A result can sound dramatically different depending on whether it is expressed as a difference, a percentage change or a ratio. These descriptions can all be correct while emphasising different aspects of the same observations. The starting point is to recover the two values and the denominator.
Distinguish subtraction from division
Cochrane separates difference measures from relative measures and distinguishes risk ratios from odds ratios. The choice changes what is being compared, so an odds ratio should not simply be rewritten as a risk ratio.Cochrane Handbook — Choosing effect measures and computing estimates of effect (opens in a new tab)
For a simple positive-valued measurement, an absolute change is the new value minus the reference. Relative change divides that difference by the reference. A fold value instead divides the new value directly by the reference.
| Description | Calculation |
|---|---|
| Reference | 40 units |
| New value | 50 units |
| Absolute increase | 50 − 40 = 10 units |
| Relative increase | 10/40 = 25% |
| Fold value | 50/40 = 1.25-fold |
A value of 1.25-fold means 125% of the reference, or a 25% increase. It does not mean a 125% increase. Reading the calculation prevents those two statements being confused.
Keep percentage points separate from percent change
Suppose an original hypothetical experiment finds that the fraction of independently assessed units meeting a defined response criterion rises from 20% to 30%. The absolute difference is ten percentage points.
Relative to the original 20%, the increase is 10/20 = 50%. The new proportion is 1.5 times the reference proportion. These are three descriptions of the same change, provided the response definition and denominators are comparable.
A statement that response increased by 10% would be ambiguous here. It might mean ten percentage points, or it might mean a relative increase from 20% to 22%. The underlying values resolve that ambiguity.
Inspect the starting value and the direction
The same relative increase can accompany very different absolute changes. A rise from 2 to 3 units and a rise from 200 to 300 units are both 50% increases. Whether either matters depends on the measurement and the research question.
Percentage changes are also asymmetric when the direction is reversed. Moving from 40 to 50 is a 25% increase, whereas moving from 50 back to 40 is a 20% decrease, because the denominator changes.
A reference of zero makes ordinary relative change undefined. A reference very close to zero can produce an enormous ratio from a small absolute difference. Negative or background-corrected references can make a casual percentage interpretation even less useful.
Do not repair an inconvenient denominator by silently adding a constant. If a transformation or offset was used, it becomes part of the definition of the reported quantity and needs to be described.
Keep numerical presentation separate from biological interpretation
A large relative change in an assay signal does not necessarily represent a large change in biological function. The signal might be a proxy, have a nonlinear relation to the target or sit near the limits of useful measurement.
Likewise, a ratio calculated from group means does not generally equal the average of individual paired ratios. Those operations answer different summary questions. Find the formula or analysis description before comparing apparently similar fold-change figures.
Uncertainty belongs with the effect measure actually reported. Limits for an absolute difference cannot simply be relabelled as percentage-change limits without accounting for the reference and the way the estimate was calculated.
When summarising a paper, give the original and comparison values if available, then the stated difference or ratio. That allows readers to see the scale of the result and reduces the temptation to choose whichever presentation sounds largest.
Sources and further detail
- Cochrane Handbook — Choosing effect measures and computing estimates of effect (opens in a new tab)
Difference/ratio distinctions and risk-versus-odds sections read. All signal values and response proportions in this article are original illustrations, not study results or claims about human benefit.
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.