Combining output modelsummary with multinomial and logistic models

[lorenzoFabbri]

I have a list of models, some of them are multinomial models from nnet, other are logistic models from base R's glm. I encounter an error when using this list as input for modelsummary:

Error: Group columns (response) were not found in the extracted data. The "group" argument must be
  a column name in the data.frame produced by `get_estimates(model)`.  If you wish to combine
  models with and without grouped estimates, you will find examples on the modelsummary
  website:
  https://modelsummary.com

The issue is that get_estimates has a missing column for response since there are only two levels. A simple example:

df <- mtcars
df$cyl <- factor(df$cyl, levels = c(4, 6, 8))
df$am <- as.factor(df$am)

multi <- nnet::multinom(
  formula = cyl ~ drat,
  data = df,
  Hess = FALSE
)

bin <- glm(
  formula = am ~ drat,
  data = df,
  family = binomial()
)

ll <- list(multi, bin)

modelsummary::modelsummary(
  ll[[1]],
  shape = model + term ~ response,
  estimate = "{estimate} ({p.value}) {stars}",
  statistic = NULL,
  coef_rename = \(.x) modelsummary::coef_rename(
    x = .x,
    factor_name = TRUE,
    backticks = TRUE,
    underscore = TRUE,
    titlecase = FALSE
  ),
  coef_omit = "Intercept",
  include_reference = TRUE,
  gof_map = NA,
  notes = list(
    'Format: estimate (p-value).',
    'Significance: "+" = .1, "*" = .05, "**" = .01, "***" = 0.001.'
  ),
  fmt = modelsummary::fmt_decimal(
    digits = 1,
    pdigits = 2
  )
)

Is there a way to force get_estimates to include a dummy response column so that the same function (with shape = model + term ~ response,) can be used for all elements of this list?

[vincentarelbundock]

@lorenzoFabbri did you find a solution for this?

I re-opened the issue because I think it's an interesting problem, but I can't think of an easy solution right now. Initially I thought that something like below would work, but in my quick attempt I couldn't get it through:

library(modelsummary)

df <- mtcars
df$cyl <- factor(df$cyl, levels = c(4, 6, 8))
df$am <- as.factor(df$am)

multi <- nnet::multinom(
  formula = cyl ~ drat,
  data = df,
  Hess = FALSE
)

bin <- glm(
  formula = am ~ drat,
  data = df,
  family = binomial()
)

tidy_custom.glm <- function(x, ...) {
    x <- broom::tidy(x, ...)
    x$response <- "6"
    x
}

ll <- list(multi, bin)
modelsummary(ll, shape = term + response ~ model)