Round Research Results Without Changing Their Interpretation

Separate calculation from display

Rounding should usually be the last presentation step. Reusing rounded intermediate values can change totals, ratios and interval endpoints. Keep the computational result at appropriate precision, then create display strings under a documented policy.

The Python decimal documentation distinguishes rounding modes and exact decimal representation. This matters when reproducing values near halfway boundaries, but choosing decimal arithmetic does not itself determine the scientifically appropriate number of displayed digits.

Choose precision for the quantity

Use units, measurement resolution and uncertainty to guide display precision. A count of participants is an integer; an estimated ratio may need several digits to avoid collapsing distinct values to the same display. Avoid showing many decimal places simply because software returns them.

Apply the same policy to comparable estimates, while allowing clearly justified differences between counts, percentages, p-values and measurement units. Preserve the unrounded values used for decisions so a display threshold does not quietly become the analysis rule.

Inspect a hypothetical rounding conflict

Suppose three proportions are each exactly one third. Displaying each as 33.3% gives a total of 99.9%, although the unrounded proportions sum to 100%. A note that totals may differ because of rounding is more honest than changing one value without a declared method.

Now consider an interval endpoint of 0.996 for a ratio. Rounding to two decimals displays 1.00. The underlying interval and any associated test must be interpreted using the unrounded result and specified method, not a visual rule based on the displayed 1.00 alone. More digits may be needed to communicate the boundary clearly.

Check every presentation layer

Compare the abstract, results text, tables, figures and supplementary data against one authoritative result file. A manual update in one location can leave an old rounded value elsewhere. Automate display formatting where feasible and review boundary cases deliberately.

Do not report a p-value as exactly zero because it rounds below the chosen decimal precision. Use an appropriate inequality or scientific notation consistent with the method’s numerical resolution. The final report should look consistent while preserving the distinctions that matter scientifically, with enough underlying precision available for verification and reuse.

Source verification: 6 November 2026. Examples are hypothetical.