Three way interactions in regression

I'm trying to figure out how to do three-way interactions with multiple regression. I am trying to predict an IV by a three DVs and the interaction between them. All my variables are continuous. I have been referred to Aiken and West but am still confused about how to start and how I can use SPSS to begin my analysis.
Any help will be realllly appreciated.
Thank you!
Hi Sifar16,

Doing the interaction in SPSS isn't difficult, but interpreting the results will be.

First of all, I assume you have one DV and three IVs.

To do it:
Use Analyze-->General Linear Model-->Univariate
Add your DV as Dependent Variable
If all three IVs are continuous, put them all in Covariates.

Now click Model, then Custom.

Put all three IVs in the Model box by clicking on each one at a time, then clicking on the arrow button.

To add interactions, click on two (or three for the 3-way) of the IVs, change the box under the arrow to say "interaction," then click the arrow.

If you have a 3 way interaction in the model, you must have all the 2-way interactions or your parameter estimates will be meaningless.

As it is, it is difficult at best to interpret a two-way interaction between two continuous variables. For a very well written article on how to do it, go to

My next question is whether you're sure you want a 3 way interaction? What is your hypothesis?

Hi Karen,

Thank you so much for your reply.

What I'm trying to test is whether personality at time 1 interacts with group norms and group identification to predict personality at time 2.

For example:
extraversion (time 2) = extraversion (time 1) * group norms (related to extraversion) * group identification
such that, if group norms are highly extraverted and the individual has high group identification, then scores on extraversion should be significantly higher.

I have also been referred to Jeremy Dawson's spreadsheets to do simple slope analysis ..

I hope to find that group norms and group identification do influence personality change, and thus I hope to find atleast two-way interactions, if not three-way (though I doubt I will find these!).

Thank you once again for your help, I really appreciate it :)

Hi Sifar16,

I'm not familiar with any of the references you mentioned, so I'm not sure in which way they'll help.

But anyway, just go ahead and run it in SPSS glm. Make sure you include all the individual IVS as well as all the two-way interactions, or none of the parameter estimates will mean anything.

One more thing--it will help a lot with both interpretation and reducing multicollinearity if you center your IVs. It's pretty common to get multicollinearity between an IV and an interaction term containing that IV, especially when they're all continuous. For more info, see:

Once you run it, let me know if you need help in interpretation.


I was not aware that SPSS computed interactions with continuous variables via GLM. Is this correct, or does SPSS require that you compute the interaction terms yourself before entering them into the model?

I hope you continue to follow this message board!