rmarginal-effectsr-marginaleffects

Replicate `avg_comparisons` by hand

I'm struggling replicating the results given by avg_comparisons.

What am I doing wrong?

library(marginaleffects)

dat <- get_dataset("thornton")
mod <- glm(outcome ~ incentive * (agecat + distance), 
           data = dat, family = binomial)

avg_comparisons(mod, variables = "incentive")$estimate
# 0.451920016659

# Counterfactual predictions
mydf <- dat

mydf$incentive <- 1
mydf$pred1 <- predict(mod, newdata = mydf, type = "response")

mydf$incentive <- 0
mydf$pred0 <- predict(mod, newdata = mydf, type = "response")

mean(mydf$pred1-mydf$pred0, na.rm=TRUE)
# 0.45174545593
Solution

  • There are missing data in your data frames. Call na.omit() first and your code will replicate the results exactly.