Thin Staffing Tracks With More Inspection Deficiencies at Nursing Homes — a Moderate, Not Dominant, Relationship

Summary

Understaffed nursing homes do rack up more inspection deficiencies, but the relationship is real and statistically decisive while being only moderate in strength — staffing is one contributor among many, not the dominant driver. Across 13,394-14,271 U.S. CMS-certified facilities, total nursing hours per resident day (HPRD) correlate with cited health deficiencies at r = -0.15 (p<0.001): more staffing, fewer deficiencies, but staffing alone statistically 'explains' only a small share of the variation (regression R² around 0.07). The cleanest view is CMS's own composite staffing star rating: facilities rated 1-star on staffing average 11.9 health deficiencies vs. 5.3 for 5-star facilities — a 55% gap. In a regression controlling for facility size (beds) and nursing staff turnover, each additional hour of total nurse staffing per resident per day is associated with about 0.69 fewer cited deficiencies (p<0.001), and this direction and significance hold up when any single state is dropped from the data (California is the most influential single state, but excluding it makes the relationship stronger, not weaker or reversed).

What was measured

Source: health.cms_nursing_home in this corpus — CMS's Nursing Homes Including Rehab Services 'Provider Information' file (dataset 4pq5-n9py), one row per CMS-certified facility (CCN), current 2026 snapshot. It carries staffing hours-per-resident-day (nurse aide, LPN, RN, total), the CMS staffing star rating, and total health deficiencies cited on the facility's most recent standard survey — all in one file, so no second dataset needed to be joined.

Grain and window: facility-level, cross-sectional (a single point-in-time snapshot — CMS refreshes this file periodically but the corpus holds only the current vintage, so no year-over-year trend could be computed here). This limits the analysis to association, not causal timing; a longitudinal panel would be needed to test whether a staffing change at a given facility precedes a deficiency change there.

How strongly staffing tracks deficiencies

Zero-order correlations (n=13,394, all p<0.001): total nurse HPRD vs. deficiencies r=-0.148; RN HPRD vs. deficiencies r=-0.149; nurse aide HPRD vs. deficiencies r=-0.141; LPN HPRD vs. deficiencies r=+0.049 (weak positive — LPN hours behave differently from RN/aide hours, plausibly because LPN hours partly substitute for RN hours in leaner facilities). Facility size (certified beds) also correlates with more deficiencies (r=0.177) and, notably, with LESS staffing (r=-0.209 with total HPRD) — larger facilities tend to be both thinner-staffed per resident and more heavily cited, so bed count is a real confound that needs controlling for.

Multicollinearity note: total/RN/aide/LPN HPRD are themselves highly collinear (VIF in the billions when combined, because total HPRD is arithmetically the sum of the other three) — they are not independent evidence of separate effects, so the regression below uses only total HPRD, not all four hour measures together.

Regression, controlling for facility size and turnover (OLS, n=13,394): deficiencies = 5.21 - 0.694×(total nurse HPRD) + 0.023×(beds) + 0.090×(turnover %). The staffing coefficient (-0.69, p<0.001) says that, holding facility size and nursing staff turnover fixed, an additional hour of total nurse staffing per resident per day is associated with about 0.7 fewer cited deficiencies. Turnover and facility size are each independently associated with MORE deficiencies, in the same direction the literature reports. The model's R² is 0.071 — staffing, size and turnover together account for only about 7% of the variation in deficiency counts, confirming that inspection outcomes are driven by many factors (survey-team variation, state agency stringency, case mix, care processes) beyond what this file captures.

Dose-response shape: binning facilities into HPRD quintiles shows a largely declining pattern (11.0 → 9.7 → 8.7 → 9.6 → 7.2 average deficiencies from lowest to highest staffing quintile), but it is not strictly monotonic — the fourth quintile ticks back up, and the overall trend-across-bins test is at the edge of significance (p=0.053). The CMS staffing star rating (a composite CMS itself builds from case-mix-adjusted RN and total staffing) gives a cleaner picture and is fully monotonic: 11.93 (1-star) → 10.38 → 9.43 → 7.86 → 5.31 (5-star) average deficiencies — a 55% decline from the lowest to highest staffing tier.

Robustness check: refitting the regression while dropping each state one at a time (all 50 states + DC, PR, GU; 53 refits) never flips the sign of the staffing coefficient and never crosses statistical significance — it stays negative and p<0.001 in every refit, ranging from -0.58 (dropping Illinois) to -1.28 (dropping California). California is the single most influential state (its omission steepens rather than weakens the relationship), meaning the finding is not an artifact of any one state driving it.

