Nursing Home Staffing Data: How to Check It Before a Tour

The file is 234 megabytes and it took eleven seconds to download. That is the entire barrier between a family and the day-by-day nursing staffing record of every Medicare-certified nursing home in the country — 1,303,830 rows covering January through March 2026, one row per home per calendar day, 14,487 homes, ninety days each.

Care Compare turns those ninety days into one number. The number is accurate. It is also an average, and averages are where the shape of a week goes to disappear.

Everything below was downloaded, read and computed on 14 September 2026, from the files named at the end.

Three files sit under the staffing page, and only one of them is daily

The consumer page at Medicare's Care Compare is the top layer. Under it are three separate public datasets, and families almost always stop at the first.

Provider Information is the Care Compare page as a spreadsheet — one row per home, around a hundred columns, refreshed monthly. The August 2026 release covers 14,690 homes and holds the star ratings, the reported and case-mix adjusted hours, the weekend figures and the turnover percentages. It sits at data.cms.gov/provider-data/dataset/4pq5-n9py.

PBJ Daily Nurse Staffing is the layer beneath: the quarterly payroll file, one row per home per day, hours broken into eight nursing categories and each of those split into employee hours and contract hours. The resident census sits in the same row.

PBJ Employee Detail goes one layer further — one row per person per day, with a system-generated employee number, a job code, an employee-or-contractor flag and the hours worked. No names. It starts in April 2020, later than the other two.

All three key on the same six-digit CMS Certification Number, which Care Compare displays on the facility's own page. Once you have that number the other files are a filter away, and you do not need 234 megabytes to look at one home: the data API accepts the number directly.

The hours number and the turnover number are not from the same year

This is the part I read wrong first, and it took another CMS file to show me.

CMS publishes the reporting window for every measure in a small dataset called Nursing Home Data Collection Intervals. In the 1 August 2026 refresh it says:

The hours describe one quarter of this year. The turnover percentage sitting a few lines below it on the same page describes all of last year. They overlap by nothing.

For most homes that gap is a technicality. For a home that changed hands, replaced a director of nursing, or lost a wing of staff in the last twelve months, it is the whole story — the turnover figure describes an operator who may no longer be there, and the hours figure describes one who has not been measured on turnover yet. I had been reading the two as a single portrait of the same months, which is a comfortable mistake to make because they are printed four centimetres apart.

The survey findings run on a third clock: complaint deficiencies in that refresh cover 1 August 2025 through 31 July 2026, with two prior twelve-month periods behind them. Three domains, three windows. The five-star rating stacks them anyway, which is worth knowing before reading the star as a snapshot of anything.

An 88-bed home in Arkansas, five stars, and a Sunday at 2.55

Here is what the daily file adds. The home is in Little River County, Arkansas — for-profit, 88 certified beds, 70.2 residents a day on average, no fines. Care Compare gives it five stars overall, five for health inspections, three for quality measures and four for staffing. Nothing is wrong with this home on paper, which is exactly why it makes the point.

Its published figures for January to March 2026: 4.59268 total nurse hours per resident day and 0.67528 RN hours. I recomputed both from the raw daily file and got 4.593 and 0.675. The two files agree, which is the check worth running before trusting any arithmetic of your own.

That recomputation is one division, and it is the same division behind every figure in this guide. In PBJ_dailynursestaffing_CY2026Q1.csv, keep the rows whose PROVNUM is the home's certification number and whose WorkDate falls in the stretch of days you are asking about. Add the eight hour columns — Hrs_RNDON, Hrs_RNadmin, Hrs_RN, Hrs_LPNadmin, Hrs_LPN, Hrs_CNA, Hrs_NAtrn, Hrs_MedAide — across those rows, and divide by the sum of MDScensus across the same rows. RN hours are the first three of the eight. Change which rows you keep and you change the question.

So split the same ninety days by day of the week:

Weekdays 5.19, weekends 3.12. The lowest single day in the quarter was Sunday 4 January at 2.55; the highest was Tuesday 10 February at 5.94. RN coverage on a Wednesday runs at roughly three times the Sunday level.

Weekend staffing is published, but only after case-mix adjustment: the provider file gives this home 3.89215 for the weekend against 5.72383 for the quarter, a gap of 32 percent. A weekday figure is published nowhere, in either form. The unadjusted weekend average has to be computed — 3.12298 total and 0.29859 RN, both matching the arithmetic above — and against the 5.19 the weekdays actually ran, the swing is 40 percent rather than 32.

That shape is not rare. Across all 14,487 homes in the quarter, the median weekday figure is 3.849 and the median weekend figure 3.264; take each home's own drop and the median of those is 14.9 percent. 2,591 homes dropped 20 percent or more. 652 dropped 25 percent or more, and 150 of those 652 hold a five-star overall rating.

