The federal government is remarkably good at keeping millions of people above the poverty line, and considerably less tidy when asked to account for every dollar it spends doing it.
Those two facts are not contradictory.
New data released by the U.S. Census Bureau show that Social Security alone kept an estimated 28.8 million people from falling below the Supplemental Poverty Measure in 2025. Refundable tax credits kept another 6.1 million above the line. The Supplemental Nutrition Assistance Program, better known as SNAP, accounted for 3.1 million. Supplemental Security Income accounted for 2.2 million, while housing subsidies accounted for another 2.1 million.
That is the good side of the ledger.
On the other side sits a federal payment system that reported an estimated $186 billion in improper payments during fiscal year 2025, including tens of billions associated with some of the same enormous programs intended to improve Americans’ economic security.
It would be tempting to choose one number and tell a simple story.
The numbers refuse to cooperate.
Two Ways to Measure Poverty
The Census Bureau reported Monday that the official U.S. poverty rate fell from 10.7% in 2024 to 10.2% in 2025, leaving about 34.5 million people below the official poverty threshold.
But Census maintains another measurement called the Supplemental Poverty Measure, or SPM.
Unlike the traditional measure, the SPM attempts to account for considerably more of the economic machinery surrounding a household. Government benefits such as SNAP, housing subsidies and refundable tax credits count as resources. Taxes, work expenses and medical expenses reduce those resources.
Under that broader calculation, the poverty rate was 13.1% in 2025 and did not change significantly from 2024.
The distinction allows Census to perform an unusually revealing exercise: remove individual benefits or expenses from the calculation and see how many people cross the poverty line.
It amounts to something resembling an antipoverty ledger.
And Social Security owns the largest entry.
Without Social Security benefits, Census estimates another 28.8 million people would have been in SPM poverty in 2025.
Refundable tax credits reduced the number by 6.1 million. Within that category, the Earned Income Tax Credit accounted for about 4 million. SNAP reduced it by 3.1 million, SSI by 2.2 million and housing subsidies by 2.1 million.
Other noncash benefits—including energy assistance, WIC and school lunch—collectively kept about 1.5 million people above the threshold. Other cash benefits accounted for another 1 million.
These estimates overlap and cannot simply be added together. A household can receive more than one benefit, and Census specifically warns that the individual components are not mutually exclusive.
Still, the direction is unmistakable.
Federal assistance changes measured poverty substantially.
Then the Bills Arrive
Census’s ledger also runs in reverse.
Medical out-of-pocket expenses pushed approximately 7.7 million people into SPM poverty in 2025. Payroll taxes, or FICA, added roughly 4.5 million. Work expenses added about 4 million, and federal income taxes pushed another 1.4 million below the threshold.
The medical number is particularly striking because it increased from approximately 7.5 million in 2024.
At the same time, several government benefits became somewhat less consequential in the poverty calculation.
Refundable tax credits kept 6.8 million people out of poverty in 2024, compared with 6.1 million in 2025. SNAP’s effect declined from 3.6 million to 3.1 million. SSI declined from 2.5 million to 2.2 million.
Social Security, meanwhile, barely moved: 28.7 million in 2024 and 28.8 million in 2025. Housing subsidies remained at approximately 2.1 million.
So while the topline SPM poverty rate barely changed, some of the machinery underneath it did.
That is one advantage of looking below the headline number.
Another is discovering what happens to all the money.
$186 Billion in Improper Payments
Federal agencies reported approximately $186 billion in improper payments during fiscal year 2025, according to the Government Accountability Office.
About $153 billion—roughly 82%—were overpayments.
The total was $24 billion higher than the previous fiscal year. GAO says the $186 billion does not represent the entire universe of improper federal payments because some programs susceptible to significant errors were not included in the total.
Several programs connected to the poverty ledger appear prominently in those numbers.
Medicaid reported an estimated $37.39 billion in improper payments, representing a 6.12% error rate.
The Earned Income Tax Credit reported approximately $21.1 billion in improper payments and an estimated error rate of 32.7%.
SNAP’s fiscal 2025 payment-error rate was 10.62%, representing approximately $10.1 billion in improper payments.
SSI’s latest available payment-integrity estimate, covering fiscal 2024, found approximately $7.33 billion in improper payments from $63.29 billion in outlays.
Those are enormous numbers.
They are not, however, enormous fraud numbers.
Improper Does Not Mean Stolen
An improper payment can involve fraud. It can also involve an innocent eligibility mistake, the wrong payment amount, missing paperwork, an underpayment or even a payment made to the correct person for the correct amount through a process that failed to satisfy federal requirements.
Medicaid provides an almost absurdly useful example.
Of the program’s $37.39 billion in estimated fiscal 2025 improper payments, CMS says 77.17% resulted from insufficient documentation, which the agency says is generally not indicative of fraud or abuse.
That does not make the problem imaginary.
