Communicate a result without erasing its limits

A result needs a defined quantity, units, sample and interval. “The value increased” is incomplete if the reader cannot tell what was measured or compared. Distinguish an absolute change from a relative change, and state which baseline supplies the denominator.

For a fictional detection fraction rising from 20% to 25%, the absolute difference is five percentage points. Relative to the original 20%, the increase is 25%. Both statements can be correct, but “a five-percent increase” would ambiguously or incorrectly describe this comparison.

A confidence interval — open definition belongs to a specified estimator and method. A small reported interval can describe precise estimation within a model while leaving selection bias — open definition or calibration uncertainty unaddressed. Do not present one calculated component as if it covered every source of doubt.

For a fictional estimate of a change of 2 units with a defined interval from −1 to 5, the interval includes zero and both signs. That does not prove the true change is exactly zero. It means the stated analysis has not pinned down the sign at that interval’s level and assumptions.

Name the denominator and the uncertainty

A useful report can be concise: what was estimated, from which observations, by which method, with what uncertainty and limitations. Separate the statistical result from practical importance. A small effect can be measured precisely, while a potentially important effect can remain poorly constrained.

If an unexpected pattern is found after examining many possible relationships, disclose that exploration and seek a suitable follow-up. Presenting only the most striking result as if it were the sole planned test can exaggerate the evidence. Honest limits make a result easier to build on: another reader can identify which new observation would most improve the conclusion.