The 3.89 and the 5.72 deserve a sentence of their own, because they are the pair the staffing star is scored on. Case-mix adjustment scales the raw hours by how much nursing this home's own residents are assessed to need; its nursing case-mix index is 1.12476, and that is what turns 3.12 into 3.89 and 4.59 into 5.72. The adjusted figure answers whether the hours were enough for these particular residents. The unadjusted figure answers how many people were in the building. A tour is mostly a question about the second one.

Counting people instead of hours changes what the gap feels like

Hours per resident day is a ratio, and ratios are easy to read past. The Employee Detail file lets you count heads instead.

Same home, same quarter, nursing job codes only — RN director of nursing, RN with administrative duties, RN, LPN with administrative duties, LPN, certified nurse aide, nurse aide in training, medication aide:

Census on both of those January days was 73.

Across the whole quarter, 76 distinct people appear under a nursing job code at this home, and 7 of them are RN-coded. That is the pool the schedule is drawn from. Seven registered nurses covering ninety days of a requirement that an RN be present at least 8 consecutive hours a day, seven days a week, is a thin bench, and the Sunday rows are what that thinness looks like on a calendar.

The same file carries an employee-versus-contractor flag. At this home 69 of 3,535 rows are marked contract, and not one of them sits under a nursing job code — its nursing hours are all its own payroll. Nationally the median home reported 0.3 percent of nursing hours as contract in this quarter and 6,219 homes, this one among them, reported none at all; the top one percent ran 44 percent or more of their nursing hours through agencies. That figure moves fast when an operator gets into trouble, and it is one of the few numbers in the public record that changes quarter to quarter rather than year to year.

An hour in this file is not an hour someone was in the building

Before treating any of these numbers as literal, it helps to know the counting rules. They are set out in the PBJ Policy Manual, version 2.8, August 2026.

• Meal breaks come out whether or not they are taken. The manual: "For each full shift that staff (exempt, non-exempt, or contract) are paid to work, a 30-minute meal break must be deducted from their shift (whether or not the employee actually takes a meal break)." A twelve-hour shift is reported as 11.5.
• Hours are rounded to tenths, not recorded to the minute. The manual is explicit that 7 hours and 33 minutes must not be reported as 7.33 — the correct entry is 7.6, or 7.55 if the facility rounds to hundredths.
• Midnight cuts the shift in two. A shift running 11pm to 7am is reported as 1 hour on the first date and 7 on the second, so night coverage never appears as a single block.
• Leave and unpaid time are excluded, and a salaried person who works ten hours but is paid for eight is reported at eight.
• Universal care workers get split. A CNA who also cooks or cleans must have that time separated by "a reasonable methodology" and reported under the other category.
• The census is not the facility's count. CMS derives the daily resident census from MDS assessments, not from anything the home submits alongside its payroll.

Submissions are due by the end of the 45th calendar day after the quarter closes, 11:59pm Eastern. The last accepted submission before that moment is final.

Five footnote codes that mean "this is not a measurement"

The Provider Information file carries a footnote column beside each staffing figure, and those codes are worth memorising, because a footnoted star is a different object from a low star. From the nursing home data dictionary, updated July 2026, with counts from the 1 August 2026 file:

Code 24 shows up in the raw file too. Counting days where the census was above zero and every RN category was zero, 1,169 homes had at least one such day in January to March 2026, and 283 had four or more. Of those 283, 273 carry a one-star staffing rating with footnote 24 — the raw file and the published rating land on the same homes, which is a reasonable sign that the day-level arithmetic here is doing what CMS's is.

Turnover footnotes are far more common than staffing-hour footnotes. In that same file, 1,153 homes have no usable total nursing turnover figure and 1,979 have no usable RN turnover figure. If a home's turnover cell looks blank or oddly generous, read the footnote before reading the number.

There is no federal hours-per-day figure to compare it against any more

For two years there was going to be one. There is not now.

The 2024 minimum staffing standards for long-term care facilities were repealed by an interim final rule, 90 FR 55687, published 3 December 2025 and effective 2 February 2026. CMS's own summary of that rule states the action was taken "in view of changes made by public law, which precludes HHS from implementing, administering, or enforcing certain provisions of the final rule until September 30, 2034."

What is left in 42 CFR 483.35, read through the eCFR versioner API on 14 September 2026 against the title 42 text issued 13 August 2026 and current through 10 September 2026:

• sufficient numbers of licensed nurses and other nursing personnel "on a 24-hour basis" — § 483.35(a)(1);
• a licensed nurse designated as charge nurse "on each tour of duty" — § 483.35(a)(2);
• a registered nurse "for at least 8 consecutive hours a day, 7 days a week" — § 483.35(b)(1);
• a registered nurse serving as director of nursing on a full-time basis — § 483.35(b)(2).

