How can I extract marginal effects from an interaction term?

Question

I am trying to extract the marginal effects from an interactive term that captures for the effects of a treatment X (X is coded as 1 or 0) on outcome Y (Y is coded in a scale from -10 to 10) moderated by variable A (A is coded between 0 and 10). However, I am not sure how to extract the marginal effects from the interaction terms at the highest and lowest measure of A

 m<-lm(Y~ X*A, data = combined)

Overall, I managed to generate plots for the marginal effect using the interplot function:

interplot(m = m, var1 = "X", var2 = "A", ci = 0.90)+
  ylab("X")+
  xlab("A")+
  theme_bw()+
  ggtitle("Figure 1. Effect of X on Y Moderated by A")+
  theme(plot.title = element_text(face = "bold"))+
  geom_hline(yintercept = 0, linetype = "dashed")

Additionally, I tried to use ggpredict to extract the marginal effects with 90% confidence interval at different levels of A:

margin1<- ggpredict(m, c ("X", "A"), ci = 0.90)

margin1

However, using ggpredict produces marginal coefficients that do not align with what I see in the summary for m, and do not align with the marginal effects plot. Instead, I get estimates that are clearly not accurate or precise. How can I extract the marginal effects as seen in the interplot?

Can you share some minimal sample data? It's easier to help and more interesting to us & others if we have some data to work with. [`sjPlot` allows you to visualise marginal effects of interaction terms](https://cran.r-project.org/web/packages/sjPlot/vignettes/plot_marginal_effects.html). — Maurits Evers, Sep 06 '19 at 04:23

score 0 · Accepted Answer · answered Sep 06 '19 at 05:37

Since you unfortunately don't provide sample data, here is a minimal example using the R built-in state.x77 dataset.

fit <- lm(Income ~ Illiteracy * Murder, data = as.data.frame(state.x77))

We are interested in the marginal effect of Illiteracy on Income

library(sjPlot)
plot_model(fit, type = "int")

Here plot_model uses the minimal and maximal values of Murder as the grouping levels (this is the default behavious, see the Plotting linteraction effects of Regression Models vignette for details).

How can I extract marginal effects from an interaction term?

1 Answers1