3
$\begingroup$

In this vignette, the plot_model function is described and some examples are given on how to plot two way interactions along with confidence bands. I am wondering how to bets interpret such plots. For example, the first plot from the vignette is:

enter image description here

How can we use this plot to conclude whether or not including the interaction between gender and Barthel Index score is a good idea? Does the fact that the confidence band of Females not include the Male fit tell us that an interaction is warranted?

I am trying to use such a plot to explain whether or not it is a good idea to include an interaction, my plot looks like:

enter image description here

$\endgroup$

1 Answer 1

1
$\begingroup$

It's of course good to check your data visually, but usually you want to include an interaction effect when it is meaningful theoretically and/or based on previous results. I.e., if you have a reason to expect the estimates to be different for your two groups.

Another thing to consider is whether your sample is powerful enough to detect an interaction effect (at least in psychology, interaction effects are typically tiny). If not, it's not a good idea to test one.

Your own plot looks like there might be an interaction effect, but the confidence bands are wide, so there's also a lot of error. If your sample is adequately powered and you have a reason to expect an interaction effect, in my opinion you can well include one and see how that goes.

$\endgroup$
2
  • $\begingroup$ my question is really about interpretation of the plots and not about interaction effects more generally - why do you conclude that there 'might be an interaction effect' based on the plot? $\endgroup$ Sep 6, 2022 at 12:23
  • $\begingroup$ Sorry, I misunderstood. My interpretation was not based on any formal evaluation, just on experience in examining visual plots along with interaction results. Sometimes, the difference in slopes depicted in you plot is enough to produce a significant interaction effect. $\endgroup$
    – Sointu
    Sep 12, 2022 at 7:58

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.