Interaction Plot Explained: How to Make, Read and Report It
An interaction plot shows how the effect of one factor depends on the level of another. It is the standard picture for two-way ANOVA and for any design in which you suspect that "it depends".
What it shows
The x-axis holds one factor, the y-axis holds the mean of the outcome, and each line represents a level of the second factor. Parallel lines suggest no interaction (the effect of one factor is the same at each level of the other). Non-parallel lines suggest interaction; crossing lines indicate a crossover interaction. The plot is descriptive: whether non-parallelism exceeds sampling noise needs a test.
Original example (illustrative)
Exam scores for a 2 × 2 design: teaching method (Lecture, Active) × prior knowledge (Low, High), equal cell sizes. Cell means: Low/Lecture 60, Low/Active 75, High/Lecture 70, High/Active 74.
The effect of active teaching is 75 − 60 = 15 points for low-knowledge students and 74 − 70 = 4 points for high-knowledge students. The interaction contrast is 15 − 4 = 11. Marginal means: Active 74.5 vs Lecture 65.0 (difference 9.5), and High 72.0 vs Low 67.5 (difference 4.5). The marginal difference of 9.5 hides the fact that the benefit is concentrated among low-knowledge learners, which is why a main effect alone can mislead when an interaction exists.
No variances are given, so this example cannot tell whether the interaction is statistically reliable. A two-way ANOVA, or a regression with a product term, would test this (Aiken & West, 1991).
How to make it in StatFAS
Open the Interaction Plot Calculator. Provide an outcome column and two factor columns. See also the ANOVA calculator.
Reading and reporting
Include error bars and state what they represent. Caption: Figure 1. Mean exam score by teaching method and prior knowledge; error bars = 95% CI. Report the interaction test alongside the plot, and when it is significant, describe simple effects rather than leaning on main effects.
Common mistakes and alternatives
Judging an interaction by eye with no test, omitting error bars, connecting categorical levels as though they were continuous trends, ignoring unequal cell sizes, and reading the interaction as causal mediation. Alternatives include grouped bar charts and tables of simple effects.
Conclusion
An interaction plot makes conditional effects visible, but only the accompanying test and uncertainty tell you how much to trust them.
1.
statfas-interaction-plot-interface.png, alt: "StatFAS Interaction Plot Calculator interface"2.
statfas-interaction-plot-example-data.png, alt: "Example data entered in the StatFAS Interaction Plot Calculator"3.
statfas-interaction-plot-settings.png, alt: "Interaction Plot Calculator settings in StatFAS"4.
statfas-interaction-plot-result.png, alt: "Interaction Plot Calculator result generated in StatFAS"References
- Aiken, L. S., & West, S. G. (1991). Multiple regression: Testing and interpreting interactions. Sage.