Local Control, Discretion, and Administrative Burden: SNAP Interview Waivers and Caseloads During the COVID-19 Pandemic

Abstract

During the COVID-19 pandemic, the U.S. Department of Agriculture waived the certification interview for the Supplemental Nutritional Assistance Program (SNAP), substantially reducing the administrative burden associated with SNAP application for both applicants and agencies. Using primary policy data collected from ten county-administered states, we find that only 27% of counties implemented the interview waiver. Further, models of local decision-making indicate that public health risk, demographic vulnerability and economic need, and political orientation in the county were not statistically significant predictors of waiver use. Finally, we find that the waiver choice did affect SNAP caseloads: using difference-in-difference models that make use of the natural experiment, we find that counties that adopted the SNAP interview waivers experienced a 5% increase in SNAP caseloads.

Introduction

The COVID-19 pandemic led to a historic increase in food hardship, particularly among families with children in the United States (Gundersen et al., 2021; Schanzenbach & Pitts, 2021; Ziliak, 2021). Household Pulse Survey data from April 23 to May 5, 2020, found that nearly 10% of all adults reported that their household sometimes or often did not have enough to eat in the past 7 days (13% among households with children), whereas only 3.4% of adults reported that their household did not have enough to eat at some point over the entire year of 2019 (Center on Budget and Policy Priorities, 2020). Demand for nutrition assistance through the Supplemental Nutrition Assistance Program (SNAP) expanded accordingly. SNAP provides a monthly benefit to support food consumption among eligible low-income households averaging $129.83 per person in FY2019 before increasing to $217.33 in FY2021. From February to May 2020, SNAP enrollment grew by 17%, three times faster than any previous 3 months since the 1970s (The New York Times, 2021; U.S. Department of Agriculture Food and Nutrition Service, 2020). The federal government reimburses states for the full cost of SNAP benefits dispersed and 50% of the administrative costs (U.S. Department of Agriculture, Office of Inspector General, 2016).

Despite the growing need for nutrition assistance programs, federally mandated application processes, such as requirements for individual interviews and in-person certification appointments, created public health risks and impeded enrollment through the associated level of administrative burden. In response, the United States Congress passed legislation to allow the U.S. Department of Agriculture to grant states waivers to reduce the administrative burden and public health risks associated with the usual enrollment practices for SNAP.

Under federal program rules, SNAP requires new and recertifying applicants to complete a certification interview prior to benefit approval, which most often occurs over the telephone. These interviews allow eligibility workers to gather accurate household information and for applicants to ask questions about navigating the SNAP program. Proponents contend that SNAP certification interviews protect program integrity by ensuring only eligible applicants receive benefits and that the benefit amount successful applicants receive is accurate. Interviews might also provide SNAP applicants with information about how to connect to additional programs. However, the certification interview is also part of a set of hurdles constituting administrative burdens in the SNAP enrollment process (Moynihan et al., 2015). Unlike online eligibility forms, which applicants can complete at their convenience, certification interviews must occur during business hours and potentially require applicants to take time off work and answer intrusive questions about their household circumstances. If applicants miss their scheduled interviews, they risk being denied benefits.

In March 2020, the USDA waived the SNAP interview requirement through May 2020 for all states. From June to September 2020, states that wanted to maintain the policy waivers had to apply for approval each month from the USDA. The Continuing Appropriations Act of 2021 (PL 116-159), however, converted SNAP policy waivers back to blanket state options from October 2020 through June 2021. ¹ However, while states were granted permission to waive the certification interview for SNAP, they were not required to do so. Within the 10 states that administer SNAP at the county level, this decision was made at the discretion of the local agency that delivers SNAP benefits. SNAP interview waivers reflect a broader COVID-era trend of intergovernmental collaboration in which higher levels of government defer to lower levels of government so that administration best fits the “peculiar needs” of local residents (Benton, 2020).

This set of circumstances provides an opportunity to study how local discretion affects program enrollment, in this case, for the SNAP program to overcome administrative burden. Furthermore, we ask which local factors were most closely related to local offices choosing to use their discretion to waive SNAP certification interviews. Finally, we take advantage of a natural experiment to examine the consequences of removing certification interviews, a significant hurdle in the SNAP application process, on caseloads. Interviews are prevalent in the U.S. safety net. SNAP, WIC, Medicaid, and TANF require interviews in some form (Holcomb et al., 2003). While a few papers investigate how transitioning SNAP certification interviews from in-person to telephone impacted enrollment (Bartfeld et al., 2015; Ganong & Liebman, 2018), the evidence is mixed and even less is known about the extent to which the interview itself (rather than its form) discourages participation. To the best of our knowledge, this is the first study to estimate the impacts of eliminating certification interviews in SNAP on program participation. Given the thick presence of interviews in other U.S. safety net programs, our results are also informative about the consequences of efforts to relax similar requirements in other safety net programs.

