Is ther a way in R and modelsummary to stack a comparisons table produced by marginal effects

Question

I ran several linear regressions and obtained comparisons from the avg_comparisons function of the marginaleffects package and then form it with the shape = term : contrast ~ model option parameters of the modelsummary package to concatenate the term and contrast. This results in the desired first part of the table. Now I wanted to make two panels and stacked them. With standard regressions this works very fine with the shape = "rbind" function, but not when I need to concatenate term and contrast before. Is there a way in the syntax of the shape option that allows to combine this two different behaviors? Or is there another way of achieving the stacking of the panels?

I ran the following code to produce the comparisons

vars <- c("var1", "var2", "var3")

panel1 <- vars %>%
    paste(., "~ Treatment") %>%
    map(as.formula) %>%
    map(lm, data = df) %>% 
    map(avg_comparisons, variables = "Treatment", vcov = "hc1") 
modelsummary(panel1, shape = term : contrast ~ model)

this produced the desired outcome for one panel. Then I tried to add another panel

vars <- c("var4", "var5")

panel2 <- vars2 %>%
    paste(., "~ Treatment") %>%
    map(as.formula) %>%
    map(lm, data = df) %>% 
    map(avg_comparisons, variables = "Treatment", vcov = "hc1") 

panels(list(panel1, panel2))

modelsummary(panels, shape = term : contrast ~ model)

This produced an error, as modelsummary does not accept the inputs as list, when shape is not "rbind".

Answer 1

Unfortunately, I can't think of a completely automatic way to do this right now. What you could do is proceed in steps: Create an intermediate representation in modelsummary_list format, combine the labels, and feed the result back to modelsummary. Example:

library(modelsummary)
library(marginaleffects)

# fit
mod <- list(
  lm(mpg ~ am + factor(cyl), data = mtcars),
  lm(mpg ~ hp + am + factor(cyl), data = mtcars)
)

mod <- lapply(mod, avg_comparisons)

# intermediate representation
tab <- modelsummary(mod, output = "modelsummary_list")

# combine labels
tab[[1]]$tidy$term <- paste(tab[[1]]$tidy$term, tab[[1]]$tidy$contrast)
tab[[2]]$tidy$term <- paste(tab[[2]]$tidy$term, tab[[2]]$tidy$contrast)

# back to modelsummary
modelsummary(tab, shape = "rbind", gof_map = "nobs")

Answer 2

I have transformed @Vincent s answer into a function, to automatize a bit more.

transform_comparison_contr_treat <- function(comparison) {
    map(comparison, ~ {
        .x$tidy$term <- paste(
            paste(.x$tidy$term, .x$tidy$contrast),
            .x$tidy$TreatmentJob
        )
        return(.x)
    })
}