The Effects of SNAP Work Requirements in Reducing Participation and Benefits From 2013 to 2017

Abstract

Objectives. To assess the effects of work requirements for able-bodied adults without dependents in the Supplemental Nutrition Assistance Program (SNAP).

Methods. We used changes in waivers of work requirements to assess the impact of requiring work on the number of SNAP participants and benefit levels in 2410 US counties from 2013 to 2017 using 2-way fixed effects models.

Results. Adoption of work requirements was followed by reductions of 3.0% in total SNAP participation, 4.5% in SNAP households, and 3.8% in SNAP benefit dollars, after controlling for the unemployment, poverty, and Medicaid expansions. Because able-bodied adults without dependents comprise 8% to 9% of all SNAP participants, our findings indicate that work requirements caused more than one third of able-bodied adults without dependents to lose benefits.

Conclusions. Expansions of work requirements caused about 600 000 participants to lose SNAP benefits from 2013 to 2017 and caused a reduction of about $2.5 billion in federal SNAP benefits in 2017. The losses occurred rapidly, beginning a few months after work requirements were imposed.

Public Health Implications. SNAP work requirements rapidly reduce caseloads and benefits, reducing food and health access. Effects on participation could be similar for work requirements in Medicaid or other programs.

Requiring low-income public program recipients to work as a condition for assistance and terminating coverage if they cannot comply has become a subject of renewed policy interest. The president has promoted broader work requirements in the Supplemental Nutrition Assistance Program (SNAP; formerly the Food Stamp Program), Medicaid, and public housing.^1–3 The Department of Health and Human Services has encouraged states to submit “community engagement” waivers to impose work requirements in Medicaid and has approved several state waivers, but some have been overturned by federal courts.^4,5 The Food and Nutrition Service (FNS; US Department of Agriculture) has proposed regulations to restrict states’ ability to waive SNAP work requirements.⁶ The president’s 2020 budget proposal includes major expansions of work requirements in multiple programs.

We provide new insight into effects of work requirements by examining the natural experiment of changes in use in SNAP. Under the 1996 welfare reform law, 18- to 49-year-old able-bodied (not disabled) adults without dependents (ABAWDs) are limited to 3 months of SNAP benefits in any 36-month period if they do not comply with work requirements. Research found that similar changes to Temporary Assistance for Needy Families reduced the number of people receiving benefits while doing little to improve employment or health.^7–11 During the Great Recession (late 2000s through early 2010s), many states received waivers of ABAWD work requirements because of high unemployment. As the recession eased and unemployment fell, waivers were cancelled. By 2017, a majority of SNAP recipients lived in areas with work requirements.¹²

A principal rationale for terminating benefits for low-income adults who do not work is to reduce dependence on public assistance, on the basis of the assumption that assistance discourages poor people from working.³ A key criticism is that work requirements cause numerous participants to lose benefits and experience hardship as a result. Many who lose benefits may already be in the labor force but not working enough to meet the requirement, may be unable to find work, or may actually be working or exempt but not be able to navigate the administrative system.

Efforts to impose stricter work requirements have been rejected because of concerns about the harm they could cause. Stricter SNAP work requirements passed the House of Representatives in 2018 but were rejected by the Senate, largely because of concerns about an estimate that more than a million participants would lose coverage if policies changed.¹³ In rejecting the Centers for Medicare & Medicaid Services’ approval of Medicaid work requirements in Kentucky and Arkansas, a federal court criticized the federal agency for not considering the effects of losing Medicaid insurance coverage.⁵

Recent research demonstrates that SNAP can improve public health. SNAP participation reduces food insecurity,¹⁴ improves self-reported health,¹⁵ and increases access to health care¹⁶ while lowering medical expenditures.^17,18 Childhood SNAP participation is associated with better health in adult years.¹⁹ Policies shrinking SNAP availability could thus increase food insecurity and jeopardize health.

We analyzed the effects of work requirements on SNAP caseloads and benefits in 2410 counties from 2013 to 2017 on the basis of changes in work requirement waivers across states and counties. We used this natural experiment to conduct econometric analyses of caseload and benefit changes, controlling for local unemployment and other factors. Our results enhance the understanding of how work requirements affect SNAP and potential effects for similar policies in Medicaid or public housing, for which less data are available.