This article proceeds in four sections: the first section reviews the literature on administrative burden and the role of discretion at the local level. The second section connects this larger literature to the current study: the federal waiver of the interview requirement for SNAP certification and our primary data collection effort to document if counties were implementing the federal waiver during the beginning of 2021. The third section describes our analysis of predictors of local adoption of the interview waiver as well as the effect of adoption on county SNAP caseloads. In the final section, we draw conclusions regarding how local agencies make decisions regarding administrative burdens during a pandemic and the limitations of the federal waiver process. Our findings speak to the flexibility of local public institutions in times of need, the extent to which administrative requirements are correlated with public health conditions and partisanship, and the consequences of relying on local discretion to reduce administrative burden.

Theory

Administrative Burden

A relatively recent literature in the public administration field has called attention to the administrative burdens recipients experience when engaging with public service providers. Moynihan et al. (2015) proposed a conceptual framework for studying administrative burdens that features social equity and behavioral science, concluding that social welfare programs for low-income populations, including the SNAP program, tend to have large administrative burdens for those who are qualified (Herd & Moynihan, 2019). Their framework recognizes burdens as consisting of three distinct costs citizens face when interacting with the government: learning costs, psychological costs, and compliance costs (Moynihan et al., 2015). Learning costs are the time and effort one must expend to learn about a program and understand its benefits, eligibility criteria, and enrollment processes (Moynihan et al., 2015). These can be thought of as search costs (collecting information about government assistance programs and appraising their relevance to the individual). Compliance costs are activities applicants must engage in to prove and maintain eligibility, such as submitting income documentation, providing proof of residence, and undergoing interviews (Moynihan et al., 2015). Psychological costs include the stigma, frustration, loss of autonomy, and stress associated with participating in government assistance programs. Barnes (2021) proposed incorporating redemption costs into the administrative burden framework, given the well-documented hassles associated with redeeming voucher-based public benefits.

The consequences of administrative burdens are numerous. Scholars have devoted considerable attention to documenting how administrative burdens can suppress the uptake of public assistance programs and lead to “administrative exclusion”: unrealized access to public services not attributable to individual preference or eligibility status (Brodkin & Majmundar, 2010). Administrative burdens can also be viewed through a “policy feedback” lens in which burdens shape, not only access to public services, but also to clients’ perceptions of government and bureaucracy more generally (Bruch et al., 2010; Moynihan & Soss, 2014).

In this study, we examine the compliance costs associated with certification interviews for SNAP applicants. Compliance costs are a common explanation for nonparticipation in government assistance programs (Herd & Moynihan, 2019). Surveys of likely eligible SNAP nonparticipants reveal application costs as a primary barrier to enrollment (Bartlett et al., 2004). Requiring SNAP recipients to recertify their eligibility more frequently reduced participation (Kabbani & Wilde, 2003). Evidence suggests that relaxing compliance costs boosts program uptake. For example, SNAP enrollment grew following policy changes that softened asset tests, extended certification periods, and simplified applicant reporting requirements (Hanratty, 2006; Ratcliffe et al., 2008). Some have documented how program modernization can ease compliance costs and encourage enrollment. For instance, the adoption of electronic benefit transfer (EBT) cards and online applications improved SNAP uptake (Schwabish, 2012; Shiferaw, 2019).

Based on this prior research, we hypothesize that counties that implement the federal waiver of the interview requirement and lower compliance costs associated with participation will have higher SNAP caseloads relative to other counties who do not waive the interview requirement (H1a). However, if the SNAP interview provides an opportunity for applicants to receive assistance helpful to completing the application, we may alternatively find that SNAP caseloads decrease in counties that adopt the interview waiver (H1b).

Discretion and the Drivers of Local Decision-Making

An important insight from recent public administration scholarship is that administrative burdens are constructed, and as such, political actors can wield them to advance ideological goals (Herd & Moynihan, 2019). However, administrative burdens are not “self-implementing” (Bell et al., 2021). Administrative burdens arising from program design are filtered through government bureaucracy and, in the U.S. federalist system, increasingly decentralized implementation, whereby local administrators may use discretion to relax or amplify them.

While throughout much of SNAP's history the program was implemented uniformly across states, beginning with the 1996 Welfare Reform Act and continuing through additional regulatory and statutory changes, the federal government provided states with a great deal of discretion in terms of the administrative processes involved in program enrollment and recertification practices. For example, while some states adopted fingerprint and drug testing requirements, other states reduced reporting requirements. The existing evidence suggests that state variation in administrative practices is associated with the size of SNAP caseloads (Bartfeld et al., 2020; Dickert-Conlin et al., 2021; Kabbani & Wilde, 2003; Klerman & Danielson, 2011; Ratcliffe et al., 2008; Stacy et al., 2018). One recent estimate suggests that if states had been forced to maintain policy and practices in place in 2000, SNAP caseload growth over the 2000–2016 period would have been 38% lower (Dickert-Conlin et al., 2021).