How this compares to published research

This corpus's cross-sectional finding lines up with the peer-reviewed literature on the same relationship. A national CMS/LTCFocus study of FY2019 data (13,086 facilities) found that a 1-unit increase in total nursing hours per resident day was associated with roughly a 17% decrease in deficiency citations, and that higher RN, licensed-nurse, and CNA HPRD each independently correlated with lower deficiency scores (Nurse Staffing Levels and Other Factors Associated with Nursing Home Quality, NCBI). Earlier work using a large national panel dataset similarly found RN and CNA staffing hours associated with fewer total and quality-of-care deficiencies (The Related Factors on Health Deficiencies in National Nursing Homes Using Panel Large Dataset, NCBI), and a classic Journals of Gerontology study specifically examined the staffing-deficiency link (Nursing Home Staffing and Its Relationship to Deficiencies, Oxford Academic). KFF's overview of nursing home deficiencies situates staffing as one of several structural quality drivers alongside ownership type and survey-team variation (A Closer Look at Deficiencies in Nursing Homes, KFF). The direction, rough magnitude, and 'real but modest' character of the relationship all match what this corpus's snapshot shows.

Caveats

• Cross-sectional, not causal. This is a single-vintage snapshot; it cannot establish that raising staffing at a given facility would reduce its future deficiency count, only that leaner and better-staffed facilities differ systematically today. A longitudinal panel (staffing change → subsequent survey outcome) would be needed for a causal claim, and that panel is not available in this corpus's current load.
• Modest explanatory power. R²=0.07 means staffing, size, and turnover jointly leave about 93% of the variation in deficiency counts unexplained — survey-team and state-agency variation, resident case mix, and care-process quality all plausibly matter more per-facility than staffing hours alone.
• Deficiency counts, not severity. total_health_deficiencies counts citations on the most recent standard survey; it does not weight by severity. The weighted health_deficiency_score correlates almost identically with staffing (r=-0.146), so this does not change the conclusion, but a severity-weighted deep dive was not run here.
• 877 of 14,271 facility-rows were dropped from the correlation/OLS sample for missing staffing, deficiency, bed, or turnover values (about 6%) — not assumed to be a random subset, but small enough not to materially bias the headline direction.

Every query behind this report

6 warehouse calls ran in this session, in order. Each is reproducible against the same snapshot.

SELECT CAST(total_nurse_hours_per_resident_day AS DOUBLE) AS total_nurse_hprd,
CAST(rn_hours_per_resident_day AS DOUBLE) AS rn_hprd,
CAST(nurse_aide_hours_per_resident_day AS DOUBLE) AS aide_hprd,
CAST(lpn_hours_per_resident_day AS DOUBLE) AS lpn_hprd,
CAST(total_health_deficiencies AS DOUBLE) AS deficiencies,
CAST(health_deficiency_score AS DOUBLE) AS deficiency_score,
CAST(certified_beds AS DOUBLE) AS beds,
CAST(nursing_staff_turnover_pct AS DOUBLE) AS turnover
FROM health.cms_nursing_home
WHERE total_nurse_hours_per_resident_day IS NOT NULL AND total_health_deficiencies IS NOT NULL

SELECT COUNT(*) AS n,
AVG(CAST(total_nurse_hours_per_resident_day AS DOUBLE)) AS avg_hprd,
AVG(CAST(total_health_deficiencies AS DOUBLE)) AS avg_deficiencies,
AVG(CAST(certified_beds AS DOUBLE)) AS avg_beds
FROM health.cms_nursing_home
WHERE total_nurse_hours_per_resident_day IS NOT NULL AND total_health_deficiencies IS NOT NULL

SELECT CAST(total_health_deficiencies AS DOUBLE) AS deficiencies,
CAST(total_nurse_hours_per_resident_day AS DOUBLE) AS total_nurse_hprd,
CAST(certified_beds AS DOUBLE) AS beds,
CAST(nursing_staff_turnover_pct AS DOUBLE) AS turnover
FROM health.cms_nursing_home
WHERE total_nurse_hours_per_resident_day IS NOT NULL AND total_health_deficiencies IS NOT NULL
  AND certified_beds IS NOT NULL AND nursing_staff_turnover_pct IS NOT NULL

SELECT CAST(total_health_deficiencies AS DOUBLE) AS deficiencies,
CAST(total_nurse_hours_per_resident_day AS DOUBLE) AS total_nurse_hprd
FROM health.cms_nursing_home
WHERE total_nurse_hours_per_resident_day IS NOT NULL AND total_health_deficiencies IS NOT NULL

SELECT staffing_rating, AVG(CAST(total_health_deficiencies AS DOUBLE)) AS avg_deficiencies, COUNT(*) AS n
FROM health.cms_nursing_home
WHERE staffing_rating IS NOT NULL AND total_health_deficiencies IS NOT NULL
GROUP BY staffing_rating
ORDER BY staffing_rating

SELECT CAST(total_health_deficiencies AS DOUBLE) AS deficiencies,
CAST(total_nurse_hours_per_resident_day AS DOUBLE) AS total_nurse_hprd,
CAST(certified_beds AS DOUBLE) AS beds,
CAST(nursing_staff_turnover_pct AS DOUBLE) AS turnover,
state
FROM health.cms_nursing_home
WHERE total_nurse_hours_per_resident_day IS NOT NULL AND total_health_deficiencies IS NOT NULL
  AND certified_beds IS NOT NULL AND nursing_staff_turnover_pct IS NOT NULL AND state IS NOT NULL