Trace a Census correction through a published analysis

Triage the notice

Record the notice title, affected collection years, product, tables, geography and what the publisher changed. Compare those fields with the archived source manifest. A correction affecting a one-year product may not affect a five-year estimate with a similar table name. If applicability is uncertain, retain the original results and mark the decision pending.

Rerun the smallest complete dependency chain

Keep the old extract. Save the corrected data separately and compare keys, estimates, margins and annotations. Identify every downstream table, map and narrative statement that uses an affected field. Refit derived models where required; editing a displayed figure alone can leave another output inconsistent.

Hypothetical impact assessment

A fictional correction changes a numerator from 12 to 15 with a denominator of 100 unchanged. The displayed rate changes from 12% to 15%, a 3-percentage-point revision. If an inclusion rule requires at least 14 events, the record also changes eligibility. Recompute the eligible set before rerunning its summaries.

Document results even if the conclusion is unchanged. Include the original extraction date, corrected version, order of rerun and which results were unaffected. Distinguish a publisher correction from a mistake in your own code; they require different preventive checks.

Primary source checked 7 November 2026: U.S. Census Bureau: ACS errata notices.

Use this in your study: Download: Source revision impact map.