What determines local bureaucratic decisions about whether to soften or harden administrative burdens in the execution of national programs? Local discretion may allow political elites to react to local conditions efficiently to advance policy goals. For example, resistance to statewide shutdowns in response to the introduction of COVID-19 into the population in March 2020 was partly predicated on the notion that the risk was not the same across all the states or even across counties within the same state. To examine if local political actors are making decisions in response to local information regarding the public health situation, we hypothesize that SNAP interview waivers were adopted more aggressively in counties with a higher risk of COVID due to higher demand for SNAP to protect both SNAP recipients and agency staff (H2). We explore this possibility by directly including a measure of county COVID rates and deaths in our model of SNAP waiver adoption.

Local policy actors may also be motivated by the vulnerabilities and needs of their local population. Federal SNAP dollars are an effective countercyclical tool that provides an economic boost to the local economy since SNAP dollars are spent at local food retailers. According to USDA estimates, $1 in SNAP benefits generates $1.50 in local economic activity (Canning & Stacy, 2019). Local agency staff are likely aware of the size and composition of the vulnerable population within the county that might benefit from SNAP benefits. Based on the idea of local political actors as responsive to constituent needs, we hypothesize that county measures of demographic vulnerability and economic need are positively associated with adoption of SNAP interview waivers (H3). To that end, we include the percent nonwhite, the percent 65 and over, the percent under age 18, the percent with a college degree or more, the percent food insecure, the percent in poverty, and the change in the share of the county's population that was employed from January 2020 to January 2021, which we term employment share change.

Finally, ideological beliefs of street-level bureaucrats have been shown to influence their perceptions of administrative burden (Bell et al., 2021). Local political authorities may therefore exert influence on the implementation of federal programs. Indeed, May and Winter (2009) document evidence that street-level bureaucrats were more likely to pursue practices that diverged from national policy when local politicians who were most proximate disagreed with national goals. A local community's political orientation may also affect how programs are implemented at the local level. Specifically, Stensöta (2012) found that the political orientation of a local community interacts with the ideology of street-level bureaucrats. Using data on the political ideology of street-level bureaucrats and their local community, Stensöta shows that, when positioned in right-leaning local communities, right-leaning street-level bureaucrats used discretion to construct less generous policy environments. More recent evidence about the role of political ideology in adapting to COVID-19 specifically suggests that local actors were strongly motivated by political ideology (Gadarian et al., 2021; Rodriguez et al., 2022). Using a measure of the share of voters for the 2020 Republican candidate for president, we hypothesize that local political ideology will be associated with the decision to adopt the SNAP interview waiver (H4).

Methods

Current Study

To document how county agencies implemented federal policy waivers for SNAP on the ground, we collected data on the enrollment processes for new SNAP applications in the 10 states with county-level SNAP-administered programs. In these states, county social service offices perform program functions and report to locally elected boards of supervisors or commissioners (Kogan, 2017; U.S. Department of Agriculture, Office of Inspector General, 2016). Counties in these states also share administrative costs with state agencies and contribute local funds toward the program (Geller et al., 2019; National Association of Counties, 2022).

From December 2020 to January 2021, we identified county SNAP telephone numbers through a review of state websites. Between January 5 and April 30, 2021, trained interviewers telephoned all 626 SNAP local information lines (112 counties shared telephone lines) across the 10 states that administer SNAP at the county level (California, Colorado, Minnesota, New Jersey, New York, North Carolina, North Dakota, Ohio, Virginia, and Wisconsin) (U.S. Department of Agriculture Food and Nutrition Service, 2018) for a sampling frame of 738 counties. To reach county offices, interviewers called each phone number up to three times on separate days. On the third attempt, interviewers remained on the line until either a local agency worker answered, or the call timed out. Interviewers were successful 88% of the time. ² Each completed call lasted less than 3 min, on average. In total, we were able to document conditions for submitting a new application for SNAP in 647 counties.

While interstate differences in policy discretion are potentially an interesting source of variation, we focused here on differences in policy implementation at the county level. For states that administer SNAP at the state level, we expect local experiences with SNAP to be similar given the uniformity of IT infrastructure and policy guidance. For these reasons, we limit our analytic sample to states that administered SNAP at the county level to observe local-level decision-making and discretion and the consequences of those decisions. While our focus on county-administered states limits the generalizability of our SNAP results, the set of states sampled is heterogeneous with respect to geographic regions, population size, and density.

Analysis

Our analysis proceeds in three parts. First, we present our descriptive findings from our primary policy data on waiver adoption at the county level. Second, we explore correlates of county waiver adoption to test our series of hypotheses (H2, H3, and H4) regarding the local factors that may predict local policy implementation. Specifically, we examine the role of public health conditions (new COVID infections and COVID-related deaths in January 2021), county demographic vulnerability and characteristics (percent racial minorities, percent over 64, percent less than age 18, population size, percent with a college degree or more), economic need (percent food insecure, percent poverty, percentage point change in county employment between January 2020 and January 2021), and, finally, local political orientation (percent 2020 Republican presidential vote share). Third, we estimate the impact of SNAP interview waivers on program caseloads.

