Medicaid as Low Tier Medicine
Nursing home care in the United States is built on a two‑tier system, and Medicaid defines the lower tier. Because Medicaid is a welfare program, older adults must first prove poverty — often by spending down a lifetime of savings — before they can receive long‑term care coverage.
Once enrolled in, LTC patients enter a parallel medical system in which expectations, staffing levels, and outcomes are markedly lower. This is not an accident of policy design; it is the design. Medicaid finances the majority of nursing home care, yet it does so within a framework that treats poor elders as a separate clinical class. What our analyses show, however, is that this lower tier does not translate into lower financial returns for providers.
History Matters: Industry Growth & Market Concentration
In 1974 — a mere decade after Medicaid was legislated into existence — a U.S. Senate Subcommittee on Long‑Term Care reported that the nursing home industry had grown to 23,000 facilities supplying 1.23 million beds. By 2024, the number of beds had remained relatively unchanged but concentrated within ~15,000 facilities. Sixty percent of bed‑days are now allocated to Medicaid patients. The industry today pulls in roughly $170 billion in Medicaid, Medicare, VA, and out‑of‑pocket revenue — a dramatic increase from the $7 billion reported in 1974.
What this history shows is simple: the U.S. nursing home sector delivers roughly the same number of beds as in 1974, but through one‑third fewer facilities and at more than twenty times the nominal cost. Adjusted for inflation, the system still costs 3–4× more per bed than it did in the 1970s. The U.S. elderly population in 1973 was 16 million. It has grown to over 60 million. That is the signature of industry concentration, reduced supply, and higher per‑capita extraction — not increased need or improved care.
To understand how poverty medicine became the operating logic of U.S. long‑term care, we have to look at the industry’s historical structure. Medicaid did not simply finance nursing home care; it reshaped the entire sector around a means‑tested, lower‑tier model. Over time, the industry consolidated, capacity stagnated, and the business model evolved to extract more revenue per bed even as the clinical tier serving Medicaid patients remained chronically understaffed and under‑resourced. This historical trajectory is essential for interpreting the empirical findings that follow.
The Industry’s Medicaid Myth as Political Narrative
The industry’s central claim is that high Medicaid census forces low staffing and low quality because Medicaid “doesn’t pay the cost of care.” If that were true, facilities with more Medicaid patients should have lower margins. But the data show the opposite pattern.
As the proportion of Medicaid bed‑days declines, staffing ratings rise, RN hours increase, turnover falls, and facilities become smaller and more clinically focused. The relationship is structural, consistent, and monotonic. Medicaid share is not depressing staffing; staffing is stratified by the tiered system Medicaid created.
Visual Evidence: The Staffing Gradient
Table 1 includes selected variables cross-tabulated with staffing ratings. Based on our many years of research, we believe that staffing rating is the primary indicator of a facility’s quality of care. Facilities with lower staffing ratings have substantially higher Medicaid share, lower RN hours, higher turnover, and larger bed counts. As staffing ratings improve, Medicaid share declines in a monotonic pattern, while RN hours and clinical focus increase. This pattern contradicts the industry claim that high Medicaid census forces low staffing.
Table 1. Staffing Rating and Key Facility Characteristics
The graphic below demonstrates how Medicaid share declines steadily as staffing ratings rise. As our regression models below will show that facilities’ bottom lines (Net Operating Income) is highest at the lowest staffing rating. Medicare share increases modestly as staffing ratings rise. This inverse relationship demonstrates that higher‑staffed facilities serve fewer Medicaid patients — a structural pattern inconsistent with the claim that Medicaid reimbursement levels determine staffing quality.
Figure 1. Percent Medicaid by Staffing Rating
Testing Hypotheses: Modeling the Data with Linear Regression Models
The staffing gradient makes the industry’s claim testable. If Medicaid underfunding truly drives poor staffing and poor quality, then Medicaid share should be a strong predictor of financial distress. To evaluate this claim, we estimated a series of linear regression models. We regressed the following predictor variables on “net operating income”
- Proportion of Bed Days – Medicaid
- Proportion of Bed Days – Medicare
- Number of Certified Beds
- Per Patient Expenditures
- Related Party/Home Office Allocations Per Bed Day
- RN HPRD
- Staffing Rating – a categorical variable with 5 levels (with the exception of one model explained in the APPENDIX A, a staffing rating of 3 was designated as the reference variable)
Although the CMS 2024 Cost Report data (the latest available) merged with the 4th quarter 2024 ProviderInfo file includes ~1500 facilities, many facilities had missing data for one or more of the variables included in our model. Therefore, not all facilities were included in the models.
