Plot interaction effects

[FDFS]

Hi!

I already know how to plot the predicted probabilities of binomial models without any interaction terms:

predicted logit = a + bX

predicted probabilities= EXP(predicted logit)/(1+EXP(predicted logit))

What I needed to know is how to do it when the model contains an interaction. For simplicity's sake the model we're talking about has only one IV (scale ranging from 0 to 10) and one moderator (ordinal ranging from 1 to 4), which means obviously the interaction is between the two of them. Which equation should I follow to plot the predicted probabilities graph of this interaction?

Thank you for your help.

 

[hlsmith]

I believe typically you would plot four lines (stratified by moderator). You would then see the lines cross of have different slopes or shapes (depending on the types of responses you got for the first IV).

 

[threestars]

I would plot first differences (sometimes incorrectly called marginal effect plots). Here's a link to an article describing how to do it in Stata, code included. You could also use Zelig in R.

https://files.nyu.edu/mrg217/public/pa_final.pdf

 

[FDFS]

I believe typically you would plot four lines (stratified by moderator). You would then see the lines cross of have different slopes or shapes (depending on the types of responses you got for the first IV).

Yes, but that I could do that even without the interaction term, right? What would be the advantage of including the interaction term, then? And would that still capture the interaction between the two variables?

 

[hlsmith]

Is your interaction term significant? If so, you would want to control for it, since your beta coefficients will be adjusted when it is in the model.

 

[FDFS]

I would plot first differences (sometimes incorrectly called marginal effect plots). Here's a link to an article describing how to do it in Stata, code included. You could also use Zelig in R.

https://files.nyu.edu/mrg217/public/pa_final.pdf

Thanks. I'll take a look. Not having Stata or R (only SPSS), anyway I can do it manually?

 

[FDFS]

Is your interaction term significant? If so, you would want to control for it, since your beta coefficients will be adjusted when it is in the model.

Yes it is. How can I control it? By setting the interaction variable to its mean?