When a Work Rule Lowers Access Without Raising Work
This series examines SNAP’s work requirement for able-bodied adults without dependents, or ABAWDs, and asks a deceptively simple question: when people lose access to food assistance after a work rule is reinstated, do they work more, or do some merely lose the benefit? The answer matters because the policy is asked to do two different jobs at once. It is supposed to encourage employment, but it is also used as a screening device—a condition intended to separate people presumed able to work from people judged more in need. To understand whether it succeeds, we need to track not only who remains on SNAP, but also employment, earnings, income, who exits, and the welfare consequences of those exits.
The central subject is therefore not “welfare” in the abstract. It is the concrete mechanism through which a SNAP eligibility condition turns work, training, exemptions, and documentation into continued access to food assistance. The practical judgment this series supports is equally concrete: a smaller caseload is not, by itself, evidence of more work or greater self-sufficiency.
That distinction is especially important because a benefit program can become smaller in two very different ways. People may leave because their earnings rose enough that they no longer need the benefit. Or they may leave because the program has become harder to enter, remain in, or prove eligibility for—even while their work and income remain unchanged. Both events reduce the number of recipients. Only the first clearly advances the employment story used to justify a work condition. Treating the two as equivalent turns an administrative outcome into an economic conclusion without measuring the missing step.
The result that forces the question
A July 2026 working paper from the Census Bureau’s Center for Economic Studies provides the clearest recent reason to separate those outcomes. Studying the reinstatement of ABAWD work requirements in selected counties and states after they had been suspended during the Great Recession, the authors find that SNAP participation fell by about 7 percent in their preferred estimate. Yet they find no statistically significant increase in employment or earnings. They also find that the people who left were disproportionately lower-income recipients. The remaining caseload therefore looked better off on average, not because everyone became more prosperous, but because the rule removed more people from the lower-income end of the distribution.
That combination is easy to misread. A participation decline can look like a success if participation is treated as a measure of dependence and employment is assumed to follow automatically. But the two are different events. A person can stop receiving SNAP because they found a job, because they could not find enough qualifying hours, because they could not document those hours, or because they could not secure an exemption or training placement. Administrative records can reveal the first outcome and the last outcome in aggregate without making every individual pathway visible.
The composition result makes the distinction more consequential. If lower-income people are more likely to leave than higher-income people, the average income of those still enrolled will rise mechanically. That rise does not show that anyone’s income increased. It shows that the population being measured changed. A policy can therefore appear to improve the economic profile of the caseload while worsening access for people with the least income. The question is not only whether the program reaches fewer people, but whether the people filtered out match the people the rule was supposed to identify.
The study itself is careful about this distinction. Its preferred triple-difference estimate compares adults just below age 50, who are generally subject to the ABAWD rule, with adults just above that cutoff, who are exempt from it. It also compares areas where the requirement was reinstated with nearby or same-state areas where waivers remained. That design does not tell us the story of every person who lost benefits. It does, however, make a broader before-and-after change less likely to be mistaken for the effect of the rule. The participation loss appears after the rule’s three-month grace period and persists through the follow-up period, while the preferred employment estimate remains close to zero.
CBO’s earlier synthesis of the research is useful precisely because it does not produce an identical conclusion. In 2022, the Congressional Budget Office judged that SNAP’s work requirement probably increased employment for some adult recipients without dependents, but reduced average income overall because more people lost benefits than increased their earnings. CBO summarized earlier studies that found modest employment gains among older ABAWDs in some states and much larger participation declines in others. It also noted that many ABAWDs who stopped receiving benefits had few or no other sources of income, and that only about 30 percent had earnings in the cited data. The difference does not mean one source can simply be discarded. The studies use different periods, states, comparison groups, and empirical designs. It means the strongest common conclusion is narrower than either political slogan: access falls in the Census study’s reinstatement settings, while earlier estimates vary across states and the size of any employment response depends on context.
The relevant comparison is therefore counterfactual. We need to ask what would have happened to similar people in similar places if the requirement had not been reinstated, not merely whether the treated group looked different afterward. That is why a credible result can show a large change in participation and a small or uncertain change in employment at the same time. The policy may be effective at changing the first margin without being effective at changing the second.
One rule, several outcomes
The ABAWD requirement creates a chain of decisions rather than a single behavioral command. During the period studied, an adult generally had to complete 80 hours of work or an approved work-related activity each month to avoid a limit of three months of SNAP benefits in a 36-month period. Paid work was one route, but volunteer work, employment and training, or workfare could also count. Pregnancy, physical or mental barriers, and other circumstances could produce exemptions. States could seek geographic waivers in places with weak labor markets. A person’s experience of the rule therefore depended on more than willingness to work: it depended on available jobs, qualifying activities, local administration, and the ability to prove compliance.
That structure creates at least five separate outcomes.
First is participation: whether someone remains enrolled and receives SNAP. This is the outcome most immediately affected when a time limit is applied. Second is employment: whether the person has a job, which may change independently of enrollment. Third is earnings: how much that work pays. Fourth is composition: the characteristics of the people who remain on the program after others leave. Fifth is welfare: whether the policy’s savings and any additional work outweigh the costs imposed on recipients, including lost food assistance and the time spent meeting or documenting the condition.
These outcomes can move in opposite directions. Participation can fall while employment stays flat. Average income among remaining participants can rise while some people who leave become poorer. Government spending can fall because benefits are no longer paid, even while social costs rise because the people screened out were not marginal recipients who were almost indifferent between receiving SNAP and losing it. A policy evaluation that reports only the first outcome has not yet answered whether the rule created work, improved targeting, or made people better off.
The distinction also changes how we interpret the word “screening.” Screening is not automatically good or bad. A condition could improve targeting if it were easier for people with higher earnings capacity to satisfy than for people with greater need. But a condition can screen in the wrong direction when the cost of qualifying, documenting, or navigating an exemption is higher for people with low income, unstable work, health limitations, transportation problems, or weak access to child care. In that case, the rule may remove people who most need the benefit while leaving a higher-income caseload behind.
What would count as an answer?
The useful question is not whether work requirements ever change anyone’s behavior. They may. The useful question is which pathway dominates in a defined setting, for which people, and at what cost. The recent administrative study gives strong evidence of a participation effect and a composition shift in its sample. Its preferred labor-market estimates do not detect a meaningful employment or earnings gain, but “no detectable effect” is not the same as proof that no individual worked more. The employment measure is based on covered employment records, earnings are available for only part of the sample, and longer-run or informal work may not appear.
CBO’s evidence adds another qualification. Work requirements are more likely to produce employment when recipients have access to jobs, case management, training, child care, or other supports that make work feasible. They are more likely to reduce income when recipients face barriers or when local labor demand is weak. Even people already working enough hours may lose benefits if they cannot complete the reporting process. The formal threshold and the practical route to satisfying it are not the same thing.
This also explains why the series will treat evidence quality as part of the subject, not as a technical appendix. A participation estimate needs a comparison group and a time pattern. An employment estimate needs a measure that captures the relevant work. A welfare estimate needs assumptions about who could comply, who could not, and how much lost assistance matters. Each question has a different denominator and a different uncertainty. A single number cannot carry all of them.
That is why this series will keep the rule itself in view while following its consequences outward. The next step is to see exactly how the 80-hour condition, the three-month limit, geographic waivers, exemptions, and proof of compliance turn a work-related standard into an eligibility decision. Without that institutional map, later numbers about participation or employment are easy to misunderstand. With it, the central test becomes visible: did the policy create more work, or did it mostly create another way for people to lose access?