Add arguments to rename polynomials with coef_rename()

[mccarthy-m-g]

This PR updates coef_rename() to handle polynomial coefficients created by the poly() function, similar to how the function currently handles factors. Here's some reprex code to check the new functionality out:

x <- list(
    lm(mpg ~ poly(cyl, 2) + drat + disp, data = mtcars),
    lm(hp ~ poly(cyl, 2, raw = TRUE) + drat + disp, data = mtcars)
)

# Results in polynomial names of the form '^1', '^2', etc.
modelsummary(dvnames(x), coef_rename = coef_rename)

# Results in polynomial names of the form 'Cyl^1', 'Cyl^2', etc.
modelsummary(dvnames(x), coef_rename = \(.x) coef_rename(.x, poly_name = FALSE))

Note that I didn't update NEWS or add any tests yet (I'm unfamiliar with tinytest, so I'm unsure how to go about it).

[vincentarelbundock]

Oh thanks, this is a really great idea! Awesome.

Two questions:

1. Could we combine the two arguments? Maybe poly doesn't have to be TRUE/FALSE and can immediately say what "style" we want.
2. Will the default style work in all output formats? For example, I believe Cyl^2 would break LaTeX compilation, and is not valid markdown either.

[mccarthy-m-g]

Welcome! Regarding your questions:

Will the default style work in all output formats? For example, I believe Cyl^2 would break LaTeX compilation, and is not valid markdown either.

This approach just lets us clean up poly() coefficients to get the same result we would get if we cleaned up polynomials that were constructed "by hand" with I():

x <- list(
    lm(mpg ~ I(cyl^1) + I(cyl^2) + drat + disp, data = mtcars),
    lm(mpg ~ poly(cyl, 2, raw = TRUE) + drat + disp, data = mtcars)
)
modelsummary(x, coef_rename = \(.x) coef_rename(.x, poly_name = FALSE))

So the intention isn't to print the exponents in superscript---just as is. I don't believe this should break any output formats unless you attempt to apply some post-processing to the string that results in something breaking. But if that's the case you'd need special handling for I() statements too.

Could we combine the two arguments? Maybe poly doesn't have to be TRUE/FALSE and can immediately say what "style" we want.

I think it makes sense to stick with a boolean argument for poly, since every other argument for coef_rename() is also a boolean.

What we could do though is reduce it to a single boolean argument for the Cyl^2 style. I only included the ^2 style based on the options you currently give for factors---but personally I don't think this style is particularly useful, so I'd just drop it unless you can imagine a use case.

[vincentarelbundock]

Got it.

Yes, let's simplify to a single bool with the style you prefer.

The ^ character should really be tested in a latex document though (or Quarto/Rmd PDF), because I think it's a special character.

If that's the case and it breaks compilation, we either need some more complicated logic to escape conditionally on output format (would be a pain to code), or we need to come up with a different default style.

[vincentarelbundock]

Oh I see your point about I(x^2).

I think that works because modelsummary escapes this character by default.

So we're fine on that front.

[mccarthy-m-g]

Awesome. I simplified the code to do the Cyl^2 style when poly = TRUE, so this gives nice output by default now:

x <- list(
    lm(mpg ~ I(cyl^1) + I(cyl^2) + drat + disp, data = mtcars),
    lm(mpg ~ poly(cyl, 2, raw = TRUE) + drat + disp, data = mtcars)
)
modelsummary(x, coef_rename = coef_rename)

Since ^ is escaped by default, we don't need to worry about testing in a latex document now, correct?

[vincentarelbundock]

Thanks a ton for this! Looks great.

I merged.

The latest release was only a few days ago, so it'll take a little bit before this gets to CRAN, but it should be available on Github and R-Universe in the next few minutes.

[mccarthy-m-g]

Happy to contribute! I've been using modelsummary frequently for the example articles of a data package I'm working on and it's been super useful.

Thanks for merging!