BACKGROUND

SNAP is a federally funded nutrition assistance program for low-income households that issues monthly allotments to purchase food at grocery stores, primarily through electronic benefit transfers. SNAP eligibility and benefits are mostly uniform across the nation, but there is some state flexibility through waivers. Eligibility is limited to households with gross incomes below 130% of poverty and net incomes below the poverty line.

In general, 18- to 49-year-old ABAWDs can participate in SNAP for only 3 months in any 36-month period, unless they report working at least 80 hours a month or participate in an approved job search or training program. ABAWDs reporting fewer than 80 hours of work a month are terminated after 3 months. Parents, those with disabilities, and those younger than 18 or older than 49 years are not subject to these rules. States may exempt an additional 15% of those subject to work requirements. SNAP offers states limited funding for employment and training programs offering job search, training, and related resources to encourage transitions from public benefits to work.

States may, but are not required to, apply for waivers to suspend work requirements at state or local levels because of poor economic conditions. To receive a waiver, a state must demonstrate to FNS that the area is eligible for extended unemployment benefits, has a recent unemployment rate of more than 10% or a recent 24-month average unemployment rate 20% above the national unemployment rate, or is designated as a labor surplus area.²⁰ After the 2007 recession, Congress slightly modified these policies and largely suspended the work requirements for much of 2009 and 2010. Some areas could have obtained waivers but chose not to seek them. Since then, as the economy improved and unemployment fell, states have qualified for and chosen to obtain waivers less often, and the number of people subject to ABAWD work requirements has grown appreciably.

We examined policies that existed in the 2013 to 2017 period. Proposals for stricter SNAP work requirements remain on the policy agenda. In February 2019, FNS proposed regulations to tighten conditions under which states may waive work requirements.⁶ The president’s 2020 budget proposes to expand SNAP work requirements to ABAWDs aged 18 to 64 years.³

METHODS

We used FNS administrative data, which included semiannual (January and July) SNAP household, participant, and benefit levels from 2012 to 2017.²¹ Data were available from 46 states and the District of Columbia (Maine, Rhode Island, Vermont, and West Virginia are missing). Most data were at the county level, although statewide-only data were available from 14 states (Alaska, Connecticut, Delaware, Idaho, Massachusetts, Missouri, Montana, Nebraska, New Hampshire, New York, Oregon, Utah, Washington, and Wyoming).

Our main analyses relied on county-level data from 33 states and the District of Columbia. This comprised a balanced panel of 2410 counties across 10 semiannual periods for a total sample of 24 100 observations, but the use of differences and lags reduced the analytic sample to 21 690 observations and 9 periods. Because of concerns about the omission of a number of states where county-level data were not available, we performed supplemental analyses that added statewide data from 14 states, and the results remained consistent with the main findings. We omitted data from 36 counties, which reported zero SNAP participants in at least 1 period. (They appeared to reflect agencies that closed or were consolidated. They had very few participants, so there should be no appreciable bias.) About seven eighths of total SNAP participants in the United States live in the 2410 counties represented.

FNS shared data about approved ABAWD waivers on a quarterly basis from 2012 to 2017; some waivers were on a statewide basis, but many were county or even municipality specific. In most cases, we coded work requirements on a binary basis, reflecting the absence of a waiver. Statewide waivers were applied to all counties in the state. If a waiver was applied at a subcounty level (or substate for cases in which only statewide participation data were available), we computed the fraction of the county (or state) population that lived in those jurisdictions, using geodata from the Missouri Census Data Center to estimate the fraction of the county (or state) population living in areas with work requirements. Although work requirements rose from 2013 to 2017, changes were not unidirectional; in a small proportion of counties, work requirements were imposed at some point during this period but then removed.

We also included estimates of county unemployment rates from the Bureau of Labor Statistics local area unemployment statistics, county-level poverty rates on the basis of the Census Bureau’s small area income poverty estimates, and state-level Medicaid expansions for childless adults from the Kaiser Family Foundation. A recent report found that Medicaid expansions and outreach affected SNAP and WIC (Special Supplemental Nutrition Program for Women, Infants, and Children) enrollment levels.²² Testing with multiple alternative models indicated that the best fit occurred using unemployment lagged by 6 months, that is, the county’s unemployment rate 6 months earlier. We considered the possible effects of other SNAP state waivers, such as broad-based categorical eligibility waivers, but they did not change from 2012 to 2017, so we did not include them.

