2
$\begingroup$

I am trying to compute probit regression that includes interaction terms. When I compute marginal effects after the main coefficients R gives me marginal effects for interaction terms and Stata doesn't. It says that Stata doesn't compute marginal effects for interaction terms because logically it's not possible. But then how does R compute them? and if it oes are they correct ?

I tried both Stata and R for the analysis and I am confused.

Improve this question
$\endgroup$
3
  • 7
    $\begingroup$ Can you add a toy example or R + Stata code with the output? $\endgroup$
    – dimitriy
    Commented Nov 21, 2022 at 1:48
  • 2
    $\begingroup$ Please give a precise definition of marginal effect in this context. To me marginalization is unconditioning and I don't know why you would uncondition on anything while looking at something as highly conditional as an interaction effect. $\endgroup$
    – Frank Harrell
    Commented Nov 22, 2022 at 12:41
  • $\begingroup$ R calculates them in the same way that they would for a term not involved in an interaction b*dnorm(xb), but as Stata suggests, they are meaningless. You cannot change the interaction while holding constant the constitutive terms. $\endgroup$
    – Dave Armstrong
    Commented Dec 16, 2022 at 0:50

1 Answer 1

Reset to default
1
$\begingroup$

Here is an example where I compute the same average marginal effects using R and Stata. The margin's vignette is also useful to explain marginal effects with interaction terms using R: https://cran.r-project.org/web/packages/margins/vignettes/Introduction.html

In Stata, the margins command can be used to compute marginal effects for interaction terms.

use "https://dss.princeton.edu/training/students.dta", clear

* Create the Math dummy variable
gen math= major=="Math"
encode gender, gen(sex)

probit math c.age##i.sex

* Compute marginal effects for interaction term
margins, dydx(*)

  Average marginal effects
  Number of   obs     =         30 Model VCE    : OIM
  Expression   : Pr(math), predict() dy/dx w.r.t. : age 2.sex

  Delta-method dy/dx   Std. Err.      z   P>z [95% Conf. Interval]        
  age   -.0087594   .0124467    -0.70 0.482   .0331544    .0156356
  sex  Male     -.361844   .1711841    -2.11  0.035   .6973588   -.0263293
  Note: dy/dx for factor levels is the discrete   change from the base level.

In R, the margins package can be used to compute marginal effects for interaction terms.

library(foreign)
library(margins)

mydata <-  read.dta("http://dss.princeton.edu/training/students.dta")

# Create the Math dummy variable
mydata$math <- ifelse(mydata$major == "Math", 1, 0)

model <- glm(math ~ age + gender + age:gender, data = mydata, 
             family = binomial(link = "probit"))

# Compute marginal effects for interaction term
m <- margins(model)
margins_summary(model)

 factor     AME     SE       z      p   lower   upper
 age -0.0088 0.0123 -0.7112 0.4770 -0.0329  0.0154  
 genderMale -0.3618 0.1716 -2.1084 0.0350 -0.6982 -0.0255
Improve this answer
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.