# How a Federal Statistic Is Revised or Retracted Almost every important federal number is published before anyone knows whether it is right. The first estimate of a quarter's economic growth lands about thirty days after the quarter ends, built from source data that are still arriving; the monthly jobs figure is revised twice before the quarter is out and then again, a year later, against tax records. This is not a flaw in the system. It is the system, and the agencies say so openly: the choice is between a provisional number now and an accurate number too late to use. That design creates a problem of a different kind, and it is the subject of this map. Because federal statistics are *expected* to move, a scheduled revision and a genuine mistake arrive by the same door, in the same kind of notice, and look very much alike to a reader. The machinery for the first is elaborate — release calendars filed with the [[Office of Management and Budget]] by December 15 each year, revisions consolidated so they land together, three-year performance reviews measuring the magnitude and direction of every revision. The machinery for the second barely exists. The directive that governs the release of the government's headline economic indicators has no section on errors at all; its nearest provision is titled "Preliminary Estimates and Revisions." The regulation that codifies what a statistical agency owes the public, issued in 2024, has no corrections provision either. So this map draws both tracks side by side: the revision that was planned before the number was published, and the correction that nobody planned for. It follows each to the place that matters most and is least discussed — what happens to the record. The answer, on the evidence of the agencies' own errata notices, is that a corrected figure replaces the wrong one in the database while the news release that carried the wrong figure into the world is sometimes left standing, uncorrected and unmarked, on the same website. This page is the companion to [[How a Federal Survey Is Approved]], which maps how a statistic is authorised before it exists, and to [[How a Statistical Agency Is Defunded or Discontinued]], which maps what happens when the program that produces it ends. > [!note] How to read this chart > `──▶` = the main path · `└─▶` = a branch or bypass · `◇` = a decision point · `★` = the step that actually decides it · `⚑` = a load-bearing stage, where the real attrition happens · `☐` = a deadline somebody can count in advance · `◎` = the releasing agency decides alone, with no outside review and no appeal · `§` = the rule is in statute, so a directive cannot change it · `⊘` = an obligation that exists nowhere, on the evidence of the sources surveyed · `◈` = a consequence written in no rule · `≠` = two errors of the same kind, given unequal process · `✕` = the figure leaves the record here · `[ ]` = a dated note. Stage numbers are this map's own; the instruments' section numbers are given at each stage. ``` HOW A FEDERAL STATISTIC IS REVISED OR RETRACTED (as of 12 November 2026) │ Legend: ──▶ main path └─▶ branch or bypass ◇ decision ★ the deciding step │ ⚑ load-bearing stage ☐ a countable deadline ◎ agency decides alone │ § in statute ⊘ an obligation that exists nowhere in the sources surveyed │ ◈ a consequence written in no rule ≠ unequal process, same kind of error │ ✕ the figure leaves the record [ ] dated note │ │ The three instruments: │ SPD No. 3 Statistical Policy Directive No. 3, release of the principal federal │ economic indicators. Revised 89 FR 11873, eff. 19 Feb. 2024. │ SPD No. 1 Statistical Policy Directive No. 1, fundamental responsibilities of │ statistical agencies. 