Our COVID measures come from the New York Times COVID database (The New York Times, 2022). The New York Times COVID database compiles information from individual states and counties and is one of the only organizations that provides county-level COVID data in a single platform for the entire United States. For demographic and economic vulnerability characteristics, we rely on 2019 5-year county estimates from the American Community Survey (ACS) published by IPUMS-NHGIS (Manson, 2020). The ACS is a Census-administered nationwide survey that asks about a variety of topics including employment, receipt of public benefits, and individual characteristics. The primary advantage of using the ACS includes the availability of narrow geographic information that allows us to merge to our SNAP waiver data. Our 2019 county population estimates are from the U.S. Census compiled by USDA's Economic Research Service (Economic Research Service, 2019). Our economic need variables are from several sources. We obtained 2019 county-level food insecurity from the 2019 County Health Rankings (University of Wisconsin Population Health Institute, 2019) and poverty data and the U.S. Census Small Area Income and Poverty Estimates (U.S. Census, 2019). We rely on the Bureau of Labor Statistics’ Local Area Unemployment Statistics for monthly county-level employment data (U.S. Bureau of Labor Statistics, 2023). We obtained county-level election return data from the MIT Election Data + Science Lab (Massachusetts Institute of Technology, 2020).

Table 1 presents the county demographic information for sample counties (n = 738), where our sample is defined as all counties in the 10 states which administer SNAP at the county level, and non-sample counties (n = 2,258), the remaining counties in the United States where SNAP is administered at the state level. We compared differences in means across these groups to assess the external validity of our study. Sample counties (i.e., counties in states that administer SNAP at the county level) differ from non-sample counties (i.e., those that administer SNAP at the state level) in many respects (compare column 1 to column 6): sample counties had fewer COVID deaths, a smaller percentage of their populations under 18, and larger populations than non-sample counties. Sample counties were also more highly educated, less food insecure, less poor, experienced larger declines in employment, and less likely to vote Republican for president in 2020. These county-level differences should be noted when considering the external validity of our results.

Table 1.

Summary Statistics, Respondent, Nonrespondent, and Nonsample Counties.

	Sample counties (from 10 states that administer SNAP at the county level)					Nonsample counties
	Respondent and nonrespondent counties	Respondent counties			Nonrespondent counties
	Full sample	Full sample	No waiver	Waiver	Full sample	Full sample
	(1)	(2)	(3)	(4)	(5)	(6)
COVID-19 cases (per 1,000)	17.12	15.86	15.69	16.32	26.11***	16.72
COVID-19 deaths (per 1,000)	0.26	0.25	0.25	0.24	0.35***	0.38***
Percent nonwhite	16.98	15.95	15.32	17.6*	24.34***	15.98
Percent > 64	18.87	19.09	19.01	19.3	17.33***	18.95
Percent < 18	21.38	21.25	21.51	20.57***	22.31***	22.47***
Population (100,000)	1.58	1.47	1.61	1.12	2.32	0.87***
Percent college or more	25.48	25.5	25.11	26.52	25.35	20.89***
Percent food insecure	11.75	11.64	11.52	11.97	12.55**	14.2***
Percent poverty	12.45	12.36	12.36	12.36	13.11	14.93***
Employment share change	−1.9	−1.88	−1.9	−1.83	−2.05	−1.33***
Percent R 2020 presidential Vote share	57.52	57.72	59.09	54.08***	56.11	67.56***
Observations	738	647	470	177	91	2,258

Note. Sample counties include all counties/county equivalents in the 10 states that administer SNAP at the county level. Nonsample counties contain all counties in the remaining states that administer SNAP at the state level. Respondent indicates counties we were able to reach via telephone (nonrespondent were unreachable by telephone). Asterisks in column 4 display results from t tests for differences in means by waiver status in respondent counties. Asterisks in column 5 display results from t tests for differences in means for the full respondent sample compared to the full sample of nonrespondent counties. Asterisks in column 6 display results from t tests for differences in means for the full sample of respondent and nonrespondent counties relative to the full sample of non-sample counties.

* p < .1. ** p < .05. *** p < .01.

Among our sample counties (n = 738), respondent counties (n = 647) (counties we were able to interview) differ from nonrespondent counties (n = 91) in that respondent counties experienced fewer COVID cases and deaths and were less racially diverse, older, and less food insecure (compare column 2 to column 5). We observed no statistically significant differences between respondent and nonrespondent counties with respect to population size, share with a college degree or more, percent poverty, employment share change, and Republican 2020 presidential vote share. Finally, Table 1 also compares characteristics of waiver (n = 177) and nonwaiver (n = 470) counties among respondent counties (compare column 3 to column 4). We find only three differences that are statistically significant: waiver counties were, on average, more racially diverse, had a slightly lower share of the population under age 18, and were less likely to have voted for the Republican for president in 2020.

