A platform’s reported share moves from 12.3% to 12.4%, and a slide announces a precise growth rate. The displayed figures support a direction under specific assumptions, but they do not preserve every decimal used to produce the underlying values. Rounding can make a tiny movement look exact and a real movement look unchanged.
EcomToolkit recommends matching the strength of a headline to the precision of its evidence. This guide uses hypothetical platform shares to explain rounding intervals, percentage-point changes, and defensible comparisons. None of the example values represents the current market share of Shopify, WooCommerce, or any other named platform.
Table of Contents
- Identify what the percentage describes
- Recover the interval hidden by one decimal place
- Bound the change instead of inventing decimals
- Distinguish percentage points from relative growth
- Explain totals that do not equal one hundred
- Write headlines that preserve the evidence
- Build a reusable platform statistics note
- The EcomToolkit view
Identify what the percentage describes
Before investigating decimals, confirm the population. A percentage of all measured websites differs from a percentage of detected ecommerce sites. A count of domains differs from a count of merchants, and neither automatically measures sales value. More decimal places cannot repair a denominator mismatch.
Record the provider, collection date, population definition, category, and published precision beside the figures you plan to compare. Use the provider’s own documentation to establish those details. The W3Techs technology overview is one entry point for checking the categories behind a technology survey; it is not interchangeable with a merchant revenue dataset.
For a broader discussion of those definitions, see the guide to platform market-share selection traps. The narrower question here is what can be concluded after the two observations are already comparable but only rounded percentages are available.
Recover the interval hidden by one decimal place
Assume a provider rounds percentages to the nearest tenth of a percentage point. A displayed value of 12.3% then corresponds approximately to an underlying value from 12.25% up to 12.35%. Endpoint inclusion depends on the tie-breaking convention, so avoid claiming more exactness than the publisher specifies.
The width of that rounding interval is 0.1 percentage points. It is not a confidence interval and does not describe sampling, detection, or coverage error. It describes the precision lost when a more detailed number was displayed with one decimal place.
The ONS guidance on rounding emphasizes balancing readability against lost precision and choosing a level appropriate for the intended use. For a platform research brief, that means a compact number may be suitable for orientation while being insufficient for a precise calculation of small changes.
| Hypothetical displayed share | Assumed display rule | Approximate underlying interval |
|---|---|---|
| 12.3% | Nearest 0.1 percentage point | 12.25% to 12.35% |
| 12.4% | Nearest 0.1 percentage point | 12.35% to 12.45% |
| 2% | Nearest whole percentage point | 1.5% to 2.5% |
| 0.0% | Nearest 0.1 percentage point, nonnegative share | 0% to 0.05% |
The final row matters when a small platform is described as having no presence. A displayed zero can represent a positive share below the display threshold. Verify whether the publisher uses rounding, truncation, suppression, or a special zero convention before drawing that conclusion.
Bound the change instead of inventing decimals
If the first observation displays 12.3% and the second displays 12.4%, subtracting the labels gives 0.1 percentage points. Under the assumed nearest-tenth rule, however, the underlying increase can be arbitrarily close to zero or approach 0.2 percentage points. The displayed change is a summary, not the exact movement.
The same labels can hide very different trajectories. Values of 12.349% and 12.351% differ by just 0.002 percentage points yet display as 12.3% and 12.4%. Values of 12.251% and 12.449% differ by 0.198 percentage points and produce the same displayed pair.
These examples are arithmetic illustrations, not plausible bounds on every provider’s measurement error. They isolate rounding while holding the population and method fixed. If collection coverage changes between dates, the analytical uncertainty is larger and requires a separate explanation.
Unchanged labels also need care. Two observations displayed as 12.3% could conceal either a small rise or a small fall within the interval. Say the reported share was unchanged at the published precision, rather than claiming that the underlying population was exactly stable.
Distinguish percentage points from relative growth
Moving from 12.3% to 12.4% is a displayed increase of 0.1 percentage points. Dividing 0.1 by 12.3 and multiplying by 100 gives approximately 0.81% relative growth in share. Those are two different ways of expressing change, and neither should be labelled simply percent without context.
