A safety net that keeps its promise

By Rebecca Cohen

Ford School of Public Policy

The law also shifts a greater share of SNAP’s administrative costs to states and expands work requirements, adding complexity to a program that many state agencies are already struggling to administer. Staffing shortages, aging technology and limited data capacity are common challenges. Now, states must find ways to reduce errors while implementing the new requirements, all without making it harder for eligible families to receive the assistance they need.

“This change happened so fast that states will have to make radical changes or face steep financial consequences,” says Pamela Herd, the Carol Kakalec Kohn Professor of Social Policy at the University of Michigan Ford School of Public Policy. “Unfortunately, it creates a powerful incentive for states to introduce new paperwork, documentation requirements, and reviews that make it more difficult for eligible families to access services.”

That tension is at the center of new work led by the Better Government Lab (BGL), a joint research center of the Ford School and Georgetown University’s McCourt School of Public Policy that Herd co-founded with Donald Moynihan, J. Ira and Nicki Harris Family Professor of Public Policy at the Ford School, and Sebastian Jilke, a public policy professor at Georgetown. Herd and Eric Giannella, an associate research professor at Georgetown and a BGL data scientist, are leading an effort to help states reduce errors without creating additional administrative burdens for eligible families.

For Herd, the work begins with the purpose of the public policy itself.

“Policies are promises,” she said. “The promise that SNAP makes is to ensure children and families have enough to eat. We are the wealthiest country in the world. It’s pretty basic.”

SNAP is one of the country’s largest safety-net programs. About one in eight Americans and one in four children receive benefits, and research estimates that as many as half of U.S. children will participate at some point during childhood.

For BGL, the central question is how states can improve program integrity without undermining SNAP’s purpose.

Finding — and fixing — the sources of error

Payment error rates are often mistaken for measures of fraud, but errors can include benefits issued at amounts that are too high or too low and can result from inaccurate or incomplete information as well as processing mistakes.

“What we’ve found is that fraud is very rare,” Herd said. “Roughly half of the errors originate from clients and half from agency staff.”

Herd explains that SNAP eligibility and benefit levels depend on details that can change frequently, including earnings, household composition, and child care and medical expenses. For example, someone may not realize that they have to report a change in household composition. A caseworker can also make an ordinary processing mistake with significant consequences, such as failing to convert weekly or seasonal income into a monthly amount.

Because state error rates are calculated using a relatively small sample of cases and weighted by the dollar value of mistakes, a handful of large errors can have an outsized effect. Understanding what drives a state’s error rate therefore requires a close look at recurring patterns.

Using historical SNAP Quality Control data, BGL researchers have released — and continue to improve — several types of resources to help states identify and address systematic errors, including data visualization tools, predictive modeling code, and guidance for data analysis. Rather than requiring additional review for every application, these resources help agencies focus limited staff time on the cases, policies and processes that contribute most to the error rate.

Giannella develops open-source modeling code that combines patterns identified in national public data with those found in internal state data. He also shares the code, outputs and modeling guidance, refining them through the Safety Net Response Network, a nationwide learning community of nonprofits that support state governments, co-led by BGL.

“Our role is to help states jump-start their modeling work so they begin to see the returns on better use of data in terms of lower error rates,” Giannella says.

Once they understand the trends, states can better target their responses through secondary reviews, additional staff training or improved communication and system prompts.

“Analyzing errors is very in-the-weeds, but it’s critical to coming up with a strategy for how states should respond,” he said. “My hope is that state agencies can do this work in-house in the long run; they will have the best sense of exactly what the data mean and what they could do with a particular answer. Agencies also recognize that building that capacity will enable visibility into other opportunities to improve program delivery.”

When process creates burden

To combat high error rates, some states have imposed additional eligibility verification requirements on participants.

Arizona offers a cautionary example of the administrative burden a broad verification strategy can create. In an effort to verify eligibility and work activity, the state has required participants to obtain additional third-party documentation, such as paystubs or letters from employers, rather than relying on self-reported work hours.

According to Herd, that approach shifts more of the work of proving eligibility to participants while generating more paperwork for state employees to process.

“The public is overwhelmed, and the Arizona administrative system is overwhelmed too, because it cannot manage all the documents coming back,” Herd said.

A missed letter, an interview scheduled during a work shift, a confusing form or a long wait for assistance can determine whether an eligible family receives food assistance. Those hurdles may be especially consequential for people with unstable work schedules, disabilities, language barriers, unreliable internet access or housing insecurity.

A lower error rate does not necessarily mean a state’s SNAP program is working better. If new requirements lead to processing delays, procedural denials or abandoned applications, eligible people may lose access even as the state’s performance appears to improve.

As Herd puts it, “If you make it harder for people to receive benefits, fewer people will receive benefits.”

Added complexity can also create the very errors states are trying to prevent, Herd explained: “Every extra step or requirement is an additional opportunity to make a mistake.”

BGL’s approach is designed to help states avoid that cycle by using data to identify where additional review is warranted rather than asking every applicant to clear more hurdles.

Building evidence for better administration

In addition to providing technical assistance, BGL researchers are examining how administrative practices and investments — including flexible interview policies, simplified reporting requirements, staffing levels and administrative spending — are associated with both accuracy and access.

They are particularly interested in states that combine relatively low administrative burdens with low error rates: How have those states balanced the two goals, and which of their practices could work elsewhere?

The project could also produce evidence about the relationship between state capacity and program outcomes. Because the law reduces the federal share of SNAP administrative support, BGL researchers will assess whether changes in administrative spending are associated with higher error rates, slower processing or reduced access — findings that could inform future legislative decisions.

The findings may also inform implementation beyond SNAP as states prepare for what Herd calls “a second wave” of administrative changes stemming from the law’s new Medicaid work-reporting and eligibility-verification requirements.

The lab’s long-term goal is a safety net designed around simplicity, accessibility, respect and accuracy — one that is easier for families to navigate and for states to administer efficiently. If the project succeeds, it could demonstrate that program accuracy and access are not competing objectives, but two essential measures of whether SNAP is keeping its promise.