key: cord-0995193-f1hesqu8 authors: Brantley, Erin; Pillai, Drishti; Ku, Leighton title: Association of Work Requirements With Supplemental Nutrition Assistance Program Participation by Race/Ethnicity and Disability Status, 2013-2017 date: 2020-06-26 journal: JAMA Netw Open DOI: 10.1001/jamanetworkopen.2020.5824 sha: 01acafaf118a4d63bac088c57d4f93b48737e2eb doc_id: 995193 cord_uid: f1hesqu8 IMPORTANCE: Increased work requirements have been proposed throughout federal safety net programs, including the Supplemental Nutrition Assistance Program (SNAP). Participation in SNAP is associated with reduced food insecurity and improved health. OBJECTIVES: To determine whether SNAP work requirements are associated with lower rates of program participation and to examine whether there are racial/ethnic disparities or spillover effects for people with disabilities, who are not intended to be affected by work requirements. DESIGN, SETTING, AND PARTICIPANTS: This nationally representative, pooled cross-sectional study examined how changes in SNAP work requirements at state and local levels in the US are associated with changes in food voucher program participation. The study combined information on state and local SNAP work requirements with repeated cross-sections from the 2012 through 2017 American Community Survey (with outcomes covering 2013 to 2017). The analytical approaches were based on difference-in-difference and triple-difference methods, after controlling for other economic and social factors. The sample included low-income adults without dependents, stratified by racial/ethnic group and disability status. The study also included parents who would otherwise meet work requirement criteria as a comparison group to estimate triple-difference models. This accounted for otherwise unobserved factors affecting trends in SNAP participation within local areas. Data were analyzed from January 2019 through March 2020. EXPOSURE: Residence in areas where SNAP work requirements apply. MAIN OUTCOMES AND MEASURES: The primary outcome is SNAP participation measured by whether anyone in the household received food vouchers at any point over the prior 12 months. RESULTS: The final analytical sample included 866 000 low-income adults (weighted mean [SE] age, 33.6 [0.01] years; 42.5% [SE, 0.07%] men). The racial/ethnic breakdown was 56.5% (SE, 0.07%) non-Hispanic white respondents, 19.4% (SE, 0.06%) non-Hispanic black respondents, 17.7% (SE, 0.06%) Hispanic respondents, 2.5% (SE, 0.02%) Asian respondents, and 3.9% (SE, 0.03%) respondents of other or multiple races. In final triple-difference models, work requirements were associated with a 4.0 percentage point decrease in participation (95% CI, –0.048 to –0.032; P < .001) for childless adults without disability, equivalent to a 21.2% reduction in SNAP participation (95% CI, –25.5% to –17.0%). For childless adults with disability, work requirements were associated with a 4.0 percentage point reduction (95% CI, –0.058 to –0.023; P < .001), equivalent to 7.8% fewer SNAP participants with disability (95% CI, –11.2% to –4.4%). When the final models were stratified by race/ethnicity, benefit reductions were larger for non-Hispanic black adults (7.2 percentage points; 95% CI, –0.092 to –0.051; P < .001) and Hispanic adults (5.5 percentage points; 95% CI, –0.072 to –0.038; P < .001) than for non-Hispanic white adults (2.6 percentage points; 95% CI, –0.035 to –0.016; P < .001). CONCLUSIONS AND RELEVANCE: Because of the association of SNAP with food security and health, work requirements that lead to benefit loss may create nutritional and health harm for low-income Americans. These findings suggest that there may be racially disparate consequences and unintended harm for those with disability. First, we estimated weighted linear probability models for each sample, using year fixed effects to account for national changes over time and Public Use Microdata Area (PUMA) fixed effects to account for time-invariant associations between PUMA and SNAP participation. This is equivalent to differencein-difference models: Yipst=β1WRpt + β2Xi + β3Mcaid_adultst + β4URpt + β5URp(t-1) + β6Povpt + αp + δt+ ipst (Equation 1) where i indexes individual, p indexes Public Use Microdata Area (PUMA), s indexes state, and t indexes year. β1 is the