When the auditors said the nursing home numbers were wrong, they were only half right

Last month, federal investigators delivered a blunt verdict on the data families use to choose nursing homes: a lot of it is not backed up. For a single month in 2024, nursing homes reported roughly 938,000 registered-nurse hours to the government they could not support with payroll records. That is about one in twenty RN hours, touching more than 40% of the nurses in the system.

It is a significant finding. It is also only half the story.

Two years ago, I looked at the same government staffing data and asked a different question. Forget for a moment whether the reported hours are real. Even if every hour were perfectly documented, does the way the government summarizes them tell you anything useful about what a resident experiences on a Saturday?

The answer was no. That analysis — Working Paper No. 01 — found that nursing homes staff at roughly 3.96 nurse hours per resident on Wednesdays and 3.24 on Sundays, an 18% drop that the quarterly average on Nursing Home Compare does not show.

This paper extends that analysis through 2024 and asks the next question: does the accuracy problem the OIG found and the distribution problem that paper found operate independently, or do they reinforce each other?

The answer is both, and the distinction is worth understanding.

The cliff got worse before it got better

The 16% weekend staffing drop documented for 2017–2020 did not resolve after COVID. It widened.

During 2021 — the worst year of the nursing staffing crisis — the national median weekend drop reached 18.01% in a single quarter. When nurses were scarce, the discretionary weekend shift was the first thing cut. Weekday coverage was largely maintained. Weekend coverage fell. The structure of the drop, not just its level, is where crisis shows up.

By 2024, the drop had declined to 15.88% — the lowest annual average in the eight-year record. The most recent quarter in the dataset, 2024Q4, shows 15.12%: the smallest weekend drop of any quarter since 2017. Whether that represents a genuine improvement or statistical noise, this analysis cannot determine. It is worth watching.

The RN floor is thinner than the headline suggests

The federal auditors focused specifically on registered-nurse hours. That turns out to matter for understanding the weekend problem too.

Total nurse staffing drops about 16% on weekends. RN staffing drops about 46%.

On a typical weekday, the median nursing home provides about 0.66 registered-nurse hours per resident. On a typical weekend, that falls to about 0.35 — nearly half. The 938,000 unsupported RN hours the OIG identified do not come off a flat baseline. They sit on top of a floor that has already been cut nearly in half by the day-of-week pattern.

The auditors asked whether the reported hours are real. This data shows that the hours are thin on weekends before anyone has adjusted for reporting errors. Both statements can be true simultaneously.

Are the two problems connected?

The two problems are not correlated at the behavioral level. Facilities that inflate their hours at quarter-end — a recognized gaming signature that regulators watch for — do not show systematically steeper weekend cliffs. The correlation across all 29 quarters is near zero (-0.016). Facilities with suspiciously flat hour-reporting are essentially absent from the data, never exceeding 0.01% of reporters in any quarter.

What this means is that the weekend cliff is not a problem of gaming facilities. It is a structural feature of how the industry schedules labor, present in facility after facility regardless of how accurately they report.

But the problems do compound at the level of consequence. The floor that unsupported RN hours sit on is already thin. If those missing hours concentrate on weekends — which is the adversarial possibility the OIG’s audit design could not test, because it did not track hours by day of week — the floor is thinner still.

Fixing the accuracy problem will not fix the distribution problem. Fixing the distribution problem will not fix the accuracy problem. They require two separate interventions, and right now neither is being actively overseen.

What neither audit found

The OIG checked whether the hours are real. This paper checked whether real hours are summarized honestly. Neither checked whether the hours, real or not, are adequate on the days that matter most.

The federal staffing minimum of 3.48 HPRD was blocked from enforcement until 2034. No regulator currently monitors the within-week distribution of staffing. CMS has collected this daily data since 2016 and still publishes a quarterly average on the website families use to compare facilities.

The auditors went looking for unsupported hours. They found them. They did not look at whether those hours, and the accurate ones surrounding them, are distributed in a way that leaves residents exposed on weekends.

That is the question this paper answers. And the answer is the same one it was in 2024, now extended through seven more years and the worst staffing crisis in recent memory: the weekly staffing pattern is a second failure mode that exists alongside the accuracy problem, independent of it, and not addressed by any current regulatory response.

The rating on the website was calculated on a Wednesday. The auditors checked whether Wednesday’s hours were real. Nobody checked what happens on Saturday.

Data: CMS Payroll-Based Journal Public Use Files, 2017Q1–2024Q4. 29 quarters, 21.5 million facility-days. Anchored to OIG report A-09-24-02005 (June 17, 2026). Full technical paper, charts, methodology, and reproduction code: Working Paper No. 02. Extends findings from Working Paper No. 01 (October 2024).

@techreport{zatara2026paper2,
  title       = {Two failure modes in nursing home staffing data: accuracy and distribution},
  author      = {Zatara, David},
  year        = {2026},
  institution = {Independent Researcher},
  type        = {Working Paper},
  number      = {2}
}