Because nursing home data is nested within states, and chains, we tested five models:
- a national model – without consideration for the regression assumption of normal distribution for variables.
- a trimmed model – elimination of outliers, which resulted in variables meeting the assumption of normalcy.
- state‑specific models (Kansas and New York) – a state with an average bed size of 60 beds and a state with an average bed size of 110 beds.
- a chain‑specific model (The Ensign Group)
Across all models — regardless of geography, ownership, size, or case mix — the coefficient for Proportion of Medicaid Beds is nonsignificant. The absence of a Medicaid effect is not a quirk of one dataset; it is a structural null. This is the empirical core of The Medicaid Myth.
Tables with coefficients (unstandardized B), Coefficients Standard Error, Standardized Coefficients, t-values, and significance level are included in APPENDIX A.
The informative patterns across the models include:
- The proportion of Medicaid bed days has no effect on NOI
- The proportion of Medicare bed days has a modest positive effect on NOI
- With the exception of the Ensign Group model and the Kansas system, the number of certified beds (bed size) has a significant negative effect on NOI. As will be discussed later, this finding is due to the sophistication of Ensign management in maintaining an optimal payor mix and the typical small bed size characteristic of the Kansas nursing home system.
- Unsurprisingly, costs such as per patient expenditures, related party/HO expenditures have a significant effect on NOI. However, the point is that holding these variables constant, Medicaid has no effect on NOI.
- RN HPRD has little to no effect on NOI.
- Much as the data demonstrates in Table 1 and is illustrated in Figure 1, the Rating categorical variable indicates that NOI is highest in the bottom two rating categories and lowest in the top two rating categories.
Editor’s Note
Publicly funded long‑term care should reflect public standards. Instead, the for‑profit nursing home sector operates like a protected cartel, controlling bed supply and payer mix according to financial priorities rather than community need. Regulators publish ratings, but owners determine staffing and access based on cash flow, not quality. And the industry’s most reliable revenue stream is the one it claims is inadequate: Medicaid — the predictable floor that stabilizes capital flows and protects returns.
Understanding Medicaid’s role in this financial architecture is essential for journalists, advocates, and legislators. The industry’s political narrative obscures how the system actually works and denies policymakers the granular data needed to fund and regulate long‑term care responsibly. The Center for Health Information & Policy provides that clarity. Our bulletins counter misinformation with transparent analytics and support efforts to build a fairer, more accountable system of care.
APPENDIX A: TECHNICAL STATISTICAL REPORT
Linear Regression Models
Utilizing the CMS 2024 ProviderInfo and Cost Report files (merged) we applied factor analysis as a data reduction technique. We identified three dimensions for explaining the financial structure of the nursing home industry: professionalism (RN staffing), costs (expenses), and scale of operations, (bed size). Cases with missing values for any variable entered into models were deleted listwise.
Through an iterative modeling process, we concluded that the financial impact of Medicaid on net operating income controlling for Medicare, per patient expenditure, related party expenditure, bed size, RN hours per resident day, and staffing rating is insignificant.
Overview of Model Specifications
This appendix presents the full regression results underlying the analyses in The Medicaid Myth, Part I. All models use facility‑level data and the same core specification:
- Dependent variable: Net Operating Income (NOI)
- Predictors:
- Proportion Medicaid
- Proportion Medicare
- Number of Certified Beds
- Per‑Patient Expenditures
- Related‑Party Expenditures per Bed‑Day
- RN Hours per Resident‑Day
- Staffing Rating (dummy variables)
Each model tests whether Medicaid share predicts financial performance under different structural conditions.
Full National Model
Interpretation summary:
- Medicare has a sizeable positive effect – as the proportion of Medicaid beds increase, NOI increases.
- The coefficient for Proportion Medicaid is significant, indicating a positive relationship between Medicaid share and NOI in the full national dataset. However, the effect is small and probably a measure of noise due to the size of the dataset (~10,000).
