if (subset == "y") {
update(model, subset = y)
} else if (subset == "z") {
update(model, subset = z)
} else {
# cat('I only accept x, y or z!')
}
}
models <- list(lm1, lm1, lm1, lm1)
subsets <- list("x", "y", "z", "nonsense")
purrr::map2(.x = models,
.y = subsets,
~ subset_lm_by(model = .x, subset = .y))
#> [[1]]
#>
#> Call:
#> lm(formula = mpg ~ hp, data = mtcars, subset = x)
#>
#> Coefficients:
#> (Intercept) hp
#> 31.21178 -0.08098
#>
#>
#> [[2]]
#>
#> Call:
#> lm(formula = mpg ~ hp, data = mtcars, subset = y)
#>
#> Coefficients:
#> (Intercept) hp
#> 32.83501 -0.07294
#>
#>
#> [[3]]
#>
#> Call:
#> lm(formula = mpg ~ hp, data = mtcars, subset = z)
#>
#> Coefficients:
#> (Intercept) hp
#> 32.53554 -0.08688
#>
#>
#> [[4]]
#> NULL