MerQuR+Statistics Glossary

Statistics Glossary → Interaction plot

Interaction plotconcept

A plot that shows the means of the outcome across the levels of one factor, with a separate line for each level of a second factor, to visualise whether the two factors interact.

The levels of the first factor are on the horizontal axis and the cell means (or estimated marginal means) on the vertical axis; each level of the second factor is a separate line. If the lines are parallel, the effect of the first factor is the same at every level of the second, so there is no interaction. If the lines converge or diverge, the interaction is ordinal (the size of the effect changes but not its direction); if they cross, it is disordinal (the direction of the effect reverses).

The plot is not a test; visible departures from parallel can be due to chance, so the F test for the interaction term and the confidence-interval bars should be read alongside it. When a strong interaction is present, interpreting the main effects on their own is misleading; simple effects (the effect of one factor at each level of the other) should be examined instead. In unbalanced designs it is better to plot estimated marginal means rather than raw cell means.

An everyday example

A study measures the effect of coffee on attention in sleepy and well-rested people. In rested people coffee raises the attention score by 2 points, in sleepy people by 12 points. In the interaction plot both lines rise, but the line for the sleepy group is much steeper: the effect of coffee depends on how rested a person is.

In MerQur

In two-way ANOVA the 'Chart' option can be set to 'Interaction plot' or 'Interaction plot + CI bars'; the CI version draws each cell mean ± t(1 - α/2, df_error)·√(MSE/n). In mixed-design ANOVA, the 'Chart CI' checkbox adds error bars to the interaction plot.

Analyses where this concept appears

Two-Way ANOVAMixed-Design ANOVA, Split-Plot

Related concepts

Interaction effectEstimated marginal means (EMM)