79 FR 71610, eff. 2 Dec. 2014. │ 5 CFR 1321 the "Trust Regulation," final rule, 89 FR 82453, eff. 10 Dec. 2024, │ under 44 U.S.C. § 3563. │ — and, standing outside all three, the Information Quality Act: four sentences │ of an appropriations act, Pub. L. 106-554 § 515, carried in the U.S. Code only │ as a note to 44 U.S.C. § 3516. │ ■ 0. ONE SUBJECT, TWO TRACKS, AND ONLY ONE OF THEM IS BUILT ⚑ │ A published federal number changes for one of two reasons, and everything │ downstream follows from which. The instruments govern the first in detail and │ are nearly silent on the second. │ │ ├─◆ TRACK A — THE SCHEDULED REVISION. Planned before publication, announced │ │ in advance, released on a calendar filed with OMB a year ahead. The │ │ number was always provisional and the agency said so. │ │ │ └─◆ TRACK B — THE CORRECTION. The number was wrong. No calendar, no │ announced date, no stated deadline for posting one, and — in SPD No. 3 │ and in 5 CFR 1321 alike — no section bearing the word. ⊘ │ ├──▶ 1. PUBLISHED PROVISIONAL BY DESIGN (SPD No. 3 § 6(a)) │ "Agencies shall clearly identify figures as preliminary or revised." │ ☐ § 2: series issued quarterly or more often are released within 22 working │ days of the end of the reference period. The clock is the point: accuracy │ is traded for timeliness deliberately, at a pace OMB sets. │ │ Worked example — the three vintages of one quarter's GDP, 2026 Q2: │ advance estimate 30 Jul. 2026 real GDP +1.5% (annual rate) │ second estimate 26 Aug. 2026 real GDP +1.5% │ third estimate 30 Sep. 2026 real GDP +2.2% ← revised up 0.7 point │ Nothing was wrong at any stage. The same quarter grew at three different │ rates in three official publications, and the third is not final either. │ [ BEA, "GDP (Third Estimate) … 2nd Quarter 2026," released 30 Sep. 2026. ] │ ├──▶ 2. THE SCHEDULED REVISION FIRES (SPD No. 3 § 6(e)) │ "Revisions occurring for routine reasons, such as benchmarking and updating │ of seasonality factors, shall be consolidated and released simultaneously." │ Three routine engines, all calendared: │ ├─ benchmark revision — the survey is re-anchored to a near-universe count │ ├─ seasonal factors re-estimated — every prior month can move │ └─ annual and comprehensive updates — the whole back series is rebuilt │ │ Worked example — the 2026 benchmark of the payroll survey: │ Released with January 2026 data on 11 Feb. 2026. Seasonally adjusted total │ nonfarm employment for March 2025 revised by −898,000, or −0.6 percent, │ to a benchmarked level of 158,377,000. Over the prior ten years the annual │ benchmark revision has averaged 0.2 percent in absolute terms — so this │ one is three times the decade norm, and it is still a routine revision. │ [ BLS, Current Employment Statistics benchmark article, 11 Feb. 2026. ] │ │ ◈ The consequence no rule addresses: a routine revision of this size and an │ error of this size are indistinguishable to a reader, because the system │ has one vocabulary for both. │ ├──▶ 3. ◇ ROUTINE, OR NOT ROUTINE? (SPD No. 3 § 6(e) · § 6(f)) │ The only fork the directive draws, and it draws it by listing examples of │ "routine" rather than by defining either side. ★ │ │ │ ├─routine──▶ consolidated, released on the regular schedule (§ 6(e)). │ │ No separate notice. No flag. Stage 2 above. │ │ │ └─not routine──▶ "shall be fully explained" and "shall be released as soon │ as corrections can be completed" (§ 6(f)). │ ☐ Note what that is not: "as soon as corrections can be │ completed" is a condition, not a date. Nothing in the │ directive says how long completing them may take. ⊘ │ ├──▶ 4. ◇ WHO FOUND IT? ⚑ ≠ │ The single most consequential branch on this map, and the instruments draw │ it nowhere. The same error gets a wholly different process depending on who │ noticed — and the track with a deadline is the one the agency does not own. │ │ │ ├─the agency found it─────▶ stage 5A. Errata. No deadline. ◎ │ │ │ └─someone outside found it─▶ stage 5B. A Request for Correction under the │ Information Quality Act. 120 days. An appeal to │ different people. A draft answer OMB sees first. │ ├──▶ 5A. THE AGENCY FOUND IT: AN ERRATUM ◎ ⊘ │ Not required by SPD No. 3, not required by 5 CFR 1321, and not, on the │ sources surveyed, required by anything. The statistical agencies maintain │ standing errata pages as a matter of practice. Each entry carries its own │ "corrections made on" or "corrections to be made on" date, set by the │ agency, case by case, against no external clock. │ │ Scale, from one agency's running list: roughly 45 entries dated 2026 and │ roughly 23 dated 2025 on the BLS errata