CliniGA CRO

CliniGA CRO Blog

Rounding and Precision in Statistical Results

Presentation precision should reflect information the data actually support

Representative research photograph: Rounding and Precision in Statistical Results
Photo: Flickr user ipasadelaide (artist: Jennie Groom) · License: CC BY 2.0 · Cropped and resized.
Published
Updated

Presentation precision should reflect information the data actually support. Too many decimal places can imply artificial certainty, while excessive rounding can obscure small differences, interval boundaries or a planned decision rule. Calculation and display need consistent but distinct treatment.

Define the question and assumptions

Keep full numerical values separate from displayed values. Specify the precision used to apply decision rules in code. Prepare consistent formatting for p-values, estimates and intervals. Very small values may need a suitable bound rather than a displayed zero. Define these conventions prospectively so that formatting does not become a tool for changing interpretation after results are known.

Hypothetical study example

In a hypothetical report, two p-values that look identical on screen receive different decision labels. The team identifies that decisions used unrounded values while display hid the distinction. It corrects formatting and notes rather than altering calculated values retrospectively to obtain a desired label or apparent consistency.

Design review checks

  • Separate numerical calculation precision from display precision in the implementation.
  • Test near-threshold examples and extremely small displayed values.
  • Reconcile rounding conventions across tables, figures and narrative statements.

Record the decision and its limits

Accepted outputs are numerically consistent and readable. Explain total discrepancies arising from rounding and verify whether a displayed zero means a true zero. Do not recalculate decision labels from formatted strings; retain the actual analysis trail. Extra digits should not be presented as removing measurement error or design uncertainty that remains in the underlying study.

Frequently asked questions

Do more decimals make a result more accurate?
No. Display detail does not improve measurement or estimation quality.
Should decisions use rounded values?
Apply the prespecified calculation rule; formatting should not change the decision.

Sources

  1. EMA — ICH E9 statistical principles and estimand addendum
  2. EMA — ICH E3 clinical study reports