← All studies · askamerica.ai
q116 · askamerica

How often do federal statistical series get redefined or discontinued, and which ones broke most recently?

Answer

Federal statistical series don't break on a fixed schedule — they break continuously, at two very different rates. Among the handful of headline series journalists actually cite (CPI, unemployment rate, GDP, poverty rate), consequential redefinitions land roughly once every several years to a decade. Among the much larger universe of secondary/detail series, discontinuation is constant: this connector's own FRED metadata catalog (econ_reference.fred_series) carries 230 series explicitly tagged DISCONTINUED, clustered in identifiable waves (104 ending in 2014, 108 in 2015, 12 in April 2021), not spread evenly.

Most recent breaks (2024–2026) are budget/staffing-driven collection cuts, not textbook rebasings:

This is a different failure mode than the classic textbook examples: it's agency capacity attrition (BLS budget down ~19% in real terms since 2009 per the American Statistical Association, with 20–30% further staff losses across statistical agencies in 2025 per CBPP) plus, in a few cases (LGBTQ+ items dropped from NCVS/NHIS under Executive Order 14168), policy-driven content changes.

Well-established older examples, for calibration:

Catalog-navigation evidence (used the connector's own metadata, not just web search)

econ_reference.fred_series is FRED's own series-discovery catalog, partitioned into 7 categories including one literally named "Discontinued/Legacy." Querying it directly:

SELECT CAST(substr(observation_end,1,4) AS INTEGER) AS end_year, COUNT(*) AS n
FROM econ_reference.fred_series
WHERE title ILIKE '%discontinued%' AND observation_end IS NOT NULL AND observation_end < '2026-01-01'
GROUP BY CAST(substr(observation_end,1,4) AS INTEGER) ORDER BY end_year

returns 2014→104, 2015→108, 2017→1, 2018→2, 2019→2, 2021→12. Inspecting titles: the 2014–2015 wave is almost entirely Census's per-state "Business Formations Within N Quarters" survivor-rate series, retired when the Business Formation Statistics program moved to its newer "Business Applications" series; the 2021 wave is the Federal Reserve retiring detailed NAICS-level industrial-production indexes for oil/gas and mining subsectors (e.g. IPG211111N, last observation 2021-04-01). An observation_end that stops well short of the present on a series still named after a live concept is exactly the declared-coverage-window signal this catalog is built to expose.

Caveat: this count reflects FRED's own tagging convention, not a census of every U.S. statistical redefinition — methodology changes that keep the same series ID (like CPI formula changes) aren't tagged DISCONTINUED and only show up in agency notices, which is why this answer combines both.

Confidence

Medium-high for the named 2024–2026 examples (each traces to the agency's own notice or a dedicated tracker cross-checked against independent reporting). Medium for "how often" as a rate — no single authoritative count of "redefinitions per decade" exists across the full federal series universe; the catalog's 230-series DISCONTINUED count is real but is a lower bound restricted to FRED's tagging practice, not a complete inventory. High for the older textbook examples (Boskin, NAICS, BEA 2013), which are extensively documented primary-source events.

Sources

Delivery

Full HTML report (narrative + dashboard) published via publish_report: http://127.0.0.1:51135/a/ea4cbfb6dbfaebd8a363b6b67386eb4f.html (local-only link, not fetchable outside this session — content reproduced in full above). Chart images saved alongside this file: chart-discontinued-by-year.png (FRED DISCONTINUED count by final observation year) and dashboard.png (published report's inlined dashboard).

Rendered from agent.md by tools/build_register.py — never edited directly.