Harmonize Measurement Units Without Losing Provenance

Confirm that the quantities match

Two columns with similar names do not necessarily measure the same quantity. Check the analyte or construct, specimen, measurement method, and reported unit before proposing a conversion. Converting the unit cannot remove differences in what was measured or how samples were collected.

Record the original numerical value and unit separately. Keep the source dictionary or table footnote available when an abbreviation is unclear. Never infer a unit solely because most values appear plausible under one interpretation.

Build a conversion map

For each source field, specify the target unit, conversion formula, reference, and conditions under which the rule applies. The BIPM SI Brochure, ninth edition updated in 2026, provides the authoritative foundation for SI units and decimal prefixes.

Use case-sensitive unit labels and write compound units unambiguously. Distinguish a concentration from a total quantity. Converting between mass concentration and amount concentration requires the appropriate substance-specific molar mass; it is not a universal scale factor that can be copied across analytes.

Verify a fictional example

Suppose a fictional length measurement is 125 millimetres. Expressing the same quantity in centimetres gives 12.5 centimetres because ten millimetres equal one centimetre. If a sample mean is 125 millimetres and its SD is 8 millimetres, the converted mean is 12.5 centimetres and the SD is 0.8 centimetres.

For a pure scaling conversion, apply the factor to confidence-interval endpoints and the absolute factor to the SD or SE. Variance scales by the square of the factor. A conversion involving an additive offset changes location summaries but does not add that offset to the SD. Keep these operations explicit when converting published summaries.

Review the full transformation

Apply the conversion to a copy or derived field, not over the only original value. Compare a small set of hand-calculated cases, including zero where meaningful, a typical value, and a boundary value. Check that a second conversion back to the original unit recovers the input within the expected rounding tolerance.

Inspect source-specific distributions after harmonization to find unresolved labels, decimal errors, or differences in measurement processes. Matching ranges alone do not validate a conversion; unusual but correct observations must not be discarded just to make sources look similar.

Save the conversion map with versioned analysis code and record when it was applied. Finish with a list of unresolved units that remain excluded from conversion pending clarification. This makes a later source correction recoverable without guessing which fields were already transformed.

Source checked on 31 October 2026; fictional conversions independently recalculated.