Research Guides
Keep Display Rounding Out of Your Research Calculations
A table can be correct even when its displayed percentages total 99.9%. It can also be wrong for reasons that have nothing to do with rounding. Before adjusting a research report to make the numbers look tidy, separate the stored measurements, the calculations and the displayed values.
This is a manual reporting workflow for data you are authorized to use. It does not describe a Colquhoun.shop precision setting or certify a source dataset.
1. Identify which numbers you actually have
Keep the original export and record whether its values are counts, estimates, rates or already-rounded summaries. A screenshot showing 1.2 thousand visits does not contain an exact count of 1,200. If greater precision is unavailable, keep that limitation rather than reconstructing digits the source never supplied.
In a working file, distinguish the source value from the calculated result and its presentation format. Changing a cell's display to one decimal place is not the same decision as replacing its stored value with that rounded number.
2. Test the presentation with a simple example
Imagine three mutually exclusive traffic groups containing 10 visits each. Each group represents one third of the 30-visit total. Displayed to one decimal place, each share is 33.3%, so the displayed shares add to 99.9%.
The missing tenth of a percentage point is not an unidentified fourth traffic group. It arises from the presentation of the three fractions. Keep the underlying counts and calculation intact; do not silently change one group's measured share to make the displayed total look exact.
The U.S. Bureau of Labor Statistics illustrates how separately rounded components can fail to sum to a rounded total, even when calculations use unrounded data. It also shows that rounded component values can stay unchanged while their total changes. BLS: Effects of Rounding on Published Estimates
3. Investigate before blaming rounding
Check the denominator, reporting period and included categories. Look for omitted rows, overlapping groups and totals that were accidentally counted as details. A substantial discrepancy needs an explanation based on the source, not a generic footnote saying that rounding occurred.
For a report you calculate from complete source values, reconcile those values first. Then apply the documented display precision and see whether it explains the visible difference. For a report available only as rounded figures, state that exact reconciliation may be impossible.
4. Choose precision for the decision
Use a consistent display rule for comparable values. If two values both appear as 2.0%, do not call them exactly equal unless the source supports that claim. Conversely, showing many decimal places does not make an uncertain estimate more trustworthy.
Ask what difference the reader needs to understand. A summary might use a compact rounded figure while a supporting table retains more detail. Label the unit and precision so someone copying the figure does not mistake a readable summary for an exact measurement.
5. Carry the rounding rule into the handoff
- Source: file, measurement definition and available precision.
- Calculation: numerator, denominator and included records.
- Display: units, decimal places and rounding convention.
- Reconciliation: underlying total and displayed total checked separately.
- Limits: unavailable precision or unexplained discrepancy.
- Publication: table, chart and export checked for consistent meaning.
When the difference is genuinely explained, a brief note can say that displayed shares may not sum to 100% because of rounding. Keep the calculation record behind that statement. The aim is not a perfectly tidy column; it is a report whose visible numbers remain faithful to the evidence.