page — an order of magnitude more │ activity than any instrument above contemplates. │ │ ├─◆ 5A(i). THE DATABASE IS CORRECTED. The wrong values are deleted and the │ │ right ones put in their place. │ │ Worked example, 19 Oct. 2026: energy price index values for the │ │ Baltimore-Columbia-Towson, MD area were published incorrectly. "The │ │ incorrect values were removed from the database on October 19, 2026." │ │ │ └─◆ 5A(ii). THE PUBLICATION IS NOT. ◈ ✕ │ Same entry, next sentence: "The area news releases will not be │ reissued with corrected data." │ This is the finding this map exists to show. The correction reaches │ the database a researcher queries and does not reach the document a │ reporter quoted. Both still sit on the same website. No rule in any │ of the three instruments requires otherwise, and no rule requires the │ superseded figure to be preserved beside the correction either, so a │ reader who saved the original number has no way, from the record, to │ learn that it changed. │ [ BLS errata entry of 19 Oct. 2026; corrections to the database │ scheduled 10 Nov. 2026. ] │ ├──▶ 6. THE RETRACTION: A PRODUCT IS WITHDRAWN ✕ │ The strongest instrument on the map, and the one with the least process │ attached. The agency deletes the product. │ │ Worked example, 30 Sep. 2026: "National rates by occupation for nonfatal │ days away from work (DAFW) were incorrect for 2018." Withdrawn: all national │ occupational rate data by occupation in the database, seven published tables │ (R97, R98, R99, R100, R103, EH2 occupation section, EH5) and four charts. │ "There was no impact on the DAFW count estimates." │ ⊘ The notice does not say the series will be reissued, gives no date for a │ replacement, and does not say the withdrawn figures remain available │ anywhere. Eight years of a published rate series left the record on one │ agency's own motion, by a page on its own website. │ ├──▶ 5B. SOMEONE OUTSIDE FOUND IT: A REQUEST FOR CORRECTION │ The one track with countable deadlines, an appeal, and a decider other than │ the author — and it is not a statistics provision at all. The Information │ Quality Act is § 515 of an appropriations act (Pub. L. 106-554), codified │ only as a note to 44 U.S.C. § 3516, implemented by OMB guidelines at │ 67 FR 8452 (22 Feb. 2002) and tightened by OMB Memorandum M-19-15 │ (24 Apr. 2019). │ │ │ ├─▶ 5B(i). FILED. Covers information disseminated on or after 1 Oct. 2002, │ │ "regardless of when the agency first disseminated the information." │ │ Agencies "usually forward RFCs and appeals requests to the office │ │ that released the information at issue" — so the first reader of the │ │ complaint is, in practice, its subject. │ │ │ ├─▶ 5B(ii). ☐ ANSWERED WITHIN 120 DAYS. The only hard number anywhere in │ │ this subject: agencies "will not take more than 120 days to respond │ │ to an RFC without the concurrence of the party that requested" it. │ │ ⚑ It comes from a 2019 OMB memorandum, not from the statute and not │ │ from any of the three statistical instruments. The 2002 guidelines │ │ set no number, requiring only "appropriate time periods." │ │ │ ├─▶ 5B(iii). DECLINED, IN PART OR WHOLE. Agencies "may reject claims made │ │ in bad faith or without justification," and undertake "only the │ │ degree of correction that they conclude is appropriate." The process │ │ is "designed to challenge technical information that drives policy, │ │ not to debate the policy itself." │ │ │ ├─▶ 5B(iv). ◇ APPEAL — AND HERE THE DESIGN IS GOOD. ★ │ │ Reviewers must not have been "involved in the review and initial │ │ response," must be "independent of the staff who prepared the initial │ │ response," and "sufficiently senior that they are effectively able to │ │ disagree" with colleagues. Some agencies use a panel of senior │ │ managers. The purpose is stated plainly in the 2002 guidelines: the │ │ disseminating office must not have "responsibility for both the │ │ initial response and resolution of a disagreement." │ │ ⊘ No deadline governs the appeal. The 120 days covers the first │ │ answer; nothing in the sources surveyed caps the second. │ │ │ └─▶ 5B(v). OMB SEES THE ANSWER BEFORE