Census's Veteran Reps Missed Fraud-Check Quotas 99% of the Time
Summary
A March 10, 2026 Commerce Department inspector general audit found the Census Bureau did not effectively implement its data collection and quality control procedures for the American Community Survey (ACS) -- the mandatory survey the Census Bureau says is used to distribute more than $2.8 trillion in federal funding a year. The reinterview program built to catch field-worker data falsification fell short across every measure OIG checked, and 21% of a sampled batch of "vacant" home codings from 2022 turned out to be wrong.
A fraud check most reps didn't clear
Reinterviews are how the Census Bureau checks whether an FR's reported data is real: a trained bureau employee calls back a sample of the FR's respondents and compares the answers. The bureau sets a monthly reinterview-completion quota for each FR based on experience level and the housing type they cover. Reviewing 2022 performance against those quotas, OIG found⧉ that only 135 of 745 (18%) inexperienced field reps working housing units met their goal, only 30 of 3,000 (1%) experienced housing-unit reps met theirs, and only 40 of 130 (31%) reps working group quarters -- nursing homes, dorms, prisons, and similar facilities -- met theirs. Combined across both housing-unit tiers, that's 165 of 3,745 field reps, or about 4.4%, hitting their quota (this outlet's arithmetic on 's separate figures; see analysis.json).
View data as table
| HU Inexperienced | 18% | 135 of 745 FRs met goal; 82% did not |
|---|---|---|
| HU Experienced | 1% | 30 of 3,000 FRs met goal; 99% did not |
| GQ (all) | 31% | 40 of 130 FRs met goal; 69% did not |
The check that skips group quarters almost entirely
The gap isn't only about quota math -- found the reinterview program's design itself has a structural hole. The bureau's reinterview goals for group quarters (GQ) apply only to the facility administrator's answers, never to the residents living there. OIG found that 215,000 of the 235,000 (91.5%) resident-level interviews⧉ conducted at GQ facilities in the 2022 reinterview sample were excluded from any reinterview at all, because bureau staff "assumed residents would be difficult to reach." Of the reinterviews the bureau did complete, 16% -- across a reviewed sample of 175 cases -- were inconclusive, meaning the bureau could not determine whether the FR had falsified data or not. And nearly half (46%) of the reinterviews that did produce a conclusive result were completed more than the bureau's own two-week recall window after the original interview -- in remote Alaska, that share was 96%.
It has happened before
This isn't a hypothetical risk. OIG's report recounts⧉ that after the 2017 ACS data release, the bureau discovered a field rep had been falsifying data -- contacting households, asking survey questions, then manipulating the responses to skip additional in-depth questions. That FR completed the reinterview process without the falsification being caught; the bad data made it into the published ACS estimates for New Castle County and the State of Delaware, and the bureau had to apply an editing and imputation fix in 2018 before releasing the following year's 1-year estimates.
To catch that kind of pattern going forward, the bureau's American Community Survey Office built a tool that flags an FR's data as anomalous compared to peers and routes an inquiry to the FR's regional office for follow-up. sampled 60 of those flagged inquiries and found regional offices responded late to 42 of the 55 it could review (76%)⧉ -- and projects that regional offices answered roughly 290 of all 415 such inquiries (69.9%) after the requested two-week due date.
View data as table
| Coded correctly | 58.4% | ~73,000 of 125,000 cases |
|---|---|---|
| Miscoded | 23.2% | ~29,000 of 125,000 cases |
| May have been miscoded | 18.4% | ~23,000 of 125,000 cases |
The data itself, not just the checks on it
didn't only test the bureau's fraud checks -- it also tested a slice of the underlying data. Reviewing a sample of cases FRs coded as "vacant" (meaning nobody was found living there, which excludes the case from the survey's population estimates), OIG found FRs miscoded 18 of 86 sampled non-remote Alaska cases (21%)⧉ -- cases that should have been logged as a different kind of noninterview instead. Projected across the full 2022 count of 125,000 such codings, estimates about 29,000 (23.2%) were miscoded outright, plus another 23,000 (18.4%) flagged as possibly miscoded because the FR's interview notes were too thin to confirm either way -- together, up to 41.6% of that batch with an identified problem (this outlet's combination of 's two figures; see analysis.json).
