Two-Way ANOVA With Interaction
You'll learn how interaction effects between two factors can change the way each factor influences the outcome, and how to test for them.
What Is an Interaction Effect?
In two-way ANOVA without interaction, we assumed that two factors (like teaching method and class size) affect the response independently—the effect of one factor doesn't depend on the level of the other. But in reality, factors often interact.
An interaction occurs when the effect of one factor changes depending on the level of another factor. For example, a new teaching method might work brilliantly in small classes but poorly in large ones. The method's effectiveness depends on class size—that's interaction.
Visualizing Interaction
Imagine plotting group means: if lines connecting means across levels of one factor are roughly parallel, there's little interaction. If they cross or diverge significantly, interaction is likely present. Non-parallel profiles suggest the factors don't act independently.
Testing for Interaction
In two-way ANOVA with interaction, we partition total variation into:
- Factor A effect (main effect)
- Factor B effect (main effect)
- A×B interaction effect (how A and B combine)
- Residual error
We conduct an F-test specifically for the interaction term. If the interaction F-statistic exceeds the critical value (or p-value < α), we conclude the interaction is statistically significant.
Interpretation Matters
When interaction is significant, interpreting main effects alone can be misleading. You must examine the combination of factor levels. The story isn't "Factor A increases the response" but rather "Factor A increases the response when Factor B is at level 1, but decreases it when Factor B is at level 2."
Key Takeaway: Interaction effects reveal when two factors combine in ways that aren't simply additive—test for interaction first, and if significant, focus on interpreting combinations of factor levels rather than main effects alone.