THE REQUESTER DOES. ⚑ │ "Agencies should share draft responses to RFCs and appeals with OMB │ prior to release to the requestor." A step the requester cannot see, │ on the only track that gives the requester anything. │ ├──▶ 7. ◇ WHO MAY ACTUALLY DECIDE? ⚑ ★ │ The column the whole subject turns on, and it comes out close to empty. │ │ │ ├─◆ THE RELEASING AGENCY. Everything in stages 5A and 6 is the agency's │ │ own call, and 5 CFR 1321 strengthens that hand: a statistical unit │ │ "must be able to determine what statistical products to disseminate │ │ without interference from parent agencies" (§ 1321.5(a)), with │ │ authority over "the content, functionality, appearance, and layout of │ │ its website" (§ 1321.4(f)). ◎ │ │ │ ├─◆ THE PARENT DEPARTMENT. Not displaced. OMB declined to make the Chief │ │ Statistician a mediator, expecting disagreements to "resolve … within │ │ the hierarchy of the parent agency" and "the head of the parent agency │ │ to retain responsibility for final determinations, when appropriate." │ │ OMB also declined to bar outside *review* before release, saying │ │ "there are times when it is appropriate for a review to occur." │ │ │ └─◆ OMB. Two powers, neither of them a power to correct: │ § 1 of SPD No. 3 — the Administrator for Information and Regulatory │ Affairs "will determine … the data series and estimates to be │ designated as principal Federal economic indicators," reviewed │ annually. OMB decides what is covered, not what is true. │ § 7 of SPD No. 3 — before any action that may violate the directive, │ an agency head "shall consult with the Administrator," who "may │ authorize exceptions to the provisions in sections 2, 3, 4, 5, and │ 6." The release calendar, the method-change notice and the whole of │ the revisions section are waivable by one official. ★ │ │ ⊘ So: no instrument surveyed names an official with the power to declare a │ published federal statistic wrong. The agency does it by practice; the │ department can intervene through its own hierarchy; OMB can designate and │ can waive. The correction itself is nobody's stated duty. │ ├──▶ 8. THE SEAT THAT HOLDS THE WAIVER ⚑ │ Both OMB powers at stage 7 run to one office — and that office has not had │ a Senate-confirmed occupant since January 2025. │ Administrator of OIRA .......... Dominic Mancini (acting), since 4 Mar. 2026 │ preceded by ................... Jeffrey Clark (acting), 5 Mar. 2025 – │ 4 Mar. 2026 │ last confirmed Administrator .. Richard L. Revesz, Jan. 2023 – │ 20 Jan. 2025, who signed the 2024 │ revision of SPD No. 3 │ Chief Statistician of the U.S. .. Stuart Levenbach, since Jan. 2026 │ [unconfirmed beyond Feb. 2026] │ — appointed by the OIRA Administrator under 44 U.S.C. § 3504(e)(7); │ not Senate-confirmed; heads the Statistical and Science Policy branch │ of OIRA and chairs the Interagency Council on Statistical Policy │ Commissioner of Labor Statistics Brett Matsumoto, 17th Commissioner, │ sworn 11 Aug. 2026, Senate-confirmed │ 7 Aug. 2026 │ preceded by ................... Erika McEntarfer, Jan. 2024 – Aug. 2025; │ then William J. Wiatrowski (acting), │ Aug. 2025 – Aug. 2026 │ ◈ Three Chief Statisticians in the thirteen months to Jan. 2026, and no │ formal public announcement of the third; the sources surveyed do not │ establish who holds that office today. │ ├──▶ 9. THE FIGURE THAT CANNOT BE CORRECTED BECAUSE IT WAS NEVER COLLECTED ✕ │ The case outside every track above, and the one the instruments cannot │ reach: there is no wrong number to fix, because there is no number. │ ├─ Consumer Expenditure Survey: "no data were collected in October and │ │ November 2025." Collection stopped 1 Oct. 2025 and resumed 5 Dec. 2025. │ │ Fourth-quarter 2025 microdata records will represent "only … one third" │ │ of the usual number of consumer units. │ ├─ Household survey: the October 2025 release "was canceled altogether," │ │ with the loss of all of that month's data, and the September and November │ │ collections delayed in processing. │ └─ ◈ The damage outlasts the gap. Response rates fell by almost five │ percentage points from the September to the November 2025 surveys, and of │ thirteen