Who Gets Screened Out—and Why That Changes the Welfare Question

The SNAP participation result becomes more consequential once we ask who leaves. In the preferred estimate, the people remaining on SNAP have higher average pre-reinstatement income, lower average benefits, and a smaller share receiving the maximum benefit. That pattern is consistent with lower-income recipients leaving at higher rates. This chapter connects that composition change to the welfare question: if a work requirement saves public spending by removing people from the rolls, are those savings large enough to justify the hassle for people who remain and the lost benefits for people who cannot feasibly comply?

The answer requires separating two kinds of analysis. The composition result is an empirical description of who remains. The welfare result is a model-based judgment about how to value their costs and the government’s savings. The first does not prove the second, but it changes which welfare costs must be counted.

The caseload changes composition

The preferred triple-difference estimate finds that, conditional on being a SNAP participant, average pre-reinstatement gross income rises by $23.58 after the work requirement is reinstated. Average SNAP benefits fall by $1.77, and the share receiving the maximum benefit falls by 2.2 percentage points. These measures use characteristics from before reinstatement, so the result is not that continuing participants suddenly earned $23.58 more or that their benefits were directly cut by $1.77. It is that the people observed among current participants are different from the people who would have been observed without the access change.

This is a composition effect. Suppose a program serves a group with both lower- and higher-income participants. If the lower-income participants leave at a higher rate, the average income of those still enrolled will rise even if nobody’s income improves. The average is responding to the population being measured. That is why a higher-income remaining caseload cannot be read as evidence that the policy made recipients more self-sufficient.

The benefit measures tell the same story from another angle. A smaller share receives the maximum benefit after reinstatement, and the average benefit among participants falls. The paper’s interpretation is that people with lower income and greater benefit eligibility are screened out disproportionately. This is not a claim that every person who lost SNAP was at the very bottom of the income distribution. It is a claim about unequal probabilities of exit across the caseload.

The prior-employment result makes the targeting concern harder to dismiss as a simple removal of people who were not working. In the preferred DDD estimate, the share of remaining participants who had been employed in the previous year does not change significantly. If the rule mainly removed people with no prior employment, we would expect the remaining caseload to show a clear increase in its prior-employment share. The study does not find that pattern in its preferred specification. The requirement appears to filter on income and benefit position more clearly than it filters on prior employment.

That finding should still be interpreted with the study’s boundaries.