A combination producing a larger response than either component alone does not automatically demonstrate an interaction. The key question is whether the effect of one factor changes with the other factor, on the scale and under the model used. That requires a comparison of effects, not just a comparison of the largest bar.
Compare effects across the other factor’s levels
Minitab’s interaction-plot guidance describes the relationship between one factor and a response as depending on the second factor. It also distinguishes displaying such a pattern from assessing it through an appropriate statistical analysis.Minitab — Interpret the key results for Interaction Plot (opens in a new tab)
For an experiment with factors A and B, the relevant conditions can be described as neither, A alone, B alone and both. On an additive outcome scale, compare the change associated with A when B is absent with the change associated with A when B is present.
| Effect of A | Subtract |
|---|---|
| Without B | A alone minus neither |
| With B | Both minus B alone |
| Interaction contrast | Effect with B minus effect without B |
This is a reading framework for reported experiments. It does not recommend combining materials or supply a protocol for doing so. The validity of the comparison depends on the study’s design and measurement.
A larger combination response can still be additive
In an original hypothetical example, the mean response is 10 units with neither factor, 14 with A, 16 with B and 20 with both. The combination is higher than either component alone.
Yet the effect of A without B is 14 − 10 = 4 units, and its effect with B is 20 − 16 = 4 units. The difference between these effects is zero. The pattern is additive on this raw response scale.
If the mean response with both were 25 instead, the A effect with B would be 25 − 16 = 9 units. The interaction contrast would be 9 − 4 = 5 units. That is an estimated departure from additivity in these illustrative means.
State the model against which a combination is judged
Additivity is a statement about a particular scale. In a second original example, responses of 10, 20, 30 and 60 units give an A-associated ratio of two both without B and with B. The ratios are constant.
On the raw difference scale, however, the A-associated change is 10 units without B and 30 units with B. A constant multiplicative effect therefore need not be a constant additive effect.
This arithmetic explains why interaction and synergy claims need an explicit reference model. A paper comparing its observations with one expected-combination model may not be answering the same question as a paper using another.
Do not choose the scale after inspection solely to obtain an interaction label. Read why the scale or model fits the outcome and hypothesis, and whether the choice was specified before the relevant result was known.
Separate statistical dependence from a mechanism claim
A significant component effect in one condition and a non-significant effect in another do not, by themselves, establish that the effects differ. The difference between those effects needs to be evaluated directly.
Likewise, evidence of an interaction in a measured response does not identify its molecular cause. Changes in measurement behaviour, biological context or other features may need investigation before a specific mechanism is supported.
The result is also bounded by the tested conditions. One combination at one setting does not establish a general interaction across every concentration, time point or model. A paper should identify which part of that space was actually observed.
A clear summary gives the interaction contrast or model comparison, its uncertainty, the outcome scale and the experimental setting. It then distinguishes the measured pattern from any proposed explanation. This prevents an appealing combination narrative from outrunning the evidence.
Sources and further detail
- Minitab — Interpret the key results for Interaction Plot (opens in a new tab)
Official definition and distinction between plotting and statistical analysis read. All four-condition values, interaction contrasts and scale examples are original, not findings about any named peptide combination.
Sources checked 19 September 2026. Worked examples are illustrative unless a supplied report is explicitly identified. This article has not undergone independent scientific peer review.