01Translate the axis label into an operation

GraphPad’s normalisation documentation describes mapping measurements onto a common scale using defined reference values, including zero and one-hundred-percent endpoints. The choice of those endpoints is therefore part of the result’s meaning, not merely a plotting preference.GraphPad Prism User Guide — Normalize (opens in a new tab)

For reading purposes, distinguish dividing by a control from subtracting a baseline. The first produces a ratio; the second produces a difference. Mapping a range onto 0–100% generally combines subtraction and division.

Two figures with the same axis label may use different operations. Find the methods or figure legend rather than assuming a familiar label guarantees a common calculation.

02See what separate scaling can conceal

In an original hypothetical example, dataset A has a low reference of 10 signal units and a high reference of 110. Dataset B has corresponding references of 10 and 60. Their raw response spans are therefore 100 and 50 units.

Mapping each dataset separately with 100 × (signal − low)/(high − low) makes both low references 0% and both high references 100%. The transformed endpoints match even though the raw response spans differ twofold.

That may be appropriate for a question about relative position within each response range. It cannot, on its own, demonstrate equal absolute response capacity between the datasets, because that difference was removed by the chosen scaling.

03Check whether values were scaled individually or together

Dividing every measurement by one common fixed number preserves their ordering and rescales their differences consistently. Dividing different subsets by different reference values can change comparisons between those subsets.

Consider two original pairs of reference and response values: 10 and 20, then 100 and 110. The average of the paired ratios is (2 + 1.1)/2 = 1.55. The ratio of the mean responses to the mean references is 65/55, approximately 1.18.

Neither calculation is merely a different display of the same summary. One gives each pair equal weight in the ratio average; the other combines the raw values first. The intended question and design determine which is appropriate.

A caption should therefore state whether normalisation occurred within each experimental unit, within each run or after pooling group summaries. That detail can be as important as the final formula.

04Do not treat the reference as perfect by default

A measured control or baseline has its own variation. Sharing it across several transformed observations can connect those observations statistically. A graph that fixes a control summary at one does not establish that the original control measurements had zero variation.

Look for the raw control values, independent experimental units and the analysis used for the transformed data. A plotted ratio alone cannot show whether the reference’s uncertainty and shared structure were considered appropriately.

Normalisation also cannot rescue an unsuitable readout. Dividing an interfering signal by a reference does not prove that the numerator specifically measures the biological target.

A concise evidence note names the normalisation rule and the comparison it supports. If raw-scale information is unavailable, state which conclusion cannot be checked from the transformed graph. That preserves the benefit of a clear visual while keeping its limits visible.