- Bed Size has a negative effect has a negative effect on NOI.
- Costs: Per Patient Expenditures & Related Parties/HO Expenditures have a significant negative effect on NOI.
- The lowest staffing ratings have a significant positive effect on NOI. Conversely highest staffing ratings have a significant negative effect on NOI.
- R Square = .91
National Model with extreme outliers trimmed:
After removing extreme outliers and leverage points, Proportion Medicaid remains nonsignificant, while Medicare share, bed size, and related‑party expenditures show strong associations with NOI.
Interpretation summary:
- Medicare has a sizeable positive effect – as the proportion of Medicaid beds increase, NOI increases.
- The coefficient for Proportion Medicaid is negative but has absolutely no effect: Unstandardized coefficient = -823, standardized coefficient = -.001, p value = .982.
- Medicare has a significant positive relationship with NOI: p value < .001.
- Bed Size has no effect on NOI: B = -127, p value = .502
- Costs: Per Patient Expenditures have a small negative effect on NOI: p value = .054, while Related Parties/HO Expenditures have a nonsignificant effect on NOI: p value = .096.
- RN HPRD has no effect on NOI: p value = .830
- The lowest staffing ratings have a significant positive effect on NOI. Conversely highest staffing ratings have a significant negative effect on NOI.
- R Square = .16
After removing extreme outliers and leverage points, Proportion Medicaid remains nonsignificant, while Medicare share shows a strong relationship with NOI.
Kansas Model (Small‑Bed State)
Interpretation summary:
In a state dominated by small facilities and rural case mix, Medicaid share does not predict NOI. This contradicts the claim that Medicaid underfunding drives financial distress in rural markets.
- The coefficient for Proportion Medicaid has no effect on NOI: Unstandardized coefficient = -613107, standardized coefficient = .034, p value = .652
- Medicare has a significant positive relationship with NOI: p value < .0
- Bed Size has a strong positive effect on NOI: B = 22080, p value = < .001.
- Costs: Per Patient Expenditures have a robust negative effect on NOI: p value = < .001. while Related Parties/HO Expenditures have a significant effect on NOI: p value = .009.
- RN HPRD has an insignificant effect on NOI: p value = .165
- The staffing ratings have no effect on NOI.
- R Square = .761
New York Model (Large‑Bed State)
Interpretation summary:
Despite large facilities, high Medicaid penetration, and a complex reimbursement system, Medicaid share is still nonsignificant. The structural null holds even in the most Medicaid‑heavy state.
- The coefficient for Proportion Medicaid has no effect on NOI: Unstandardized coefficient = -613107, standardized coefficient = -.023, p value = .669
- Medicare has a significant positive relationship with NOI: p value < .001.
- Bed Size has a strong positive effect on NOI: B = 22080, p value = < .001.
- Costs: Per Patient Expenditures have a robust negative effect on NOI: p value = < .001. while Related Parties/HO Expenditures have a significant effect on NOI: p value = .009.
- RN HPRD has a modest effect on NOI: p value = .018
- The staffing ratings of 4 and 5 have a strong negative effect on NOI.
- R Square = .720
Ensign Group Model (Chain‑Specific)
Interpretation summary:
Within a sophisticated, publicly traded chain with advanced financial engineering, Medicaid share again fails to predict NOI. This demonstrates that the null effect persists even in highly optimized corporate environments.
- The coefficient for Proportion Medicaid has no effect on NOI: Unstandardized coefficient = -191515, standardized coefficient = -.023, p value = .768
- Medicare has a significant positive relationship with NOI: p value < .006.
- Bed Size has a strong positive effect on NOI: B = 15567, p value = < .001.
- Costs: Per Patient Expenditures have an insignificant effect on NOI: p value = < .930. while Related Parties/HO Expenditures have a significant effect on NOI: p value = .013.
- RN HPRD has an insignificant effect on NOI: p value = .213
- A staffing rating of 1 has a negative effect on NOI: p value = .004
- R Square = .56
Notes on Interpretation
A short, reader‑friendly note:
- “ns” = nonsignificant (p > .05)
- Coefficients represent the expected change in NOI for a one‑unit change in the predictor
- Dummy variables compare each staffing rating to the omitted reference category
- All models use the same specification for comparability