We measured how changes in the implementation of work requirements (i.e., the absence of ABAWD waivers) affected SNAP caseloads and benefits over time, controlling for unemployment, poverty, and the presence of Medicaid expansions. We used a county-level 2-way fixed effects model that controlled for characteristics of each county across 10 semiannual periods from 2013 to 2017, when work requirements expanded dramatically. This covers a period when work requirements were being reintroduced in many parts of the country; we were interested in the public health implications of extending work requirements to populations that had not been previously subject to them. More recent data were not available.

For outcomes, we used 3 county-level SNAP measures reported for January and July of each year: the (1) number of households participating in SNAP, (2) number of individuals participating in SNAP (i.e., including children or other dependents), and (3) total monthly value of SNAP benefits issued across all participants in the county, adjusted to 2017 dollars by the consumer price index for food at home. These outcomes included all participants in SNAP because FNS does not report similar information on ABAWDs specifically. Basically, the model is the following:

where Δln(SNAP_tj) is the change in SNAP over 6 months; WR represents measures of work requirements in each county at time t; X is a vector of independent variables, including the county’s unemployment and poverty rates and presence of state Medicaid expansions; C is a set of dummy variables for each county; and T is a dummy variable for each semiannual period. The parameter estimates (b) are the estimated coefficients and ε is the error term. The subscript j refers to the county, and the subscript t is the semiannual period.

Because of the large variations in county sizes, we used the natural log of each measure. To further mitigate the influence of county size, we estimated the change in the natural log between the index period and the previous 6-month period. The coefficients estimate how the percentage change in SNAP levels in a 6-month period in each county is associated with the independent variables, as shown by Equation 2:

To estimate national-level effects and adjust for the varying size of counties, we weighted all analyses by the baseline county measures (households, participants, or issuance levels, respectively). We used robust SEs to adjust for state-level clustering, autocorrelation, and heteroscedasticity.

Our modeling approach used each county as its own control and focused on changes that occurred between 2013 and 2017. The fixed effects methodology controlled for time-invariant characteristics of each county, such as otherwise unmeasured differences in racial, ethnic, or educational characteristics or industry composition, and focused the analysis on the effects of changes in work requirements or other time-variant independent variables. The use of time dummies controlled for federal policy or other secular changes at the national level. Hausman tests led us to reject alternative formulations of random effects models.

RESULTS

Figures 1 and 2 illustrate overall trends in SNAP participation and work requirements from 2013 to 2017 (supplemental figures about unemployment and poverty rates and SNAP benefit levels are available from the authors). Figure 1 shows that the share of SNAP participants living in counties with work requirements rose dramatically: from 10% at the beginning of 2013, doubling to 22% by January 2014, increasing to 53% by January 2016, and then 59% by July 2017. (For brevity, results presented here are for the 2410 counties in the 33 states with county-level data; results are similar if we include statewide data from another 14 states.)

FIGURE 1—

Weighted Percentage of SNAP Participants Living in Counties With Work Requirements: US Counties, 2013–2017

Note. SNAP = Supplemental Nutrition Assistance Program.

FIGURE 2—

Average County SNAP Caseloads: US Counties, 2013–2017

Note. SNAP = Supplemental Nutrition Assistance Program.

Figure 2 shows that the average number of SNAP participants in a county fell by 13.0% from 15 646 in January 2013 to 13 676 by July 2017 and that the average number of households fell by 8.5%. The average value of SNAP benefits in a county fell by 20.5% from $2.14 million in January 2013 to $1.70 million in July 2017, using constant 2017 dollars. In 2014, payments fell after a temporary enhancement of SNAP benefit payments during the Great Recession expired.²³ The weighted average unemployment rate fell from 8.7% in January 2013 to 4.8% by July 2017, and the weighted poverty rate fell from 17.6% to 15.1%.

Alternative model testing indicated that the best fit arose when we used work requirements adopted in the quarter immediately preceding every semiannual period (e.g., the effect of October policies on January participation). Because SNAP policy permits 3 months of SNAP participation without work, caseload reductions should lag behind the imposition of work requirements by at least 3 months. Variants of our models showed that caseload declines followed adoption of work requirements; they did not precede them.

