2

I have performed a logistic regression with the following variables:

  • X: Categorical with three levels (St, An, Int)
  • M: Moderator, which has two levels (Mob,Desk)
  • Y: Dichotomous dependent variable (0/1)

I found no main effect of X on Y or of M on Y but I did find a significant interaction effect between level 'Int' of X with M.

Hence, I want to show a graph in which I show the cross over interaction but I can't figure out how.

I already read this question (Plotting logistic regression interaction (categorical) in R) but I couldn't rewrite this to my own dataset.

PS: I am using R in case anyone would kindly like to provide an example.

Penguin_Knight
  • 11,988
Ank
  • 21

1 Answers1

2

There could be many ways. One of them is to compute the predicted odds, and turn that into probability for each possible combination between X and M. Here is a short example:

set.seed(144)

x <- as.factor(sample(c(1,2,3), 100, replace=T))
m <- as.factor(sample(c(1,1,2,2,2), 100, replace=T))
y <- sample(c(1,0), 100, replace=T)

m1 <- glm(y ~ x*m, family=binomial)

possibleCombo=data.frame(x = c(1,2,3,1,2,3), m = c(1,1,1,2,2,2))

yhatProb <- predict(m1, possibleCombo, type="response")*100

possibleCombo <- data.frame(possibleCombo,yhatProb)

library(ggplot2)
p1 <- ggplot(possibleCombo, aes(factor(x),yhatProb)) + 
      geom_bar(aes(fill=factor(m)), position="dodge", stat="identity")
p1

The result looks something like this:

enter image description here

Penguin_Knight
  • 11,988