Separately, FRs skipped the required step of contacting a recommended contact⧉ -- an owner, landlord, or similar -- for nearly half of sampled vacant-unit interviews, relying instead on neighbors or, in one case, a postal worker.
The takeaway
- The Census Bureau's fraud check for the survey behind $2.8 trillion in federal funding missed its own targets almost everywhere looked in 2022: just 1% of experienced field reps met their reinterview quota, and 91.5% of group-quarters resident interviews were excluded from any reinterview at all.
- This has already gone wrong once: a field rep falsified ACS data in 2017 that made it past the reinterview process and corrupted published estimates for New Castle County and Delaware, requiring a 2018 editing fix. found the bureau's follow-up tool for catching similar cases now gets a timely regional-office response only about 30% of the time.
- A sample of the underlying data showed real errors, not just process gaps: 21% of sampled 2022 "vacant" home codings were wrong, projecting to roughly 29,000-52,000 of 125,000 such codings bureau-wide with a confirmed or possible error. made 17 recommendations; the Census Bureau concurred with all of them.
Figures are from the U.S. Department of Commerce Office of Inspector General's audit report "Audit of Data Collection and Quality Control Procedures for the U.S. Census Bureau's American Community Survey"⧉ (Report No. -26-010-A, March 10, 2026) and Commerce 's Semiannual Report to Congress, March 2026⧉ (used only for institutional context), both fetched directly from oig.doc.gov and read in full. Wayback captures exist for both documents. The audit covers ACS data collection conducted in calendar year 2022; the Census Bureau's own response, attached to the audit as Appendix 5, concurred with all 17 of 's recommendations.
The combined 41.6% vacant-interview exposure share, the 4.4% combined housing-unit reinterview-completion rate, and the 34.2% combined supervisory-non-review rate are this outlet's own arithmetic on figures 's report states separately (methods and caveats in analysis.json); 's report does not itself state these combined figures. No individual field representative, supervisor, or bureau official is named in this reporting; 's report itself does not name the FR involved in the 2017 Delaware falsification. A blind adversarial verifier, working from the primary documents alone with no access to this draft, independently checked every itemized fact; see verification.json.
Sources(2) ▾
- U.S. Department of Commerce, Office of Inspector General, Office of Audit and Evaluation, Audit of Data Collection and Quality Control Procedures for the U.S. Census Bureau's American Community Survey (Report No. OIG-26-010-A) (2026-03-10) — Source for every figure on the American Community Survey's 2022 data-collection and quality-control failures: the ISR (internet self-response) testing-documentation gap, the vacant-interview miscoding rates, the supervisory-review gaps, the reinterview selection and completion goal misses, the group-quarters reinterview exclusion, the inconclusive/late reinterview rates, the FR Data Tool inquiry response-time data, the 2017 Delaware falsification precedent, and the $2.8 trillion federal-funding-use context (citing the Census Bureau's own June 2023 report). oig.doc.gov · original document
- U.S. Department of Commerce, Office of Inspector General, Semiannual Report to Congress, March 2026 (October 1, 2025-March 31, 2026) (2026-04-27) — Source for institutional context: this ACS audit (-26-010-A) is one of the 10 audit, evaluation, and inspection reports Commerce issued in the six months ending March 31, 2026, and the semiannual report's own framing places the ACS's data-reliability problems inside the Department's broader 'Providing Core Services and Data' challenge area, alongside the bureau's preparations for the 2030 Census. oig.doc.gov · original document
Comments
Always open. Logged-in readers can annotate paragraphs in place.
The American Community Survey (ACS) is the Census Bureau's mandatory, always-running household survey -- the source for the age, income, commute, and housing data that the bureau itself says⧉ steers more than $2.8 trillion in federal funding to states, tribes, and communities each year. A March 10, 2026 audit from the Commerce Department's inspector general⧉ found the bureau "did not effectively implement its data collection and quality control procedures to ensure that ACS estimates were reliable." The clearest evidence: the bureau's own defense against field-worker data falsification -- a program that has each field representative's (FR's) work double-checked through a follow-up call -- missed its targets almost everywhere looked. Among the bureau's most experienced door-to-door interviewers, just 1% hit their monthly reinterview quota in 2022.