avg_slopes() and AER package Tobit

[arantzagorostiaga]

Hi everyone

I have an issue when calculating avg_slopes() with the ouptut of tobit() function from the AER package.

When I run avg_slopes(), the average marginal effects are exactly equal to the tobit coefficients. It seems that only the linear part of the model is considered, rather than the full Tobit model.

Here’s an example:

library(AER)
library(marginaleffects)
data <- read.csv("https://vincentarelbundock.github.io/Rdatasets/csv/AER/Affairs.csv")

model <- tobit(
  affairs ~ age + yearsmarried + religiousness + occupation + rating,
  data = data,
  x = TRUE)

summary(model)

avg_slopes(model)

I expected the output to be:

mean(pnorm(model$x %*% model$coef / model$scale)) * model$coef[-1]

What am I missing here?

[vincentarelbundock]

Can't try right now. But did you check different values of the type argument?

[vincentarelbundock]

May be relevant: https://stats.stackexchange.com/a/149529/4874

[arantzagorostiaga]

Many thanks for your prompt reply! I understand now, it's the effect on the conditional mean of the latent variable which is what predict() returns.

Thanks again for your help and for the great marginaleffects package.