Two of those can be waived, on different grounds. Paragraph (e) lets a state release a nursing facility from the 24-hour licensed nurse and the 8-hour RN requirements if the facility shows it could not recruit despite diligent efforts, "including offering wages at the community prevailing rate," and the state finds the waiver will not endanger anyone; it is reviewed annually. Paragraph (f) is narrower and belongs to the Secretary: a rural skilled nursing facility, in an area short of skilled nursing capacity, with one full-time registered nurse on duty 40 hours a week, can be released from providing RN services beyond those 40 hours. Under both, the facility must notify residents and their representatives, and notice goes to the Office of the State Long-Term Care Ombudsman. So if a rural home's RN rows look unusually thin, a waiver is one of the explanations, and the ombudsman is the office that would know — the federal Eldercare Locator gives you the number for the right one.

Absent a federal floor, the honest comparison left is to other homes. From the August 2026 file: the median home reports 3.681 total nurse hours and 0.584 RN hours per resident day. The tenth percentile is 3.064 and 0.302; the ninetieth is 4.849 and 1.134. Median total nursing turnover is 44.9 percent, median RN turnover 41.7 percent. State medians for total hours run from 3.27 in Texas and 3.29 in Illinois up through 4.61 in Hawaii and 4.95 in Oregon to 6.82 in Alaska — though the Alaska median rests on 16 homes and the Texas one on 1,141, which is its own reason not to read a state figure as a standard. Measuring a home against a national median can flatter or punish it purely by geography. A number of states also set their own minimums in licensing law. That is a state question, and the state survey agency or licensing board is where to put it.

The sheet by the front door is today's, and they have to keep 18 months of them

Everything above describes a quarter that closed months ago. There is one staffing record that describes this morning, and it is required to be hanging on a wall.

Under § 483.35(g), a nursing home must post daily, at the beginning of each shift, the facility name, the current date, the resident census, and the total number and actual hours worked by registered nurses, licensed practical or vocational nurses, and certified nurse aides. It must be posted in a "clear and readable format" and "in a prominent place readily accessible to residents and visitors."

Two further sentences in that paragraph are the useful ones. The facility must, "upon oral or written request, make nurse staffing data available to the public for review at a cost not to exceed the community standard." And it must keep the posted daily data for at least 18 months, or longer where state law requires it.

So a tour has a question in it that costs nothing. Ask to see the posted sheet. Write down the census and the three hour figures on it. Then ask to review the same sheets for the four most recent Sundays. You are entitled to review them, the home is required to still have them, and the answer arrives while you are standing in the lobby rather than five months later in a spreadsheet.

If the visit is happening because a hospital discharge planner named a date, the staffing file and the inspection report's F-tags are the two documents worth reading before a bed is accepted — and the coverage clock that starts on the day of admission is a separate question with deadlines of its own.

Sources

• Payroll Based Journal Daily Nurse Staffing, CMS open data, file PBJ_dailynursestaffing_CY2026Q1.csv — 1,303,830 rows, 14,487 facilities, 90 dates, dataset modified 29 July 2026. Downloaded and analysed 14 September 2026. Every national figure above is a group-by on PROVNUM of the eight Hrs_ columns over MDScensus, cut by the weekday of WorkDate.
• Payroll Based Journal Employee Detail Nursing Home Staffing, CY2026Q1 — retrieved 14 September 2026.
• Provider Information, the Care Compare provider-level file, NH_ProviderInfo_Aug2026.csv, processing date 1 August 2026, 14,690 homes across 99 columns — downloaded 14 September 2026. Source of the star ratings, the reported and adjusted hour columns, the nursing case-mix index, the turnover percentages and every footnote count above.
• Nursing Home Data Collection Intervals, dataset qmdc-9999 in the CMS provider data catalog, 1 August 2026 refresh — retrieved 14 September 2026.
• Nursing Home Data Dictionary, updated July 2026, Table 15 (footnote codes) — read 14 September 2026.
• Electronic Staffing Data Submission Payroll-Based Journal Long-Term Care Facility Policy Manual, Version 2.8, August 2026, CMS — read 14 September 2026, linked from the CMS staffing data submission page.
• 42 CFR 483.35 (nursing services, including the daily posting requirement) and 42 CFR 483.70(p) (mandatory submission of staffing information based on payroll data) — read through the eCFR versioner API on 14 September 2026, title 42 issued 13 August 2026, current through 10 September 2026.
• Medicare and Medicaid Programs; Repeal of Minimum Staffing Standards for Long-Term Care Facilities, 90 FR 55687, interim final rule with comment period, published 3 December 2025, effective 2 February 2026 — read 14 September 2026.
• Medicare Care Compare — a home's CMS Certification Number appears on its Care Compare page and is the key to every file above.

The Arkansas home is described by county rather than by name. Nothing here is a finding about that facility; it is a worked example, and the same method runs on any certification number.

This page is general information, not legal, medical, or financial advice. See the terms.