Table 1 presents key results about the effects of work requirements on the number of SNAP households and participants in each county. Because ABAWD waivers may be triggered by falling unemployment, we first analyzed models without accounting for waivers. As seen in models A and C, a 1 percentage point increase in unemployment in the previous 6 months was associated with a 1.1 percentage point increase in SNAP participation. After adding the effects of work requirements in models B and D, there was a 4.5% reduction in the number of SNAP households the quarter after work requirements are imposed and a 3.0% reduction in SNAP participation in the following quarter (P < .001 for both effects). The coefficient for unemployment was essentially unchanged (1.2%), indicating that effects of work requirements are essentially independent of effects of previous unemployment levels.

TABLE 1—

Two-Way Fixed Effects Models of Effects of Work Requirements on Changes in SNAP Caseloads: US Counties, 2013–2017

	Households, b (SE)		Participants, b (SE)
Dependent Variable	Model A	Model B	Model C	Model D
Lagged unemployment (%)	0.011 (0.003)***	0.012 (0.003)***	0.011 (0.004)**	0.011 (0.003)***
Poverty rate (%)	0.000 (0.001)	−0.001 (0.001)	0.000 (0.001)	0.000 (0.001)
Medicaid expansion	0.004 (0.007)	0.003 (0.005)	0.016 (0.012)	0.016 (0.012)
Policy in previous quarter		−0.045 (0.010)***		−0.030 (0.007)***
R2
Within	0.086	0.125	0.066	0.082
Between	0.000	0.000	0.000	0.000
Overall	0.003	0.002	0.003	0.002

Note. SNAP = Supplemental Nutrition Assistance Program. We studied 2410 counties for 21 690 observations in 9 semiannual periods.

*P < .05; **P < .01; ***P < .001.

Table 2 presents results about issuances, the total monthly value of SNAP benefits in the county. Unemployment increases were associated with a 1.0% to 1.1% increase in SNAP benefits, whereas benefits fell by 3.8% in the quarter after work requirements were added (P < .001).

TABLE 2—

Two-Way Fixed Effects Models of Effects of Work Requirements on Changes in Total Monthly SNAP Benefits (in 2017 $): US Counties, 2013–2017

	Benefits, b (SE)
Dependent Variable	Model A	Model B
Lagged unemployment (%)	0.010 (0.004)*	0.011 (0.003)**
Poverty rate (%)	0.001 (0.001)	0.000 (0.001)
Medicaid expansion	0.016 (0.012)	0.015 (0.011)
Policy in previous quarter		−0.038 (0.009)***
R2
Within	0.080	0.088
Between	0.001	0.002
Overall	0.026	0.019

Note. SNAP = Supplemental Nutrition Assistance Program. We studied 2410 counties for 21 690 observations in 9 semiannual periods.

*P < .05; **P < .01; ***P < .001.

These estimates reflect changes in overall SNAP household and participant levels, but it is important to recall that only a fraction of total SNAP caseloads is in the ABAWD target population. On the basis of American Community Survey data, we estimated that the ABAWD target population of nondisabled 18- to 49-year-old adults without children represented 8.6% of all SNAP participants in 2013, declining to 8.2% in 2017. Using SNAP quality control data, FNS similarly estimated that in 2016 the ABAWDs represented 8.8% of all SNAP participants, of which 26.0% worked (including those who worked > 20 hours a week). FNS reported that the average income of ABAWDs was 33% of poverty, lower than for other SNAP participants, and the average size of an ABAWD SNAP household was 1.2 persons.²⁴

Put in this context, our finding that work requirements were associated with a 3.0% reduction in total SNAP participants indicates that more than one third of the target population of ABAWDs lost benefits soon after work requirements were adopted.

We can interpret these results in the context of effects on an average county. The 3.0% reduction in SNAP participants is equivalent to a reduction of about 450 participants per county, whereas the 4.5% household estimate is comparable to about 330 households per county. The magnitude of effects is larger for more populous counties and smaller for less populous ones.

We estimated that expanding work requirements from counties where about 10% of SNAP participants lived in 2013 to counties where 59% lived in 2017 led to the loss of about 550 000 SNAP individual participants (or 400 000 households) for the 2410 counties in our analytic sample. The value of SNAP benefits lost was $2.2 billion in 2017. If we assume that the 2410 counties are representative of all the counties in the nation, then more than 600 000 participants were lost, and federal SNAP benefits fell by about $2.5 billion because of the expansion of work requirements from 2013 to 2017.

