Working Paper No. 02 · July 2026
Two ways the same number lies
When federal auditors found the reported hours were wrong, they were answering half the question
The June 2026 OIG audit found 938,000 unsupported RN hours in a single month. This paper extends the weekend staffing analysis through 2024 — through the exact period the auditors examined — and asks whether the accuracy problem and the distribution problem operate independently or compound.
Preprint — not peer-reviewedThree numbers to read first
Abstract
In June 2026, the HHS Office of Inspector General reported that for March 2024, nursing homes submitted approximately 938,000 registered-nurse hours to the Payroll-Based Journal that were not supported by required documentation — about 5% of RN hours, affecting an estimated 42% of RNs. That audit asked whether reported hours are accurate. A prior analysis showed that even accurate hours, summarized as a quarterly average, misrepresent what residents experience on weekends [Zatara 2024]. This paper extends the daily PBJ analysis through 2024Q4 using 29 quarters and 21.5 million facility-days. The weekend staffing cliff widened during the 2021 staffing crisis (peak 18.01% in 2021Q2), had partially recovered to 15.88% by 2024, and reached its lowest point in the eight-year record (15.12%) in 2024Q4. RN staffing drops approximately 46% on weekends versus 16% for total nurse hours, meaning the accuracy problem OIG identified lands on a floor that is already nearly twice as exposed as the headline metric suggests. However, the two failure modes are structurally independent: quarter-end hour inflation and weekend cliff depth are uncorrelated (r = -0.016 across all quarters). The compounding is real but operates through level, not behavioral overlap, and requires two separate regulatory interventions.
1. Why this matters now
On June 17, 2026, the HHS Office of Inspector General published an audit of RN hours reported to CMS's Payroll-Based Journal for March 2024. The finding was direct: roughly 938,000 RN hours — about 5% of the total — could not be verified against required documentation. The unsupported hours affected an estimated 42% of registered nurses in the system. OIG made four recommendations; CMS concurred with two, declined one, and acknowledged it does not follow up on its staffing audits to confirm corrections.
That audit answered whether reported hours are real. A prior analysis asked a different question [Zatara 2024]: even when hours are accurate, does the quarterly average CMS publishes from them represent what residents experience? The answer was no — nurse coverage peaks midweek and drops sharply on weekends, a pattern the quarterly average erases. Forty-three percent of facilities clearing the 3.48 HPRD federal standard on their quarterly average fell below it every weekend.
This paper extends that analysis through 2024 and asks a third question: are these two failure modes independent, or do they compound? The answer is both, and the distinction matters for anyone designing a regulatory response.
2. The 8-year arc
The weekend cliff did not resolve after COVID. It widened during the staffing crisis and has only partially recovered.
Five periods in the 2017–2024 staffing story
Weekend drop % by period · OIG audited March 2024 (2024Q1: 16.05%)
Baseline 16.30%
Artifact
Crisis peak
Recovery
Below base
3. The full quarterly trend
Across 29 quarters, the crisis signature is visible: weekday HPRD was largely maintained during 2021 while weekend HPRD fell. When staff were scarce, weekend shifts were the first cut. The gap widened not because the midweek floor fell but because the weekend floor fell faster.
Weekday vs weekend HPRD — 29 quarters, 2017–2024
Shaded band: staffing crisis (2021Q1–2022Q2). Dashed line: 3.48 proposed federal minimum.
Weekend drop % by quarter — deviation from the pre-COVID baseline
Orange bars are above the pre-COVID mean (16.30%). Blue bars are below it.
4. The RN floor the OIG audit sat on
The OIG audit examined registered-nurse hours specifically. RN staffing follows a much steeper weekend curve than total nurse staffing, and the difference is not small. On a typical weekday, the median facility provided approximately 0.66 RN hours per resident. On a typical weekend, that fell to 0.35 — a reduction of nearly half.
The 938,000 unsupported RN hours OIG identified do not come off a flat baseline. They sit on top of a weekend floor that is already roughly half the weekday level. Whether a given unsupported hour is a weekday or weekend hour changes its consequence significantly.
RN-only weekend drop vs total-nurse weekend drop
Median 2017–2020. OIG audited RN hours specifically.
5. Are the two failure modes compounding?
This paper tests two behavioral proxies for accuracy problems. Neither can confirm misreporting at any individual facility. Together they answer the structural question.
Quarter-end hour inflation — inflating hours in the final weeks of a quarter — is a recognized gaming signature that regulators including Washington State explicitly watch for. If facilities that game quarter-end reporting also have steeper weekend cliffs, the two problems reinforce. Across all 29 quarters, the mean correlation between quarter-end spike magnitude and weekend cliff depth was -0.016. Facilities in the top decile of quarter-end spiking showed a mean weekend cliff of 16.95%; bottom-decile facilities showed 17.51%. The difference is 0.56 points in the opposite direction from the hypothesis.
Improbably flat reporters — facilities with near-identical hours every day — represented at most 0.01% of all reporters in any single quarter, and zero in most quarters. Not a meaningful signal.
Quarter-end spike decile vs weekend cliff depth
If the problems compound behaviorally, high spikers should show steeper cliffs. They do not.
The verdict
The two failure modes are structurally independent at the behavioral level. Fixing the accuracy problem OIG identified will not fix the distribution problem. Fixing the distribution problem will not fix the accuracy problem. They require two separate regulatory interventions. The compounding is real — the RN weekend floor is thin, and unsupported RN hours sit on top of it — but it operates through level, not through shared behavioral patterns.
What this analysis does not tell you
Whether the 2024 decline is a trend. 2024Q4 at 15.12% is the lowest quarter in the record. One year below baseline warrants watching, not a conclusion.
Whether unsupported hours cluster on weekends. OIG's audit design did not track hours by day of week. If missing hours concentrate on weekends, the actual RN floor is lower than reported data shows.
Whether the weekend cliff causes harm. This paper measures reported staffing inputs. Connecting patterns to patient outcomes requires inspection and claims data.
Accuracy of individual post-2020 facilities. OIG audited one month in one year. This paper cannot speak to facility-level accuracy for 2021–2024.
Source. CMS Payroll-Based Journal Daily Nurse Staffing Public Use Files, all quarters 2017Q1–2024Q4. Retrieved July 2026. 29 quarters, approximately 21.5 million facility-days.
Gaps. 2018Q1–Q3 files returned HTML error pages at retrieval and are excluded. The series therefore has 10 pre-pandemic quarters (2017Q1–2020Q2 as in Paper 01) plus 19 new quarters (2020Q3–2024Q4).
RN hours. Summed from columns HRS_RNDON, HRS_RNADMIN, and HRS_RN. Total nurse hours additionally include HRS_LPNADMIN, HRS_LPN, HRS_CNA, HRS_NATRN, and HRS_MEDAIDE.
EOQ spike. Facility-level ratio of mean HPRD in the final 14 days of a quarter to mean HPRD in the preceding days. Facilities with fewer than 60 valid days excluded from this analysis.
Flat reporters. Facilities with a coefficient of variation below 2% across all daily HPRD readings in a quarter, with at least 60 valid days.
OIG report. A-09-24-02005, issued June 17, 2026. All OIG figures are from that report's published findings and highlights.
All code, dated dataset snapshots, and the Python environment are in the repository.
zatara-research-data/02-two-ways-the-number-lies