We begin with a set of simple linear probability models for the use of waivers within the county with standard errors clustered at the state level in which we enter each set of variables in four groups. Model 1 examines county public health conditions. Model 2 examines demographic vulnerability, economic need, and county characteristics. Model 3 focuses on the share of voters that supported the 2020 Republican presidential candidate. In Model 4, we include all the covariates, and in Model 5, we also include state fixed effects to control for unmeasured characteristics at the state level that are constant over time. We also assess the sensitivity of our results to probit estimation in Table A1 in the Appendix.

Finally, we explore the consequences of those local decisions on SNAP caseloads using a difference-in-difference (DD) regression model (H1). We obtained monthly county-level SNAP caseload data from each state in our sample. We analyze SNAP caseloads 5 months prior to SNAP interview waiver availability (January–May 2019) and the same 5 months after USDA made waivers available (January–May 2021). Specifically, we estimate the following regression equation:

$$Y_{ct}={\beta }_0+{\beta }_{1}Wai{v}_c+{\beta }_{2}Pos{t}_t+{\beta }_3(Waiv*Post{)}_{ct}+{\gamma }_s+{\epsilon }_{ct}$$ (1)

Our outcome is SNAP caseloads (per 100,000) in county c in month t. $Wai{v}_c$ is a binary indicator for counties that reported waiving the SNAP interview. $Pos{t}_t$ is an indicator for the months in which SNAP interview waivers were available (1 = January–May 2021; 0 = January–May 2019). The coefficient ${\beta }_3$ on our interaction term, $(Waiv*Post{)}_{ct}$ , is the focus of our main hypothesis and measures the difference in SNAP caseloads in the counties waiving the interview relative to counties with an interview in the post period, or our DD coefficient. We also control for state fixed effects ( ${\gamma }_s$ ) in a separate regression. We cluster standard errors at the state level.

We also check the robustness of our findings to the inclusion of county fixed effects and month-year fixed effects using the following two-way fixed effects regression:

$$Y_{ct}={\beta }_0+{\beta }_1(Waiv*Post{)}_{ct}+{\gamma }_c+{\gamma }_t+{\epsilon }_{ct}$$ (2)

where ${\gamma }_c$ are county fixed effects and ${\gamma }_t$ are month-year fixed effects. This result will signal if our coefficient of interest from equation (1) is biased by factors that are constant within counties (such as the progressivity of the population) or within months and years. The results shown in Table A2 of the Appendix are robust to this specification. We also test for the presence of differential trends in outcomes between waiver and nonwaiver counties before the USDA made SNAP interview waivers available to assess the possibility that pre-trends might bias our results. Specifically, we estimate an event-study regression of the following form:

$$Y_{ct}={\beta }_0+{\beta }_{1}Wai{v}_c+{\mathrm{\Sigma }}_{n=-N,n\ne -1}^N{\beta }_{n}Wai{v}_{ct}^n+{\gamma }_t+{\epsilon }_{ct}$$ (3)

where $Y_{ct}$ and $Wai{v}_c$ are unchanged from equation (1). $Wai{v}_{ct}^n$ are a set of dummy variables indicating each county's timing relative to waiver availability (we omit month-year prior to availability) and ${\gamma }_t$ are month-year fixed effects. We cluster standard errors at the state level.

Results

We begin by presenting information on the implementation of the federal waiver removing the requirement for a certification interview for SNAP benefits (Table 2). Only 27% of counties reported waiving the SNAP certification interview despite explicit permission from the federal government to do so. For the remaining counties, interviews were conducted following standard protocol, despite the federal waiver.

Table 2.

County Implementation of SNAP Interview Waiver by State, Spring 2021.

	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)	(10)	(11)
	CA	CO	MN	NJ	NY	NC	ND	OH	VA	WI	Total
No Waiver	81%	93%	100%	83%	86%	51%	100%	97%	60%	3%	72%
Waiver	19%	7%	0%	17%	14%	49%	0%	3%	37%	97%	27%

Note. Items reflect unweighted percentages among respondent counties (n = 647 out of 738 total counties). About 3% of localities in Virginia reported an unknown waiver status, so values do not sum to 100.

Local discretion at the county level was present as evidenced by the significant level of within-state variation in the SNAP interview requirement: only Minnesota and North Dakota reported the same process (interview required) in every county while the other eight states reported differences across counties, with only some counties implementing the federal waiver no longer requiring a certification interview. However, only four of the nonwaiver counties in the sample required applicants to be physically present at the county office to enroll in SNAP (results not shown). Therefore, SNAP applicants could satisfy the interview requirement via the telephone in nearly all the counties maintaining the interview requirement.

