Lawmakers invested trillions of dollars in economic recovery efforts during the COVID-19 pandemic, with the goal of providing support to families who lost their jobs or wages as well as those who were least able to buttress the impacts of the pandemic. These pandemic impacts extend beyond COVID-19 mortality risk to an increased likelihood of job loss, housing eviction, household debt, and food insecurity (e.g., Despard et al., 2020).

There were important shifts in access to safety net supports during this period, including multiple rounds of economic stimulus payments, enhanced unemployment benefits, an expanded child tax credit, and Supplemental Nutrition Assistance Program (SNAP) emergency allotments (EA). The introduction of these pandemic policy responses had an important impact on poverty and food hardship (Bitler et al., 2020; Parolin & Wimer, 2020). However, there has been little opportunity to examine how the range and timing of federal programs implemented or curtailed during this period affected multiple dimensions of consumer shopping behaviors, including grocery spending and payment used.

In this paper, we examine the impact of removing COVID-era SNAP EAs on household grocery purchasing behavior and mode of purchasing among SNAP participants. We use Circana Consumer Network data, which contains food-at-home purchasing information for a large sample of respondents, enabling us to explore food acquisition behavior. It also includes demographic information, enabling us to measure heterogeneity in policy impact.

We address two research questions: (1) To what extent did SNAP households' food spending shift (in terms of total grocery expenditures, payment mechanisms used, and share of grocery purchases on different food groups) following the removal of SNAP EAs? and (2) did these policies have differential impacts on racial or ethnic groups or by family composition?

We find, for both the entire sample and for specific demographic subgroups, that households participating in SNAP reduced their grocery purchases after SNAP EAs were removed. On average, total aggregate monthly grocery spending fell by nearly $13, or 4.6%. We also find evidence of changes in how consumers paid for groceries. The share of grocery spending using SNAP benefits declined by 23.4%, while credit card usage increased by 24.1%, and cash equivalents increased by 15.0%.

RELEVANT LITERATURE

Two streams of research are especially important for this study: (1) literature on how safety net programs changed during the pandemic, how individuals responded to changes in program generosity, and the implications for their well-being; and (2) literature on how people's purchasing behaviors changed during the pandemic, with a focus on substitution behaviors, payment mechanisms used, and household consumption decisions. This study contributes to both of these literatures by examining how the removal of time-bound safety net supports can shape household food purchasing and financial decisions.

DATA

In this section, we describe the source data, analytical sample, and outcomes. We briefly discuss data limitations, but a fuller discussion is available in Muth et al. (2016).

ANALYTIC APPROACH

To identify the impact of removing SNAP EAs on consumer spending and purchasing behavior, we use the methodology outlined in Borusyak et al. (2024). This method is an imputation approach that fits a regression using not-yet-treated outcomes to impute a counterfactual outcome for each treated unit. These individual treatment effects are aggregated to an overall treatment effect. In our analysis, removal of SNAP EAs is considered the treatment.

We select this model because it generates an efficient and robust estimator that calculates individual treatment effects that can be aggregated, while other heterogeneity-robust staggered difference-in-difference estimators may require balanced panels or other assumptions (see Baker et al., 2022 for a summary). Borusyak et al. (2024) rely on several key assumptions for robust and efficient estimation: (1) Parallel trends assumption on potential outcomes for pretreatment periods, (2) no anticipation effects, and (3) restricted causal effects (Baker et al., 2022).

Following Borusyak et al. (2024), the unit (here, state) and period fixed effects are fitted by regressions using only the untreated observations (we assume that once states are treated, they remain treated). Then, these fixed effects are used to impute the untreated potential outcomes to obtain the estimated individual treatment effect for each treated unit. The individual treatment effects are then summed using panel weights provided by Circana designed to align the sample with representative estimates for the contiguous US (excluding Hawai'i and Alaska). Standard errors are clustered at the state level, coefficients represent average treatment effects (ATE) for all post-treatment periods, and we include individual and state controls (as detailed above). A full discussion of methodology is available in supplementary material.

For our estimates, we focus on the population of consumers who paid for groceries with SNAP EBT at any time between when SNAP EAs were implemented (March 2020) and when the first state removed SNAP EAs (March 2021). This population identifies the consumers who are likely to see a change in the resources available for food after SNAP EA removal.

