CODEWITHSUNDEEP
mlogitmarginal-effectslogitsr-marginaleffects
elongat
elongat

Reputation: 11

Logit function and marginal effects

I have 2 approaches to calculate my marginal effects.

The first model is the following

model1 <- 
  feglm(
    FTAs ~ 
      log(dist) +
      log(gdp_d) +
     ...,
   
    family = binomial(link = "logit"),
    data = data
  )

And I calculate the marginal effects via

model1_marginaleffects <- 
  slopes(
    model1
  )
summary(model1_marginaleffects)

My second approach was the following which automatically calculates the marginal effects.

model2 <- 
  logitmfx(
    FTAs ~ 
      log(dist) +
      log(gdp_d) +
      
    
    data = data
    
  )

print(logitmfx_model3)

Does anyone know where the difference is coming from? The signs (+/-) are the same but the values totally differ. Do the functions take another baseline to calculate the marginal effects. If yes, what is the difference?

I looked into the documentations: https://cran.r-project.org/web/packages/marginaleffects/marginaleffects.pdf as well as https://cran.r-project.org/web/packages/mfx/mfx.pdf, but I do not understand the difference.

Upvotes: 1

Views: 195

Answers (1)

Vincent
Vincent

Reputation: 17823

I believe that marginaleffects computes the "average marginal effect (or slope)", whereas logitmfx computes the "marginal effect (or slope) at the mean".

Those are distinct quantities. Here is a nice tutorial: https://www.andrewheiss.com/blog/2022/05/20/marginalia/

Note that you can easily compute both quantities with the marginaleffects package. The discrepancy you observe is simply because the defaults differ, and not because marginaleffects is incapable of replicating logitmfx.

Also, please consider that it is much easier to answer a question like this when it is accompanied by a minimal working example. See:

How to make a great R reproducible example

Upvotes: 2

Related Questions