DISCUSSION

Our analyses indicate that imposing work requirements leads to substantial reductions in SNAP caseloads and benefits issued, even after controlling for unemployment, poverty, and Medicaid expansions. Our estimates suggest that, although improving economic conditions also contributed to caseload and benefit reductions, work requirements led about 600 000 adults to lose SNAP benefits, more than one third of all ABAWDs, and reduced benefits by $2.5 billion in 2017 alone. The caseload reductions occurred quickly, within a few months of implementation. This is consistent with reports of large and rapid caseload losses in selected areas after SNAP work requirements went into effect^25–27 and similar to the losses that occurred when Arkansas adopted work requirements for Medicaid.²⁸

These analyses refer to the effects of policies in the 2013 to 2017 period. In a proposed regulation to broaden SNAP work requirements, FNS estimated that more than 750 000 participants could lose benefits because of the revisions.⁶ The policies included in the president’s 2020 budget proposal to also impose work requirements for ABAWDs aged 50 to 64 years would likely have large effects too.

In light of research indicating that SNAP reduces food insecurity and is associated with improved health and lower health expenditures,^14–18 our analysis suggests that work requirements could create hardships for low-income adults, including increased food insecurity and impaired health. ABAWDs tend to have lower incomes than do other SNAP participants²⁰; those being terminated because they are unemployed are likely to be especially poor.

Strengths and Limitations

A strength of this study is that it uses FNS administrative data that are not subject to survey or sampling errors; there is greater precision in the measurement of the level and timing of participation and benefits. In comparison, survey data like those of the American Community Survey are subject to respondent recall errors and reference a broad period (e.g., any SNAP in the past 12 months). On the other hand, the administrative data lack detail about characteristics of SNAP participants or households, including how many are ABAWDs. Although it is reasonable to infer that work requirements primarily affect ABAWDs, a portion of the caseload decline might be because of spillover effects on other SNAP participants. Second, states may exempt up to 15% of ABAWDs who would otherwise be subject to work requirements, which could reduce impacts. We lacked county-level data to include this in our models, but the impact should be limited. Most states used no or few of their exemptions; only 4 used half or more of their available exemptions.²⁹ Third, the FNS waiver data show when waivers were approved, not when they were implemented, but we believe they were usually implemented quickly. Fourth, the national economy was generally improving in the 2013 to 2017 period, and unemployment was falling; results might differ in periods when unemployment is increasing.

The findings of this analysis are relevant to SNAP and to efforts to impose work requirements elsewhere. The Trump administration and many state policymakers seek to increase the use of work requirements in SNAP, Medicaid, and public housing. The federal government has approved requests from several states for Medicaid work requirements (Arizona, Indiana, Maine, Michigan, Ohio, Utah, and Wisconsin), although approvals for Arkansas, Kentucky and New Hampshire were overturned in court and Maine decided to not go forward), and other state Medicaid proposals are pending.³⁰

An important question, which we did not address in this study, is whether work requirements are successful in encouraging poor people to find work and raise their incomes. In recent analyses, Han³¹ and Stacy et al.³² found that SNAP work requirements did not significantly improve employment; Stacy et al. also found large caseload reductions. An analysis by Harris³³ found very small increases in employment and also found large SNAP caseload reductions. A substantial body of earlier research, including welfare to work experiments and demonstration projects, also found that work programs offered little to no long-term employment benefit and no significant health improvements.^7–11 A recent study found that Medicaid work requirements in Arkansas did not promote higher employment.³⁴

SNAP food purchases bolster grocery store income and employment, which can ripple out to other parts of the economy, such as agriculture and construction. A US Department of Agriculture analysis indicated that SNAP benefits create an economic multiplier effect: every $1 billion in SNAP benefits increases local economic activity by $1.79 billion and employment by 8900 jobs.³⁵ By reducing SNAP benefits by $2.5 billion, as we estimated for 2017, work requirements could actually lower employment, rather than boosting it.

Public Health Implications

Overall, there is scant evidence that work requirements are effective in helping people gain employment or become more self-sufficient, and there is strong evidence that work requirements create hardships. We found that work requirements caused substantial SNAP caseload reductions, controlling for economic conditions. Regulatory and budget proposals to stiffen work requirements in SNAP or Medicaid may cause many more people to lose benefits. Previous research indicates that the loss of SNAP benefits will likely harm nutrition and health.