Reading an Interaction Plot
Interpret parallel, diverging, and intersecting slope lines in statistical software plots to diagnose antagonistic and synergistic factor combinations.

Contents Overview
How the Plot Is Built
An interaction plot places the levels of one factor on the horizontal axis and draws a separate line for each level of a second factor. Each point is the mean response for that combination.
Statistical packages, including Minitab, produce these plots directly from a factorial worksheet.
Key Principle
The effect of one factor may depend on the level of another.
Three Typical Patterns
Roughly parallel lines suggest little or no interaction: the effect of the first factor is about the same at both levels of the second.
Lines that diverge or converge suggest the size of an effect depends on the other factor. Lines that cross suggest the direction of the effect may reverse.
Example Plot Shapes
Surface prep: Level 1 Surface prep: Level 2Parallel lines
PatternPress duration changes the response by about the same amount for both surface preparations: little or no interaction.
Diverging lines
PatternThe benefit of a longer press is larger for one surface preparation: the size of the effect depends on the other factor.
Crossing lines
PatternThe direction of the press-duration effect reverses between preparations: a strong interaction pattern.
Not results. These sketches show shapes an interaction plot could take; they are not measured data and do not indicate which pattern the study would produce.
What the Plot Does Not Prove
The plot shows means only. Small differences in slope can arise from ordinary variation, especially with few runs. Whether an interaction is meaningful must be judged from the fitted model and its error estimate.
Always check the scale of the vertical axis; a narrow axis can make tiny differences look dramatic.
Key Takeaways
- Each line represents one level of the second factor.
- Parallel, diverging and crossing lines suggest different interaction patterns.
- Confirm visual impressions with a model that includes error.