Cross-derivative

[vincentarelbundock]

Ai & Norton call this the "interaction effect":

library(marginaleffects)
mod <- glm(am ~ mpg * wt, data = mtcars, family = binomial)
slopes(mod,
  variables = "mpg",
  newdata = datagrid(wt = c(mean(mtcars$wt), mean(mtcars$wt) + 0.001)),
  hypothesis = "(b2 - b1) / 0.001 = 0")