observable post-shutdown cohorts, "all but one are below trend." │ Households that entered the sample before the lapse age out by January │ 2027 — until then they partly mask the shortfall. │ [ BLS, Monthly Labor Review, 26 Jun. 2026; Federal Reserve Bank of Atlanta, │ Policy Hub macroblog, 9 Apr. 2026. ] │ └──▶ 10. THE ACCOUNTABILITY LOOP, AND WHERE IT LEADS § ⚑ Two reporting duties exist. Neither reports a correction to the public. ├─ SPD No. 3 § 8: the releasing agency evaluates its own performance, │ addressing "the magnitude and direction of all revisions," and files it │ with OMB's statistical policy office every three years. The Administrator │ reviews each one, and they are summarised in the annual report to Congress │ under § 3514 of the Paperwork Reduction Act. § │ — Note the measure: revisions, counted in magnitude and direction. Errors │ are not a category in it. ├─ IQA guidelines § IV.6: each agency reports annually to the Director of OMB │ on "the number and nature of complaints received by the agency regarding │ agency compliance," and how they were resolved, due by January 1. │ ⊘ Individual requests need not be docketed publicly. The one track with a │ deadline and an appeal is also the one whose caseload the public learns │ about only in aggregate, a year late, in a report to OMB. └─ 5 CFR 1321 § 1321.9: the parent agency's inspector general reviews compliance, beginning no sooner than two years after issuance and at least once every three years — but the review "would not extend to review of the quality of the RSAU's statistical products or methodology." ⊘ ═══ THE SHAPE OF IT ═══════════════════════════════════════════════════ A revision that was planned has a calendar, a consolidation rule, a three-year performance review and a line in a report to Congress. A number that is simply wrong has an agency web page, no deadline, and a correction that may stop at the database and never reach the release. The only route with a clock, an appeal and an independent reviewer was written into an appropriations act in December 2000, for all federal information, by people who were not thinking about statistics at all. ═══════════════════════════════════════════════════════════════════════ ``` ## Notes 1. **Two tracks, one vocabulary.** The decisive fact about this subject is that the word "revision" covers both a change scheduled years in advance and a change forced by a mistake. Readers cannot tell them apart from the notice, and the instruments do not help: [[Statistical Policy Directive No. 3]] distinguishes "routine" from "other than routine" revisions by example rather than definition (§ 6(e), § 6(f)), and offers no separate category for an error. 2. **Where the correction stops.** The clearest evidence on the map comes from the agencies' own errata notices. In the entry of 19 October 2026 concerning area energy price indexes, the incorrect values were removed from the database, and the notice states in the same breath that the area news releases "will not be reissued with corrected data." Nothing in [[Statistical Policy Directive No. 1]], [[Statistical Policy Directive No. 3]] or the 2024 Trust Regulation requires a publication to be corrected once the database has been, and nothing requires the superseded figure to be retained alongside the new one. 3. **The 120 days come from a memorandum.** The only countable deadline in the whole subject — the cap on an agency's answer to an outside correction request — appears in OMB Memorandum M-19-15 of 24 April 2019, not in the [[Information Quality Act]] itself and not in the 2002 implementing guidelines, which required only "appropriate time periods." A deadline created by memorandum can be withdrawn by memorandum. 4. **The appeal is the best-designed step on the map, and the least used.** An appeal must go to reviewers who were not involved in the first answer, are independent of the staff who wrote it, and are senior enough to overrule them. That is a stronger guarantee than anything the statistical directives contain. It is also available only on the outside-complaint track, and agencies are instructed to share draft responses with [[Office of Management and Budget]] before sending them to the requester. 