I am running a GEE (Generalized Estimating Equations) Linear Scale Response regression. I have 4 IVs: Time pressure and Approach as as factors, and BIS and BAS sensitivity as covariates. My DV is "Choice". Under Model tab, I need to specify my model effects. What I would like to do is to test the main effects for all 4 IVs + the 3-way interaction between Time pressureApproachBAS (BIS is included only as a main effect as I want to control for the effect of BIS). This gives me a significant 3-way interaction effect. However, I wonder if this is correct? Can I specify the model effects that I am interested in? Don't I need to run full factorial models (that include all possible main, 2-way and 3-way interaction effects)? If I run the full factorial model, the significance of the 3-way interaction disappears.
1 Answer
$\begingroup$
$\endgroup$
2
If you have a 3 way interaction, you will (in nearly all cases) want to include the 2 way interactions that are part of that 3 way interaction and all the main effects in the 3 way interaction.
Similarly, if you have a 2 way interaction, you should (in nearly all cases) include the main effects.
You also shouldn't look exclusively at significance; look at effect sizes.
answered Jul 3, 2015 at 11:06
-
$\begingroup$ Thank you Peter! By effect sizes, you mean the B coefficient values? $\endgroup$– QuestionCommented Jul 3, 2015 at 11:08
-
$\begingroup$ Yes, exactly. Look at those and see what they mean in your model. $\endgroup$ Commented Jul 3, 2015 at 11:10