Next, we examine predictors of local discretion to adopt the SNAP interview waiver in Table 3. In Model 1, we find no evidence that public health conditions were related to the adoption of the SNAP interview waiver (H2). This is somewhat surprising, given espoused beliefs regarding the need to let local officials use science to guide local access to services, and contrary to our hypothesis. In Model 2, we find that counties with larger populations were less likely to adopt the SNAP interview waiver. Measures of demographic vulnerability and economic need, on the other hand, were not statistically significant predictors of waiver use. This finding does not support our hypothesis that local officials may be looking out for the best interests of the most vulnerable members of their local population (H3). Finally, in Model 3, we observe that counties which had a higher level of support for the Republican presidential candidate in 2020 were also less likely to adopt the SNAP waiver. The size of this coefficient suggests that a 10 percentage point increase in Republican vote share was associated with a 4 percentage point decrease in the probability of SNAP waiver adoption, a nontrivial change from a base of 27%, which supports our hypothesis about the role of political ideology in the implementation process (H4). Taken together, results from Models 1–3 suggest that perhaps political ideology trumped the local needs of the county population, providing tenuous support for Hypothesis 4.

Table 3.

Predictors of SNAP Interview Waiver Adoption, OLS Model.

	(1)	(2)	(3)	(4)	(5)
Public health conditions
COVID-19 cases (per 1,000)	0.003			0.002	−0.002
	(0.008)			(0.005)	(0.001)
COVID-19 deaths (per 1,000)	−0.045			0.008	−0.033
	(0.056)			(0.056)	(0.037)
Demographic vulnerability, economic need, and county characteristics
Percent nonwhite		0.004		0.001	−0.002
		(0.003)		(0.006)	(0.002)
Percent > 64		−0.006		−0.002	−0.002
		(0.009)		(0.007)	(0.003)
Percent < 18		−0.025		−0.020	−0.002
		(0.014)		(0.011)	(0.007)
Population (100,000)		−0.006*		−0.007*	−0.003
		(0.003)		(0.004)	(0.002)
Percent college or more		0.000		−0.005	0.002
		(0.005)		(0.011)	(0.003)
Percent food insecure		0.004		0.001	−0.002
		(0.021)		(0.018)	(0.008)
Percent poverty		−0.011		−0.012	0.002
		(0.014)		(0.015)	(0.006)
Employment share change		0.018		0.019	0.001
		(0.028)		(0.026)	(0.005)
Local political orientation
Percent R 2020 presidential vote share			−0.004**	−0.007	−0.002
			(0.002)	(0.008)	(0.002)
State FE	No	No	No	No	Yes
Observations	647	647	647	647	647

Note. Standard errors clustered at the state level and shown in parentheses.

* p < .10. ** p < .05. *** p < .01.

However, when we estimate a model that includes all covariates, Model 4, we find that only the size of the population remains statistically significant. That is, the coefficient on Republican vote share is no longer statistically significant. When we include state fixed effects in Model 5, we observe that that population becomes statistically insignificant. Model 5 requires variation across states to be identified. Thus, the two states in our sample without variation in waiver adoption are not contributing to identification in Model 5. ³ In the end, we do not find robust support for any of our hypotheses 2–4 regarding what drives the local discretion of political actors around policy implementation decisions.

Finally, we turn to our central hypothesis (H1) regarding administrative burden and the effect of SNAP waiver implementation decisions on county SNAP caseloads. Table 4 presents results from our difference-in-difference model described using equation (1). Model 1 presents results without controlling for state fixed effects and Model 2 includes state fixed effects, which adjusts for time invariant state-level factors that may bias our DD coefficient on $(Waiv*Post{)}_{ct}$ . While both waiver and non-wavier counties experienced an average increase in the SNAP caseloads in the post-waiver of 1,009 (per 100,000), waiver counties experienced an additional increase of 478 (per 100,000) SNAP participants over and above nonwaiver counties. This increase represents a sizable (4.8%) increase in SNAP caseloads due to the adoption of the SNAP waiver. The coefficient on $Wai{v}_c$ is statistically insignificant in both models with large standard errors; the change in the sign on $Wai{v}_c$ from Model 1 to Model 2 may indicate that state conditions that are constant across the time period do bias results on $Wai{v}_c$ but there is no evidence that our DD coefficient, $(Waiv*Post{)}_{ct}$ , changes with the inclusion of state fixed effects. In results presented in Table A2 in the Appendix, we evaluate the robustness of our DD coefficient using the two-way fixed effects regression described by equation (2). In this specification, we replace state fixed effects with county fixed effects and replace $Pos{t}_t$ with month-year fixed effects. The coefficient on the interaction term $(Waiv*Post{)}_{ct}$ is unchanged and remains significant (p < .05), suggesting that waiving the interview requirement led to an increase in SNAP caseloads and supporting Hypothesis 1a or over 1b.

Table 4.

Difference-in-Differences Estimates of the Effect of SNAP Interview Waiver on SNAP Caseloads.