Calculating a long decimal expansion from rounded inputs does not restore missing precision. Reporting 0.813008% growth suggests information the published labels do not contain. If relative change is useful, identify it as an approximation calculated from the rounded figures and use restrained precision.
For very small shares, rounding has a larger relative effect. A move from a displayed 0.1% to 0.2% looks like a doubling, but the underlying interval combinations can imply quite different ratios. Avoid dramatic growth headlines when the display resolution is large relative to the starting share.
A value displayed as zero is especially unsuitable as the denominator for relative growth. Obtain unrounded data or describe the change in percentage points at the supported precision. Do not replace zero with an arbitrary small constant to make a percentage-growth formula produce an answer.
Explain totals that do not equal one hundred
Suppose three exhaustive, mutually exclusive hypothetical categories each have an underlying share of one third. Displaying each as 33.3% gives a total of 99.9%. The missing tenth is a presentation artifact; it does not establish that a fourth category is absent from the report.
Conversely, do not assume every imperfect total is rounding. Categories may overlap, exclude unknown cases, or use different populations. Confirm that the shares are intended to form one exhaustive partition before expecting them to sum to one hundred in the first place.
When underlying values are available, calculate totals from those values and round the result for presentation. If only rounded values are available, label any total as an approximation. Avoid adjusting an individual platform’s published share solely to force a visually tidy total without explaining the adjustment.
This matters when creating an other-platforms row by subtraction. Subtracting several rounded figures from one hundred carries their rounding uncertainty into the remainder. A small remainder can therefore be a poor basis for claims about the collective size or movement of minor providers.
Write headlines that preserve the evidence
The goal is not to surround every statistic with an unreadable technical note. Use a concise headline and a nearby definition that explains precision, population, and period. Reserve interval calculations for decisions that actually depend on a small difference.
| Evidence available | Defensible wording | Wording that overstates it |
|---|---|---|
| 12.3% then 12.4%, same method | Reported share rose by 0.1 percentage points | Exact growth was 0.813008% |
| 12.3% at both dates | Unchanged at published precision | No underlying change occurred |
| Two platforms both show 2.1% | Equal after rounding | Exactly equal adoption |
| A platform displays 0.0% | Below the stated display resolution, if confirmed | No sites use this platform |
| Rounded categories sum to 99.9% | Check rounding and category definitions | The source lost 0.1% of sites |
Equal displayed values do not establish a ranking. If one platform is 2.06% and another is 2.14%, both display as 2.1% under the assumed rule. When that difference would influence a decision, obtain more precise comparable data rather than selecting a winner from the formatting.
A wider displayed lead may survive rounding while still being affected by survey methodology. Keep those issues separate. Rounding bounds can establish what the displayed precision permits, but they cannot establish whether the underlying technology detection or sampling process is unbiased.
Build a reusable platform statistics note
Save the original observation with its URL, access date, displayed value, and population definition. Preserve the published format instead of silently normalizing every source to three decimal places. Include any documented rounding convention, and mark an assumed convention explicitly when it is necessary for an illustrative bound.
For repeated comparisons, keep the same source and population where possible. If a source changes its method, separate the series or explain the break. A clean spreadsheet with uniform decimal places can conceal a methodology change just as easily as a poorly labelled chart.
Finally, connect the statistic to the actual platform decision. Adoption share may be background context for ecosystem research, but a tiny rounded movement does not measure implementation fit, migration cost, or merchant operating capability. The precision exercise should improve judgment rather than encourage more elaborate calculations on a weak decision input.
The EcomToolkit view
A small number deserves careful language, not extra invented decimals. Keep percentage points distinct from relative growth, distinguish rounding from measurement uncertainty, and ask for better data when a decision depends on a difference the published figures cannot resolve.
If your platform comparison relies on incompatible or over-precise statistics, request an EcomToolkit audit to review the evidence and the operational questions it should support.