For the subgroup analysis of heterogeneous policy effects by race and ethnicity of the household head and among families with children, we run each specification above for each outcome of interest on a separate subsample of SNAP households. We report results for multiple subgroups that have sufficient numbers of consumers to report effects: (1) Non-Hispanic White-headed households, (2) non-Hispanic Black-headed households, (3) Hispanic-headed households, (4) households with a child under 6 years old, (5) households with a child between ages 6 and 17 years old, and (6) households without children. We explore these groups separately in an effort to understand more about how groups at higher risk of experiencing food insecurity respond to policy shifts. Prior research has documented how Black and Hispanic families and families with children have consistently higher rates of food insecurity than White peers and peers without children, and often have fewer resources (e.g., savings, credit liquidity, etc.) at their disposal (Odoms-Young & Bruce, 2018). These subgroup samples are not mutually exclusive, where if a household is Hispanic-headed and has a child under 6 years old, they would be represented in each subsample analysis.

We are also interested in articulating the dynamic effects of SNAP EA removal on SNAP households. This analysis explores how treatment effects change over time, as households may respond differently to the loss of additional EA benefits 1 month after removal than 6 months after. It also enables us to estimate pretreatment trends to assess the validity of the parallel trends assumption. To do this, we use Borusyak et al.'s (2024) event study methodology and disaggregate treatment effects across 6 months following EA removal.

RESULTS

We present results from our preferred specification for the entire sample and six subgroups as well as dynamic effects for the entire sample several months before and after policy removal (additional estimates can be found in the Tables A1–A9). We report the average monthly dollar change in total grocery spending, the average percentage point change in the share of grocery purchases made with SNAP EBT cards, credit cards, or cash and cash equivalents, and the average percentage point change in the share of grocery purchases made in each food group. We also report the average of each variable in March 2021, which is the last month SNAP EAs were active in all states, and a scaled coefficient estimate relative to the baseline mean (in March 2021) to get a sense of the scale of policy impact on grocery purchasing behavior (see Table 2).

All treatment effect results are interpretable for the group of consumers who paid for groceries with SNAP EBT at any time in the months between when SNAP EAs were implemented (March 2020) and when the first state removed SNAP EAs (March 2021) and who are living in states that removed SNAP EAs before the end of 2022. This population is the “treated” population, or those consumers who are likely to see a change in the resources available for food purchasing after the removal of SNAP EAs (in contrast with consumers who are not receiving such benefits).

We find that the decline in SNAP resources after EAs were removed is partially accounted for by (1) declines in total aggregate monthly grocery expenditures and (2) shifts to other payment mechanisms. Consistent with the decrease in SNAP benefits available to households after EAs were removed, the share of grocery spending paid for using SNAP EBT benefits declined by an estimated 10.2 percentage points (Table 2). We find that SNAP participants decreased their total aggregate monthly spending on groceries by $12.52 on average in those states where EAs were removed. In addition, the share of monthly grocery spending using other payment sources increased: The use of credit cards to pay for groceries increased by 3.0 percentage points, while cash payments increased by 6.4 percentage points.

We put these coefficients into context by calculating the percent change these coefficients represent relative to the “baseline” mean of each variable (defined as the mean in March 2021).¹¹ Here, the results suggest that changes in SNAP EBT spending are both statistically and nominally significant, as expected: the total share of spending using SNAP EBT benefits declined by 23.4%, which is largely offset by shifts to using credit cards (an increase of 24.1% relative to March 2021) and cash (an increase of 15.0% relative to March 2021) to pay for groceries. Changes in total grocery spending are also meaningful, representing a 4.6% decline in monthly grocery spending.

While we also examine changes in the share of grocery dollars spent by SNAP households on vegetables, fruit, meat and protein, dairy, grains, and prepared foods, we see either nonsignificant results (at the p < 0.05 level) or small-to-moderate effects (in terms of magnitude). We find evidence that after the removal of SNAP EAs, SNAP households did not shift their overall fruit or meat and protein purchasing (Table 2). While we find statistically significant increases in vegetable, dairy, and grain purchases, these changes are small-to-moderate in magnitude (0.2, 0.3, and 0.4 percentage points respectively, or a relative percentage increase of 5.0%, 3.5%, and 7.0%). We also find some evidence of decreases in prepared meals, but these are small in magnitude (−0.5 percentage points, or a 3.3% relative decline).