coefficient of interest and indicates the association between living in an area with a work requirement and SNAP participation. WR is the work requirement variable for each PUMA, described in more detail below. Xi is a vector of individual covariates, comprising age (indicator variables), gender, marital status, education, household size and home ownership (vs. renting). Mcaid_adult is the state's Medicaid eligibility for childless adults in the current year, URpt and URp(t-1) are PUMA-level unemployment rates in the current and prior years, Povpt is the current-year poverty rate, α is a set of PUMA fixed effects, δ represents year fixed effects, and ipst is the error term. The PUMA-level unemployment and poverty rates were estimated using ACS data. The unemployment rate is defined as the number of unemployed among all adults (ages 16 or older) in the civilian labor force. The poverty rate is defined as the number of people with incomes below the poverty line among all people. Next, we estimate triple difference models, using parents who would otherwise meet able-bodied adult without dependents (ABAWD) work requirement criteria as a comparison group for ABAWDs: Yipst=β1WRpt + β2ABAWDi+β3WR*ABAWDipt+ β4Xi+ β5Mcaid_adultst+ β6Mcaid_parentst + β7URpt + β8URp(t-1)+ β9Povpt + αp + δt+ ipst Where ABAWD is an indicator for being an ABAWD (exposed) versus a parent (unexposed). β3 is the coefficient of interest and indicates the interaction of living in an area with a work requirement with being an ABAWD. This model incorporates parental Medicaid eligibility (Mcaid_parent) since parents are included. Other terms are as above. For the disabled sample, we compare disabled parents with disabled childless adults. As an alternative to comparing parents vs. childless adults, we estimated models comparing childless adults ages 45 to 49 to those ages 50 to 54: Where ABAWD_age is an indicator for the 45 to 49 age group (vs 50 to 54). β2 is the coefficient of interest representing the interaction of living in a work requirement area with being in the ABAWD age range. Other terms are as in Equation 1. The model does not include a separate term for the 45 to 49 age group since this would be collinear with age indicator variables. The ACS asks respondents to report any food stamp participation in the past 12 months; the survey is fielded throughout the year. We obtained information from the Food and Nutrition Service on waivers of work requirements for each quarter. We constructed an estimate of each PUMA's exposure to work requirements that reflects the average level of work requirements over the survey year and prior year. We weighted the quarters closer to the middle of the two-year period higher (e.g., fourth quarter of the prior year and first quarter of the index year) because these quarters would be most likely to be included in the 12-month window of a date drawn randomly between January 1 and December 31st. Specifically, we use the following formula to calculate the work requirement variable: Work -0.035 -0.06 -0.019 -0.025 -0.044 requirement (-0.045 to -0.026) (-0.097 to -0.039) (-0.047 to 0.009) (-0.036 to -0.014) (-0.066 to -0.022) Race/ethnic group (ref= non-Hispanic white) NH black 0.112 n/a n/a n/a 0.073 (0.103 to 0.121) n/a n/a n/a (0.057 to 0.089) Asian -0.041 n/a n/a n/a -0.141 (-0.050 to -0.031) n/a n/a n/a (-0.182 to -0.101) Hispanic 0.004 n/a n/a n/a 0.009 (-0.004 to 0.011) n/a n/a n/a (-0.010 to 0.029) Other race 0.062 n/a n/a n/a 0.045 (0.050 to 0.074) n/a n/a n/a 0.114 n/a n/a n/a 0.063 (0.108 to 0.120) n/a n/a n/a (0.052 to 0.074) Asian -0.031 n/a n/a n/a -0.12 (-0.040 to -0.022) n/a n/a n/a (-0.152 to -0.087) Hispanic 0.009 n/a n/a n/a 0.011 (0.003 to 0.015) n/a n/a n/a (-0.002 to 0.024) Other race 0.067 n/a n/a n/a 0.049 (0.058 to 0.075) n/a n/a n/a 0.076 n/a n/a n/a 0.065 (0.064 to 0.087) n/a n/a n/a (0.050 to 0.081) Asian -0.063 n/a n/a n/a -0.114 (-0.078 to -0.047) n/a n/a n/a (-0.169 to -0.059) Hispanic -0.022 n/a n/a n/a 0.021 (-0.035 to -0.010) n/a n/a n/a (-0.000 to 0.043) Other race 0.031 n/a n/a n/a 0.056 (0.010 to 0.051) n/a n/a n/a Linear probability models with area and year fixed effects. Models also control for age, gender, marital status, education, household size and home ownership. b Comparing childless adults to parents