5. **Retraction has less process than correction.** Withdrawing a published series requires no notice period, no replacement date and no statement of where the withdrawn figures went. The removal of the 2018 national occupational injury rate tables on 30 September 2026 — seven tables, four charts and the database series — was effected by an errata page, and the notice promises no reissue. 6. **The waiver seat is vacant in the formal sense.** Section 7 of [[Statistical Policy Directive No. 3]] lets the Administrator of the [[Office of Information and Regulatory Affairs]] authorise exceptions to the release-schedule, method-change, pre-release and revisions provisions. Section 1 gives the same official the power to decide which series count as principal economic indicators at all. That office has been filled in an acting capacity since 20 January 2025 — by [[Jeffrey Clark]] to 4 March 2026 and by [[Dominic Mancini]] since — and the [[Chief Statistician of the United States]], whom the Administrator appoints, changed hands twice in 2025 and 2026, most recently to [[Stuart Levenbach]] in January 2026 with no formal public announcement. No public source found in this survey establishes who holds the Chief Statistician's office as of today, which is why the chart marks it. 7. **The uncollectable case.** Stage 9 is not a correction problem and is included because it is the limit of the whole system. Where the 2025 lapse in appropriations stopped collection outright — two months of the Consumer Expenditure Survey, the whole of the October household survey — no revision or correction can recover the figures, and the residue shows up in response rates for more than a year afterwards. The relevant page for the lapse itself is [[Government Shutdown]]. 8. **What the inspector-general review does not cover.** The compliance reviews required by the 2024 Trust Regulation reach a statistical unit's compliance with the regulation, and expressly do not reach "the quality of the RSAU's statistical products or methodology." The one recurring outside audit of a statistical agency is therefore not an audit of its numbers. 9. **Deliberately left open.** Three questions this map raises and does not answer. Whether an agency's refusal of a correction request is reviewable in court at all — the sources surveyed are silent on judicial review and on whether § 515 creates a private right of action, and this map does not assert either way. Whether the fifty states have anything comparable for their own official statistics, which no source in this survey counts. And how the thirteen principal statistical agencies differ from one another in correction practice, which would need each agency's own information-quality guidelines read side by side; the [[Statistical Officials Crosswalk]] is the nearest existing comparison. 10. **Officeholders (verified November 2026).** Commissioner of Labor Statistics: Brett Matsumoto, sworn 11 August 2026 after Senate confirmation on 7 August 2026, following [[Erika McEntarfer]] (January 2024 – August 2025) and a year of acting leadership. Administrator of OIRA: [[Dominic Mancini]], acting, since 4 March 2026. Chief Statistician: [[Stuart Levenbach]] since January 2026, marked unconfirmed beyond February 2026 for the reason in note 6. ## See also - [[Federal Statistical System Tree]] - [[How a Federal Survey Is Approved]] - [[How a Statistical Agency Is Defunded or Discontinued]] - [[Federal Disclosure Avoidance Crosswalk]] - [[Statistical Officials Crosswalk]] - [[Statistical Policy Directive No. 1]] - [[Statistical Policy Directive No. 3]] - [[Information Quality Act]] - [[Chief Statistician of the United States]] - [[Dominic Mancini]] - [[Office of Information and Regulatory Affairs]] - [[Office of Management and Budget]] - [[Bureau of Labor Statistics]] - [[Bureau of Economic Analysis]] - [[Census Bureau]] - [[Consumer Price Index]] - [[GDP]] - [[Paperwork Reduction Act]] - [[Foundations for Evidence-Based Policymaking Act]] - [[Interagency Council on Statistical Policy]] - [[Federal Agencies Hub]] - [[Foundations Hub]]