These results suggest that SNAP participants are responding to changes in SNAP benefits sufficiency by shifting to paying for groceries with cash or credit and by curbing their total spending. Interestingly, consumers do not decrease their total aggregate grocery spending by the same amount as the SNAP benefits that they lost, suggesting that consumers may not have substantially fungible food needs even when they lose benefits. This supports the findings found elsewhere (e.g., Hoynes et al., 2015) that show that SNAP participants are inframarginal and use SNAP benefits in similar ways as other forms of payment. We also find some preliminary evidence that SNAP households may have marginally changed their overall purchasing (in dollar terms) across key food categories—with evidence of slight reductions in prepared foods and greater relative spending on vegetables, grains, and dairy products. These findings align with other research demonstrating that when faced with a lower budget constraint, consumers change the mix and types of food they purchase (French et al., 2019). However, it is important to note that food group outcomes are measured in dollar terms, which do not capture changes in the volume purchased or the nutritional quality of selected items; switches to cheaper alternatives within these food categories; or other consumption changes. As such, we do not seek to over-interpret these results, acknowledging these key limitations in data availability. While the Circana data include the amount of food purchased (e.g., weight, count), we do not explore these outcomes in this paper. Further research could do so, which could lead to insights into whether and how SNAP recipients change the mix or amount of food types in response to policy changes.

DYNAMIC EFFECTS ANALYSIS

Table 3 disaggregates results across the 6 months preceding and following removal of SNAP EAs. Looking at pre-trends, changes in the share of grocery purchases paid for with SNAP EBT and cash as well as the share of grocery purchases spent on meat/protein, vegetables, dairy, and grains are not statistically significant at the p < 0.05 level in any pre-period. For the share of grocery purchases paid for with credit cards and total aggregate grocery spending, there is one period out of six that is statistically significant at the 0.05 level. This suggests that there is little difference in consumer purchasing behaviors between SNAP participants living in states that removed SNAP EAs before the end of 2022 relative to those living in states that did not do so.

We find that over the 6 months following SNAP EA removal, declines in total aggregate grocery spending are sustained and statistically significant at the p < 0.05 level for all periods. Similarly, we see that shifts in grocery purchasing behavior are sustained—SNAP households decrease the share of grocery purchases paid for with SNAP EBT and increase the share paid for with cash and credit cards in all 6 months after allotments are removed. There is some evidence that the share of grocery purchases paid for with credit cards increases over the 6 months following policy removal. For example, 1 month after removal, the share of grocery purchases put on credit cards increases 1.0 percentage point, but 6 months post-policy removal, this increases by 3.2 percentage points.

These results suggest that the changes SNAP beneficiaries make after SNAP EAs are removed may not be temporary but instead reflect how families are trying to establish strategies that consistently meet their food needs under changing levels of benefit generosity. It is possible that slight increases in the share of grocery purchases paid for with credit cards over the 6 months following EA removal signal changes in households' ability to continue affording food with existing resources—but further investigation is needed to confirm this hypothesis.

SUBGROUP ANALYSIS

The results in Table 4 show policy impact estimates for six subgroups of SNAP participants. We may expect that groups with higher rates of food insecurity may be less able to shift spending to other payment mechanisms and may instead respond to reductions in benefit generosity by cutting spending more sharply or by shifting their food purchasing patterns to a greater degree.

We find similar patterns in declines of total monthly grocery expenditures after SNAP EAs are removed for all subgroups. SNAP households headed by Black or White individuals decreased grocery expenses an average of 8.0% and 7.1% relative to baseline. Similarly, we find that households with children cut their total grocery spending by an estimated 6.5% for households with children under age 6 years, and 10.6% for households with children between ages 6 and 17 years (relative to 2.7% for families without children).

