Simplest way to calculate AME, MEM, MER?

Question

What is the simplest way to calculate average marginal effect, marginal effect at the mean and marginal effect at representative values for a logit model?

I found this example, but the explanation is messy and frankly I don't understand it: https://cran.r-project.org/web/packages/margins/vignettes/Introduction.html

I am using a STATA dataset ANES.dta with information from the 2000 presidential election in the USA. This is what the content of the dataset looks like:

dat <- structure(list(age = c(49, 63, 40, 47, 26, 48, 41, 18, 31, 22
), gender = c(1L, 1L, 2L, 2L, 2L, 2L, 2L, 2L, 1L, 2L), race = c(1L, 
1L, 1L, 1L, 2L, 1L, 1L, 1L, 5L, 1L), education = c(3L, 3L, 3L, 
4L, 4L, 2L, 2L, 2L, 3L, 3L), income = c(4L, 3L, 3L, 3L, 4L, 3L, 
3L, 3L, 4L, 3L), attendance = c(2L, 5L, 5L, 5L, 4L, 5L, 4L, 1L, 
0L, 3L), lib_con = c(59, 49, 94, 24, 29, 19, 39, 49, 79, 49), 
pro_choice = c(2L, 4L, 3L, 4L, 4L, 2L, 4L, 1L, 1L, 4L), vote = c(1, 
0, 1, 0, 0, 0, 0, 1, 1, 0), black = c(0, 0, 0, 0, 1, 0, 0, 
0, 0, 0)),row.names = c(NA, -10L), class = c("data.frame"))

Here is the head of the dataset:

age gender race education income attendance lib_con pro_choice vote black
1   49      1    1         3      4          2      59          2    1     0
4   63      1    1         3      3          5      49          4    0     0
5   40      2    1         3      3          5      94          3    1     0
8   47      2    1         4      3          5      24          4    0     0
9   26      2    2         4      4          4      29          4    0     1
10  48      2    1         2      3          5      19          2    0     0

And here is the code for my model:

rm(list=ls())

library(foreign)
dat <- read.dta("ANES.dta", convert.factors = FALSE)
dat_clear <- na.omit(dat)
head(dat_clear)

m1_logit <- glm(vote ~ gender + income + pro_choice ,
                data = dat_clear, family = binomial(link = "logit") , 
                na.action = na.omit)
summary(m1_logit)

Hi Nikolay, welcome to Stack Overflow. It will be much easier to help if you provide at least a sample of your data with `dput(data)` or if your data is very large `dput(data[1:10,])`. You can edit your question and paste the output. You can surround it with three backticks (```) for better formatting. See [How to make a reproducible example](https://stackoverflow.com/questions/5963269/) for more info. — Ian Campbell, May 23 '20 at 15:32

Ian Campbell · Accepted Answer · 2022-02-18T17:33:28.777

You can calculate the average marginal effects with margins::margins:

library(margins)
margins(m1_logit)
#Average marginal effects
#glm(formula = vote ~ gender + income + pro_choice, family = #binomial(link = "logit"),     data = dat, na.action = na.omit)
#
#   gender  income pro_choice
# -0.08049 0.08049    -0.1607

You can then calculate the marginal effects on the data or any arbitrary values with marginal_effects:

 marginal_effects(m1_logit)
#   dydx_gender dydx_income dydx_pro_choice
#1  -0.07200314  0.07200315     -0.14380458
#2  -0.08333712  0.08333712     -0.16644077
#3  -0.15510887  0.15510887     -0.30978309
#4  -0.03829122  0.03829123     -0.07647519
#5  -0.08333712  0.08333712     -0.16644077
#6  -0.21028256  0.21028254     -0.41997574
#7  -0.03829122  0.03829123     -0.07647519
#8  -0.07215379  0.07215380     -0.14410546
#9  -0.01377665  0.01377665     -0.02751471
#10 -0.03829122  0.03829123     -0.07647519

Simplest way to calculate AME, MEM, MER?

1 Answers1