It means the government could not demonstrate that those payments satisfied all applicable requirements.
The distinction becomes even stranger in the Affordable Care Act premium-tax-credit program. CMS reported that approximately 43% of that program’s improper payments involved situations in which the government paid the right recipient the right amount—but violated a statutory or regulatory requirement while doing it.
Financially correct.
Procedurally improper.
There may be no phrase more suited to the federal government.
Actual fraud nevertheless exists, and treating every payment error as innocent would be just as misleading as calling every error theft.
The Holes in the Fence
SNAP provides a good illustration of the difference between a known improper payment and a potential control weakness.
USDA reported $10.1 billion in SNAP payment errors for fiscal 2025. The agency says its payment-error rate measures whether states correctly determine household eligibility and benefit amounts and includes both overpayments and underpayments.
A separate USDA program-integrity review found something different.
Reviewing eligibility information from 29 state agencies, USDA identified records involving dummy or missing Social Security numbers, duplicate participation within states, interstate duplication and deceased individuals appearing as active participants.
The agency estimated that dummy Social Security numbers and intrastate duplication alone represented more than $1.5 billion in annual financial exposure.
“Exposure” is not the same as confirmed loss, and neither is synonymous with fraud.
But it identifies holes in the fence.
The distinction matters because a prosecution can tell taxpayers how much one fraudster stole. A control weakness raises the more consequential question of how many fraudsters the government might not have identified.
And What About Noncitizens?
Immigration status creates another place where the numbers require more care than the political shorthand normally permits.
Some noncitizens are legally eligible for particular federal benefits. Others are not. Mixed-status households may contain an ineligible adult and eligible U.S.-citizen children.
A benefit received by such a household therefore cannot automatically be described as money improperly paid to a noncitizen.
Social Security Administration records provide an unusually clear example.
In December 2025, 7,392,201 people received federally administered SSI payments.
Of them, 7,082,134 were citizens and 310,067 were noncitizens—about 4.2% of recipients.
Noncitizen recipients received a lower average monthly payment: $596.77 compared with $719.70 for citizens. Nearly 252,000 of the noncitizen recipients were 65 or older.
Those figures establish that noncitizens receive a measurable share of SSI benefits.
They do not establish that those payments were illegal.
That requires a separate eligibility determination.
And fraud cases involving noncitizens using false identities or otherwise defeating eligibility controls should likewise be categorized as fraud—not used to characterize every lawful noncitizen beneficiary.
The distinction may be less satisfying than a sweeping conclusion, but accounting generally is.
The Government Knows Some Things Better Than Others
The federal government can estimate with remarkable specificity how individual programs affect poverty.
Census can tell us that Social Security moved 28.8 million people across its Supplemental Poverty Measure threshold.
It can estimate that SNAP moved 3.1 million.
It can tell us that medical expenses pushed 7.7 million the other direction.
Elsewhere in Washington, however, the accounting becomes considerably fuzzier.
GAO says federal agencies identified $186 billion in improper payments during fiscal 2025—and also says that total does not capture the full government-wide universe.
Treasury estimates that nearly one-third of EITC payments were improper. USDA reports a SNAP error rate above 10%. CMS reports tens of billions in Medicaid errors while explaining that most resulted from inadequate documentation rather than apparent fraud.
And the Social Security Administration can tell us exactly how many SSI recipients are noncitizens, while that fact alone tells us almost nothing about whether their benefits were properly or improperly paid.
The deeper one digs into the ledger, the harder it becomes to reduce it to “government programs work” or “government programs waste money.”
Both propositions are too simple.
What Taxpayers Are Buying
Census’s new statistics establish something important.
The American safety net has a measurable effect.
Remove Social Security, SNAP, refundable tax credits, SSI, housing subsidies and other assistance from household resources and millions more people fall below the Supplemental Poverty Measure.
That is not an ideological claim. It is what the government’s own statistical model measures.
The government’s other records establish something equally important.
Programs distributing hundreds of billions of dollars make mistakes. Agencies send money in incorrect amounts, fail to maintain required documentation, pay people who are not eligible, underpay people who are eligible and sometimes get defrauded.
The scale is large enough that even small error rates produce numbers measured in billions.
In fiscal 2025, the government estimated $186 billion in improper payments across 64 programs. GAO says federal improper-payment estimates since fiscal 2003 total roughly $3 trillion.
Those numbers do not erase the 28.8 million people Social Security kept above the SPM threshold.
And those 28.8 million people do not erase the payment errors.
The more useful accountability question is what lies between them:
How much poverty is the federal government preventing, how much does it spend doing so, how much money leaves the system improperly, how much of that represents actual fraud—and how confidently can Washington tell taxpayers the difference?
Census has provided a surprisingly good answer to the first question.
The federal government’s own auditors suggest there is considerably more work to do on the rest.
Bureaucracy Times 