We also find similar patterns in payment behaviors, with shifts away from SNAP EBT and toward cash and credit options, although the magnitude of such shifts varies by subgroup. For example, we see greater relative percent increases in shifts to paying with credit cards for White- and Hispanic-headed households (23.6% and 29.7%, respectively) relative to Black-headed households (9.7%). These shifts toward paying for groceries with credit cards are qualitatively meaningful and reflect a large increase in credit card spending post-SNAP-EA removal. We see smaller relative percentage increases in the share of grocery expenditures paid for with credit among households with children (3.2% for households with children under age 6 years and 4.5% for households with children between ages 6 and 17 years). Childless families see the starkest increases in the share of grocery purchases paid for with credit cards post-EA removal (39.9%).

DISCUSSION

We find compelling evidence that SNAP participants responded to the removal of EAs by reducing their grocery expenditures and purchasing a greater share of their groceries using credit or cash. They may have also made small changes to the types of food purchased—increasing dairy, grains, and vegetables while reducing prepared food purchasing. These trends are broadly consistent across different household types, including race and the presence of children, although the magnitude of these responses varies.

Our findings that SNAP households are increasingly paying for grocery purchases using credit cards after the removal of SNAP EAs are important and, in connection with other data, could point to the implications of this policy on material hardship. Data from the Urban Institute's 2023 Well-Being and Basic Needs Survey find that 22.7% of SNAP participants reported paying for groceries with a credit card and not paying the full balance or making the minimum payment (Gonzalez & Karpman, 2025). These consumers could incur interest and penalty fees on carried balances, which are more expensive due to increases in interest rates (Aladangady et al., 2023). It is possible that increased credit card spending associated with the removal of SNAP EAs means that nearly 1 in 4 SNAP households using credit cards to pay for groceries could be incurring interest on these purchases. These households could struggle to repay these increased debt burdens, potentially undermining their long-term creditworthiness. More investigation of the overall economic well-being implications of shifts in grocery spending will be integral to detailing the impact felt across household budgets as a result of changes in SNAP benefit generosity.

Additionally, our finding that SNAP households decrease their overall grocery spending in response to SNAP EA removals could have implications for household health. While we only look at expenditures in this study and find modest declines in dollars paid for groceries, these results need to be interpreted in the context of rising grocery costs over the period of study. During our sample period, food-at-home costs increased by 23% (US Bureau of Labor Statistics, 2024). With low-income families spending 32.6% (an average of $5278) of their income on food (USDA, 2024), it would be challenging for these households to absorb food price increases without experiencing additional hardship. Indeed, a body of literature articulates the connection between food price increases and local rates of food insecurity (Waxman et al., 2022).

Food price increases may mean that it is increasingly difficult for families to curb grocery spending without reducing nutritional quality, diet variability, and the amount of food consumed. Typically, families will use a mix of these strategies to align spending with a lower budget constraint. All of these coping strategies could have important effects on health, where less nutritionally dense diets could make it harder to manage chronic diet-sensitive diseases, and reduced diet quality and amount could contribute to increased risk of chronic disease (Gundersen & Ziliak, 2015). Although this is outside the scope of the current study, it would be fruitful to explore these dynamics in future work to explore the effect of such policies on the nutritional security of families.

CONCLUSION AND POLICY IMPLICATIONS

Evidence on how changes in public benefits generosity impact household purchasing behavior is integral to helping policymakers design effective programs that support families' nutritional security and well-being. We find compelling evidence that reductions in SNAP benefit generosity in 2021 and 2022 resulted in families reducing their monthly grocery expenditures by an average of 4.6% and increasing the use of credit cards to pay for groceries by 24.1%. If increased credit card spending on groceries induced debt repayment challenges for families, this could indicate growing levels of financial distress. Similarly, if households saw not just declines in grocery expenditures but also declines in the amount of food consumed or diet quality, this could result in declines in nutrition and health outcomes important to policymakers.

This study is limited in several ways, including (1) not having data on all states, (2) a lack of available consumer data for the full period SNAP EAs were active, (3) limitations in the types of food purchases captured in the data, and (4) data limitations in being able to calculate standardized measures of diet quality as a core outcome.

The contribution of this study lies in using granular, consumer-level purchasing data on real SNAP participants to articulate how different households responded to changes in SNAP benefit generosity. While this study's findings provide evidence on how consumer spending shifted in response to policy changes, additional research is needed to fully flesh out the financial and health impacts of changes in SNAP benefit generosity on households.