Three-Way (2x2x2) Between-Subjects ANOVA in SPSS

Question

I'm just wondering how I can decompose a significant 3-way interaction in SPSS (all three IVs are categorical) without splitting the file, so that my follow up degrees of freedoms in the follow up tests remain the same as the omnibus degrees of freedom?

Answer 1

Instead of using split file (which would run separate ANOVAs for each subgroup), you can use planned contrasts or interaction terms within either the GLM drop downs or in your syntax in SPSS. This way, all comparisons are made within the same model, which would preserve your error term and degrees of freedom from the omnibus ANOVA.

So if we assume your factors are: A (2 levels), B (2 levels), and C (2 levels), and you found a significant A × B × C interaction, you can use the following approach using the drop downs for the general linear model.

Step 1: Run the Full 3-Way ANOVA

Go to:

1. DV: Your dependent variable
2. Fixed factors: A, B, and C

Click “Model” → choose Full factorial

Click “Options” → Select A, B, C, and their interactions → Check "Descriptive statistics" and "Estimates of effect size"

Now run it and verify that the 3-way interaction is significant.

Step 2: Use Custom Contrasts (or Estimated Marginal Means)

To break down the interaction, you would now examine the 2-way interactions at each level of the third variable using EMMeans and simple effects:

1. In the Univariate dialog box, click EMMeans. ---> Select A*B by C
2. Click “Compare main effects” or use the “Display means for” box to choose A and B at each level of C.

Now SPSS will give you tests of simple 2-way interactions (A×B) at each level of C.

This keeps the full model intact, so your df are preserved.