How Researchers Know Medicaid Freezes Harm Beneficiaries

The basic question is not whether a Medicaid freeze sounds harsh. It is how researchers can tell, before and after, that a funding squeeze actually changes beneficiary outcomes. The answer depends on what the freeze means: a federal payment cap, a state provider-payment hold, an eligibility redetermination freeze, or a broader funding cap. Those mechanisms do not produce the same evidence trail, so a serious case study has to keep projections, counterfactuals, and observed disenrollment separate.

Federal payment freeze to states
State provider-payment hold
Eligibility redetermination freeze
Funding cap or block grant

Abstract illustration of three evidence streams converging on one point.

Three Ways To Test The Claim

The cleanest way to see the evidence problem is to put the methods side by side. One projects forward from assumptions, one reconstructs a counterfactual from administrative records, and one tracks what happens when coverage access tightens in the real world.

Approach	Data source	What it tells you	Main limit
Microsimulation modeling [1]	Policy and health-system inputs in Basu et al.	Projects uninsured counts, mortality, hospital use, hospital closures, and broader economic effects under proposed Medicaid constraints	Depends on assumptions about behavior and policy response
Retrospective counterfactual claims analysis [2]	MACPAC T-MSIS data from 41 states, 2018-2022	Tests whether per-capita caps would have bound against actual state spending	Hypothetical policy, not observed harm from an enacted freeze
Disenrollment tracking [5]	Medicaid unwinding and coverage-loss tracking	Shows what coverage contraction looks like when access is restricted	Not a freeze study in the strict sense; the mechanism is different

The Microsimulation Does The Heavy Lifting

Basu et al. provide the strongest forward-looking estimate. In JAMA Health Forum, the microsimulation projected 7.6 million to 11.7 million newly uninsured people, 1,484 to 2,284 excess deaths per year, 94,802 to 145,946 preventable hospitalizations per year, 101 rural hospitals at high risk of closure, and a $135.3 billion GDP reduction [1].

That range is the point, not a flaw to hide. The model is not reporting observed harm; it is estimating what happens if the proposed Medicaid constraints change coverage, utilization, and health-system finances the way the authors assume they will. If state behavioral responses shift, the output shifts too. The wider ranges are the visible cost of asking the model to do real counterfactual work.

The most useful reading is not to ask whether the exact number will come true. It is to ask whether the direction is stable, whether the mechanism is explicit, and whether the model makes its own uncertainty easy to inspect. On those terms, the study is valuable because it turns a policy change into a chain of downstream estimates instead of pretending the whole effect can be seen in one headline number.

The Counterfactual Check

The CBPP analysis matters because it asks a skeptic's question: if a per-capita cap had already been in place, would actual state spending have hit it? Using MACPAC T-MSIS data from 41 states and spending data from 2018 through 2022, it reconstructed that counterfactual and found that nearly every state would have exceeded the cap [2].

Ohio is the sharpest example in the analysis, with projected losses of $190 million in 2019 and nearly $2 billion by 2022 [2]. That does not mean the policy was observed in practice. It means administrative claims data can be used to test whether a cap would have been binding under real spending conditions, which is a different and more credibility-building question than a pure projection.

The Real-World Signal

The disenrollment literature is not a freeze study in the strict sense, but it is the closest behavioral signal in the set. When pandemic continuous coverage ended, Medicaid coverage loss accelerated, and about one-third of redetermined enrollees lost coverage [5].

That evidence does not prove the same mechanism as a funding freeze. It does something narrower and still useful: it shows what coverage contraction looks like when access tightens in practice. In other words, it helps readers see the downstream behavior that the projection is trying to anticipate.

What Convergence Changes

The broader state-level projections line up with that direction of harm. The Commonwealth Fund projected $665 billion in state Medicaid budget reductions over 2025-2034 and 1.2 million job losses [3]. RAND's state-level review of key Medicaid provisions also points in the same direction [4]. The methods are not identical, and they do not deserve to be treated as if they were. But they are independent enough that convergence matters.

This is the part worth keeping straight for GRE analytical writing and MCAT social and behavioral sciences work: separate the data source, the identification strategy, and the claim. The same habit shows up in GRE analytical writing and MCAT social and behavioral sciences: when one method's limits are visible and another method covers those limits, confidence rises.

References

Projected Health System and Economic Impacts of 2025 Medicaid Policy Proposals — JAMA Health Forum, July 2025
Medicaid Per Capita Cap Would Harm Millions — Center on Budget and Policy Priorities, January 2025
How Medicaid, SNAP Cutbacks Would Trigger Job Losses Across States — The Commonwealth Fund, June 2025
State-Level Impacts of Key Medicaid Provisions in the One Big Beautiful Bill Act — RAND
Medicaid: What to Watch in 2026 — KFF, 2026