	(1)	(2)
Waiv	726.2	−1200.1
	(985.97)	(975.34)
Post	1009.0***	1009.0***
	(201.05)	(201.21)
Waiv × Post	478.4**	478.4**
	(207.46)	(207.61)
State FE	No	Yes
Mean of dependent variable	9,997	9,997
DD estimate as % of mean dep. var.	+4.8%	+4.8%
Observations	6,020	6,020

Note. Standard errors clustered at the state level and shown in parentheses.

* p < .10. ** p < .05. *** p < .01. Post-period is defined as Jan–May 2021 (pre-period is Jan–May 2019).

Our difference-in-difference model assumes SNAP caseloads for waiver and nonwaiver counties would have followed similar trends in the absence of the waiver. This assumption is untestable but more plausible if SNAP caseloads for both groups exhibited parallel trends in the pre-waiver period (Angrist & Pischke, 2008). Figure 1 presents event-study estimates testing for differential trends between waiver and nonwaiver counties before and after the USDA made interview waivers available. Event-study coefficients prior to waiver availability (left of the vertical line) are reliably near zero and not statistically significant. In other words, we do not find evidence that SNAP caseloads were trending differently for waiver and nonwaiver counties prior to waiver availability. If anything, SNAP caseloads appear to be trending down slightly for waiver counties relative to nonwaiver counties in the pre-waiver period, which would bias our estimates against finding a positive effect of interview waivers on caseloads. However, event-study coefficients are positive and statistically significant in 3 out of the 5 months following waiver availability (right of the vertical line), consistent with the positive and statistically significant DD estimate in Table 4. Moreover, the event-study estimates in the post-wavier period are similar in magnitude to the DD estimates.

Figure 1.

Event-study estimates of impact of interview waiver on SNAP caseloads (per 100,000).

Note. Figure plots coefficients on indicators for time until or since SNAP interview waiver availability. Month zero is the first month waivers were available (vertical line denotes month prior to availability). Model includes month-year fixed effects. Standard errors are clustered at the state level.

A potential threat to the parallel trends assumption is if counties imposed the SNAP interview waiver in response to or in anticipation of rising caseloads. We leverage a few findings from our analysis to assess the plausibility of this potential violation. First, trends in SNAP caseloads appear parallel between waiver and nonwaiver counties throughout the entire pre-period. If counties were imposing waivers in response to rising caseloads, we would expect outcomes for waiver counties to begin trending upward relative to nonwaiver counties slightly before waiver adoption. We find no such evidence (see Figure 1). If anything, SNAP caseloads appear to decline for waiver counties relative to nonwaiver counties in the pre-waiver period. Moreover, if counties were implementing the waiver in anticipation of rising caseloads, we would expect changes in local economic conditions to be predictive of waiver adoption. Specifically, we would expect counties experiencing more economic distress following COVID to have higher SNAP caseloads and potentially a greater incentive to implement the interview waiver. However, recall from Table 3 we found that our measure of changes in local economic conditions was not a statistically significant predictor of waiver adoption, both in the stepwise model (column 2) and with the full set of controls (columns 4–5). Together, these findings do not support the notion that counties implemented SNAP interview waivers in response to or anticipation of rising caseloads, providing further support for Hypothesis 1a over 1b.

Discussion and Conclusion

This study makes use of a natural experiment, the COVID-era federal waivers available to SNAP agencies to reduce the administrative burden associated with program application and recertification, to explore a series of hypotheses around drivers of local discretion (Hypotheses 2–4) and the importance of administrative burden on SNAP caseloads (Hypothesis 1). We find that, when given the opportunity to do so, most counties in county administered states did not implement the SNAP interview waiver, which would have reduced the administrative burden associated with the program. We also find that, contrary to our expectations, the local public health situation, the demographic vulnerability and economic need, and the political orientation of the local population were not statistically significant predictors of adoption (Hypothesis 2–4). Finally, we find that these local points of discretion around administrative burden may be quite consequential: counties that adopted the SNAP interview waiver had 5% higher SNAP caseloads compared to counties that decided to not adopt the waiver, providing support for Hypothesis 1a.

Why might SNAP agencies have had such low waiver utilization? Since most states already offered SNAP interviews over the phone, completing the interview requirement potentially represented an administrative burden for the applicant and the agency but may not have been seen as posing a direct public health risk. Thus, low utilization of the waiver among the counties may reflect a lack of understanding of how the interview waiver may pose a burden on applicants.

These results are consistent with research by Heflin et al. (2020) in their examination of predictors of SNAP policies that would increase eligibility for noncustodial parents paying child support. In that study, which explored the state-level decision to treat child support paid more favorable for SNAP benefit determination over 2001–2017, they found that the state decision was not related to political ideology or economic conditions at the state level. However, there was suggestive evidence that the level of per capita state SNAP administrative costs and the state income tax revenue were associated with the decision to adopt or revoke the child support income exclusion.

