USDA published payment accuracy figures for every state this week. The national error rate is 10.62 percent. Under the 2025 reconciliation law, those rates now decide how much each state must pay toward its own SNAP benefits beginning in 2028. The error rate is not fraud. It mostly reflects how well a state runs the program and how broadly it serves the people who qualify. This is what the new numbers show, and what cutting them would cost.
Payment error rates get talked about as if they were a fraud statistic. They are not. Each year USDA and the states pull a statistical sample of SNAP cases and re-determine every one of them in depth, then measure the share of benefit dollars that were issued too high or too low against the correct figure. That share is the payment error rate. For fiscal year 2025, released June 24, 2026, it was 10.62 percent nationally.
It is worth being precise about what that does and does not mean. A 10.62 percent rate does not mean one in ten cases is wrong, or that one in ten recipients got the wrong benefit. It means that across the benefit dollars reviewers checked, overpayments and underpayments together came to 10.62 percent of the total. The average household benefit is about $352 a month. A 10.62 percent error rate is the equivalent of that benefit being off by roughly $37 a month in either direction, about a dollar a day. Errors cluster in a minority of cases rather than spreading evenly, so no single family is necessarily off by that much, but the scale is the point: this is a measure of administrative precision on modest sums, not large-scale waste.
What the rate actually tracks is administrative bandwidth, whether a state can keep up with the constant churn of reported and unreported changes in income and circumstance and still get the math right, compounded by access, how many eligible people it is serving in the first place. By USDA's own accounting most error is the agency's doing, not the recipient's, and roughly one in eight error dollars is an underpayment, money owed to families that never reached them. Fraud and trafficking are measured separately and run near one percent. Starting in 2028, this accuracy score becomes a bill.
The 2025 reconciliation law maps a state's error rate to a cost-share percentage and applies it to the state's annual SNAP benefit spending. Below 6 percent, a state owes nothing. At 10 percent or above, it covers 15 percent of its benefit costs. These figures use each state's ongoing FY2023 benefit issuance, with the expired COVID emergency allotments removed, and reflect the law as enacted.
The penalties do not begin until FY2028, and the law is full of off-ramps, one of them upside down. A state already below 6 percent owes nothing in the first year, and any state can still escape by dropping under the line before its determination year. But the states with the worst error rates, at or above roughly 13.3 percent, get the longest runway of all: they can defer the penalty to FY2029 or even 2030. A higher error rate buys more time, not less. That paradox was not designed as policy. It is a byproduct of the reconciliation negotiation, the price of holdout votes written into the statute, and the seven jurisdictions it shields include the four worst-performing states in the country. So this is not a fixed bill. It is a target with a crooked incentive already built into it, and the rest of this piece asks what driving the number down actually requires.
| State | Error rate (FY2025) | Cost-share | Annual liability | Per participant / yr |
|---|
The penalties land on rates that climbed sharply after 2021. Total payment error rate by state, FY2017–FY2025, U.S. average in black. No official rates exist for FY2015–16 or the FY2020–21 COVID review pause; the lines are dashed across that gap to connect the two eras without inventing data for the missing years.
Why does one state's error rate run two or three times another's? Almost all of it comes down to two things: how well a state runs the program day to day, and how many of its eligible residents it actually serves. Together those two explain about 62 percent of the gap between states. The first, call it capacity, sets how much error a state makes at all. The second, access, sets what kind of error it is: whether it shows up as overpayments to families already on the rolls, or as eligible people who get wrongly turned away.
Capacity is the clearest single signal, and it shows up in timeliness. Each dot is a state, plotting how promptly it processes recertifications against its payment error rate. When a household's income changes and the paperwork sits in a queue, the old benefit keeps going out and the overpayment accrues, so the faster a state keeps up, the lower its error, and the line is close to straight. Dollars spent per case show no such relationship. Capacity does.
Recertification processing timeliness (FY2024) against payment error rate (FY2025), each dot a state. Correlation −0.54. Timeliness bites overpayments in particular: when a reported income increase sits in a queue, the old, too-high benefit keeps going out and the overpayment accrues.
Access pulls the other way, and it is worth being precise about what "access" means here, because it is easy to misread. This is not about who is eligible. Eligibility rules are federal and do not change from one of these states to the next. It is about what share of the people who already qualify a state actually reaches and keeps enrolled. A state serving a larger share of its eligible residents is carrying more active cases, and every case is a live touchpoint: an income change to process, a recertification to complete, a calculation to keep current. More eligible families being served means more of those moments, and more chances for one of them to go wrong. The higher error rate is a function of administrative surface area, not of letting the wrong people in. A more restrictive state runs a smaller, steadier caseload and posts a lower number for it. Neither is fraud.
Program Access Index, the share of a state's eligible low-income population that actually receives SNAP (2023), against payment error rate (FY2025). Correlation +0.46. The slope reflects more eligible people being served, more cases and more touchpoints, not more ineligible people on the rolls.
The two errors a state can make. Payment accuracy is one axis. The other is how well a state handles the cases it denies, terminates, or closes, measured separately as the Case and Procedural Error Rate. Crossing the two sorts every state into one of four operational profiles. Each tab shows one of them on its own, with its own trend line. Most states fall into the two profiles that line up, strong on both or weak on both, because both track the same underlying capacity.
Because the metric is blind to wrongful denials, the cheapest way to lower it is to serve fewer people. That is the incentive the penalty creates, and it is not hypothetical. Since H.R.1 passed, the states with the steepest caseload declines have been the ones rolling back self-attestation and shortening certification periods, the two quiet administrative levers that push eligible people off the rolls, with the largest drops where both are combined (companion analysis). And access is not a neutral dial. SNAP is the country's largest anti-hunger program, and the evidence that it reduces food insecurity is deep and long-settled.
Among states with similar poverty, the ones that reach more eligible people have less food insecurity. Both axes below are adjusted for each state's poverty rate, so this is not just poverty showing through. The downward tilt is the protective effect of access, and it holds even though high-need states tend to expand enrollment, a bias that works against finding it.
SNAP access (Program Access Index) and food insecurity, each residualized on the state poverty rate. Partial correlation −0.38. Observational, but pointing the same way as decades of causal research on SNAP and hunger.
This is the asymmetry the penalty ignores. A payment error is recoverable. An overpayment gets clawed back; an underpayment gets corrected and repaid. The money moves the wrong way and then it moves back. A month without enough food for a child who was eligible does not get refunded. The harm to a family turned away, or pushed off the rolls, lands and stays.
The 2025 law put these two things on the same scale. On one side a state's payment error rate, on the other its own budget, and the instruction to balance them. But the weights are not the same kind of thing. A payment error is a synthetic quantity, a figure the program defines, measures against itself, and can recover. Hunger is not synthetic. Its costs fall on children's development, on health, on whether people can work and learn and build a life, and those costs do not reverse.
The penalty treats the two as commensurable. They are not, and pretending they are is the policy.
Read these as correlations, not precise estimates. This analysis cross-references measures drawn from different years in several places: payment error rates are FY2025, processing timeliness is FY2024, access and benefit figures are FY2023, food insecurity is a 2022–2024 average, and poverty is 2023. Each is the most recent figure available, and the underlying conditions move slowly, but the year mismatches mean the relationships here are correlational and directional rather than exact or causal. They point the same way the established research does, and they could be tightened with same-year and longitudinal data. Treat them as strong indications of how the program behaves, not as final estimates.