inst/tinytest/test-ellipses.R

In modelsummary: Summary Tables and Plots for Statistical Models and Data: Beautiful, Customizable, and Publication-Ready

library(modelsummary)

# exponentiate logit coefficients

mod <- glm(am ~ mpg, mtcars, family = binomial)

# not exponentiated
raw <- modelsummary(
  mod,
  gof_omit = ".*",
  output = "dataframe",
  exponentiate = FALSE)
truth <- c("-6.604", "(2.351)", "0.307", "(0.115)")
expect_equivalent(raw[[4]], truth)

# exponentiated conf.int
raw <- modelsummary(
  mod,
  gof_omit = ".*",
  statistic = "conf.int",
  output = "dataframe",
  exponentiate = FALSE)
truth <- c("-6.604", "[-12.328, -2.772]", "0.307", "[0.122, 0.587]")
expect_equivalent(raw[[4]], truth)

# exponentiated std.error
# As noted by Alex P Hayes:
# https://github.com/tidymodels/broom/issues/422
# I believe this is the correct behavior: quantiles (and thereby confidence
# intervals) are invariant under monotonic transformation (exp()) but
# variances are not.
raw <- modelsummary(
  mod,
  gof_omit = ".*",
  output = "dataframe",
  exponentiate = FALSE)
truth <- c("-6.604", "(2.351)", "0.307", "(0.115)")
expect_equivalent(raw[[4]], truth)