Our study, like all non-experimental research, is not without limitations. First, we study a rather unique natural experiment related to the relaxation of federal guidelines related to a global pandemic that is unlikely to reoccur. It is unclear if the findings here—both in regards to the determinants of local decision-making at the county level and the importance of administrative burden to SNAP caseloads are replicable under more typical conditions. Second, because we study the county-level implementation of SNAP, our study is limited to the 10 states that administer SNAP at the county level. It is possible that these findings are not replicable in other geographic areas. Third, our aggregated SNAP caseload data does not permit us to decompose enrollment effects by different types of participation. We cannot determine, for example, whether increased program retention or initial enrollment drives our estimates.

We find that the elimination of SNAP interview requirements increased caseloads, suggesting that interview requirements in SNAP represent an administrative burden that suppresses program participation. However, proponents argue that SNAP interviews might safeguard program integrity or provide applicants with information about how to navigate the program. Of course, administrative cost savings associated with waiving the SNAP interview might obviate these concerns. Future research ought to consider how relaxing SNAP interview requirements affected other outcomes of interest, such as administrative costs, and undermined other policy goals, such as program integrity.

In summary, while waiving specific application requirements is an important federal policy lever to facilitate access to social programs that provide nutrition assistance during a public health crisis, SNAP interview waivers in these 10 county-administered SNAP states were not broadly or uniformly implemented both within and across states. Moreover, we find that county implementation decisions were largely insensitive to local conditions, suggesting a potential disconnect between federal intent and state and local implementation of SNAP interview waivers (Benton, 2018). On-going discussions to make the SNAP interview and other federal policy waivers adopted during COVID-19 permanent should take into consideration the extent to which local agency practices actually respond to the availability of federal policy waivers and whether waiver implementation is tailored to meet the needs of local communities. Local discretion in the implementation of federal social welfare programs may undermine the efficacy of policy options designed to reduce administrative burden, despite the local benefits of these efforts.

Author biographies

Colleen Heflin is Associate Dean, Chair, and Full Professor in the Public Administration and International Affairs Department at the Maxwell School at Syracuse University. Her research focuses on documenting the causes and consequences of food insecurity, identifying barriers and consequences of participation in nutrition programs, and understanding the changing role of the public safety in the lives of low-income Americans.

William Clay Fannin is a doctoral candidate in the Department of Public Administration and International Affairs at Syracuse University and graduate research associate in the Center for Policy Research. His research explores the consequences of social policy design and implementation on low-income households. L.

Leonard Lopoo is the Paul Volcker Chair in Behavioral Economics and Full Professor in the Department of Public Administration and International Affairs and Director of the Maxwell X Lab at the Maxwell School at Syracuse University. He has research interests in the fields of child and family policy and behavioral public policy.

Appendix

Table A1.

Predictors of SNAP Interview Waiver Adoption, Probit Model.

	(1)	(2)	(3)	(4)	(5)
Public health conditions
COVID-19 cases (per 1,000)	0.003			0.003	−0.001
	(0.008)			(0.005)	(0.002)
COVID-19 deaths (per 1,000)	−0.045			0.020	−0.028
	(0.056)			(0.064)	(0.038)
Demographic vulnerability, economic need, and county characteristics
Percent nonwhite		0.004		0.001	−0.002
		(0.003)		(0.005)	(0.003)
Percent > 64		−0.006		−0.003	−0.003
		(0.009)		(0.006)	(0.004)
Percent < 18		−0.026*		−0.021*	−0.004
		(0.015)		(0.012)	(0.007)
Population (100,000)		−0.012		−0.014*	−0.004
		(0.008)		(0.008)	(0.005)
Percent college or more		0.001		−0.005	0.000
		(0.005)		(0.010)	(0.003)
Percent food insecure		0.004		0.001	−0.002
		(0.022)		(0.019)	(0.010)
Percent poverty		−0.012		−0.014	−0.001
		(0.015)		(0.016)	(0.008)
Employment share change		0.021		0.026	0.004
		(0.031)		(0.031)	(0.010)
Local political orientation
Percent R 2020 presidential vote share			−0.004**	−0.008	−0.003*
			(0.002)	(0.008)	(0.002)
State dummies	No	No	No	No	Yes
Observations	647	647	647	647	523

Notes: Coefficients represent average marginal effects (AME). Standard errors clustered at the state level and shown in parentheses.

* p < .10, ** p < .05, *** p < .01.

Table A2.

Estimates of the Effect of SNAP Interview Waiver on SNAP Caseloads, Two-Way Fixed Effects Model.

	(1)
Waiv X Post	478.4**
	(207.58)
County FE	Yes
Month-year FE	Yes
N	6,020

Note. Standard errors clustered at the state level and shown in parentheses.

* p < .10. ** p < .05. *** p < .01. Post period is defined as Jan–May 2021 (pre period is Jan–May 2019).