Support for car::lht and better term handling

[grantmcdermott]

(Mostly) fixes #1090.

I was very tempted to create a new, dedicated glance.anova method and pull some columns that are currently returned as part of the tidy.anova object. (Most obviously, the "df.residual" and "rss" columns.) But for now I'll stick to a more conservative fix that just supports car::linearHypothesis and returns more information about the model contrasts. OTOH, the fact that glance.anova currently still returns a not very sensible data frame is probably something that needs to be fixed at some point.

Some examples:

devtools::load_all("~/Documents/Projects/broom")
#> ℹ Loading broom

library(broom)

a <- lm(mpg ~ wt + qsec + disp, mtcars)
b <- lm(mpg ~ wt + qsec, mtcars)

ab <- anova(a, b)

# new change: term column shows comparison models
tidy(ab)
#> # A tibble: 2 × 7
#>   term                   df.residual   rss    df    sumsq statistic p.value
#>   <chr>                        <dbl> <dbl> <dbl>    <dbl>     <dbl>   <dbl>
#> 1 mpg ~ wt + qsec + disp          28  195.    NA NA       NA         NA    
#> 2 mpg ~ wt + qsec                 29  195.    -1 -0.00102  0.000147   0.990

library(car)
#> Loading required package: carData
#> 
#> Attaching package: 'car'
#> The following object is masked from 'package:broom':
#> 
#>     recode
tidy(lht(a, "wt = disp"))
#> # A tibble: 1 × 10
#>   term     null.value estimate std.error statistic p.value df.residual   rss    df
#>   <chr>         <dbl>    <dbl>     <dbl>     <dbl>   <dbl>       <dbl> <dbl> <dbl>
#> 1 wt -disp          0    -5.03      1.23      16.6 3.39e-4          28  195.     1
#> # … with 1 more variable: sumsq <dbl>

^{Created on 2022-06-06 by the reprex package (v2.0.1)}

[simonpcouch]

@grantmcdermott, thank you for this PR! I apologize for not getting back to your original issue—as always, very well-considered and -argued.

I'm on board for this as well as the glance.anova addition. If you have the energy for it, feel free to add that to this PR or in a separate one, whichever you find more fitting.

I appreciate you linking out to the modelsummary issue; good to know where the both of yall are at in terms of how you think about broom's lifecycle and reliability.

I pushed one small change to pass the "hard" check. [edit: I had said I would request a change, but I see why your logic is the way it is now.🙂]

[simonpcouch]

Feel free to ignore the pkgdown checks—that machinery needs updating.🌝🌚

[grantmcdermott]

Thanks Si!

I can add the glance.anova method to this PR if you'd prefer. Here's what that would entail:

• Dropping any "df.residual" and "rss" columns from the returned tidy.anova object.
• Porting those columns to the glance method instead. (And maybe renaming "rss" to "deviance" to be consistent with other glance methods?)
• In some default cases—e.g. anova(lm(mpg ~ wt, mtcars))—the glance method would just be an empty data frame, since the return object doesn't produce appropriate glance-like columns.

Lmk your thoughts and I'll try to submit ASAP.

[simonpcouch]

Porting those columns to the glance method instead. (And maybe renaming "rss" to "deviance" to be consistent with other glance methods?)

On board!

In some default cases—e.g. anova(lm(mpg ~ wt, mtcars))—the glance method would just be an empty data frame, since the return object doesn't produce appropriate glance-like columns.

On board!

Dropping any "df.residual" and "rss" columns from the returned tidy.anova object.

I think I'd hold off on this. These columns have been around, at least by position, since some of the first commits to broom. I see the argument for why they shouldn't be there, but I'd imagine this would affect a good few reverse dependencies.

[grantmcdermott]

just to let you know i haven't forgotten this... need to get grading done first, though :'-|

[simonpcouch]

@grantmcdermott No rush. :)

[grantmcdermott]

Thanks for bearing with me @simonpcouch. I think these last few changes should do it.

I added the following note to the glance.anova() help documentation.

#' Note that the output of `glance.anova()` will vary depending on the initializing 
#' anova call. In some cases, it will just return an empty data frame. In other 
#' cases, `glance.anova()` may return columns that are also common to
#' `tidy.anova()`. This is partly to preserve backwards compatibility with early
#' versions of `broom`, but also because the underlying anova model yields 
#' components that could reasonably be interpreted as goodness-of-fit summaries
#' too.

Example:

devtools::load_all("~/Documents/Projects/broom")
#> ℹ Loading broom

a <- lm(mpg ~ wt + qsec + disp, mtcars)
b <- lm(mpg ~ wt + qsec, mtcars)

ab <- anova(a, b)

glance(ab)
#> # A tibble: 1 × 2
#>   deviance df.residual
#>      <dbl>       <dbl>
#> 1     195.          29

## Example where glance returns an empty DF
glance(anova(a))
#> # A tibble: 0 × 0

^{Created on 2022-06-13 by the reprex package (v2.0.1)}

[simonpcouch]

Awesome—I'm away from work for the week but will give this a more thorough look + merge if things look good next week. :)

[simonpcouch]

This looks great! No edits from me—will just update NEWS.

Was a little bit nervous about the tidy.anova column repositioning and renaming of res.df, so I ran some revdepchecks:

We checked 199 reverse dependencies, comparing R CMD check results across CRAN and dev versions of this package.

 * We saw 0 new problems
 * We failed to check 0 packages

Woop woop! This may be breaking for some non-testable dependencies, but this feels like a change worth making.

[github-actions]

This pull request has been automatically locked. If you believe the issue addressed here persists, please file a new PR (with a reprex: https://reprex.tidyverse.org) and link to this one.