Social Safety Net
Paper Session
Saturday, Jan. 3, 2026 8:00 AM - 10:00 AM (EST)
- Chair: Brendan Cushing-Daniels, Gettysburg College
From Storefronts to Screens: The Impacts of Online Grocery Shopping on Public Food Assistance Users
Abstract
Many anti-poverty programs are in-kind, and adoption of new technology can alleviate the challenges associated with redeeming benefits. This project investigates how the availability of online grocery purchasing in public food assistance programs—including SNAP—affects food access, benefit spending patterns, and program participation. Authorization to accept Electronic Benefit Transfer (EBT) payments online was disproportionately adopted by large food retailers in urban areas. Exploiting the staggered roll-out of online purchasing authorization across retailers, I estimate that online exposure led to a $16 increase in monthly online EBT spending per household. Households substitute away from in-store spending at large food retailers. Following a large drop in SNAP benefits, participants decrease online grocery spending more than dollar-for-dollar, suggesting that online grocery shopping is a luxury service and EBT consumers are willing to pay for convenience under higher incomes. Finally, online grocery purchasing availability increases local SNAP participation by 4 percent, primarily by increasing retention of existing participants. These results suggest that policies which reduce benefit redemption frictions can improve the effectiveness of in-kind benefit programs.Measurement Error in Survey Incomes Over Four Decades
Abstract
Survey data increasingly miss dollars for major income sources (Meyer et al. 2015), threatening their reliability as a foundation for research and policy. Previous studies have gone beyond comparisons of aggregates to document measurement error at the individual level, tending to focus on single income sources like retirement income (Bee and Rothbaum 2017) or SNAP (Meyer et al. 2022) over a limited set of states and/or years.We provide a comprehensive assessment of how individual-level measurement error has changed over time, using data for twelve income sources spanning nearly four decades. We link the Current Population Survey Annual Social Economic Supplement (CPS) – the official source of poverty statistics – from 1984 to 2022 to administrative records for earnings (IRS Forms 1040 and W-2; SSA’s Detailed Earnings Record), interest and dividends (1040s, 1099-INTs, and 1099-DIVs), retirement income (1099-Rs), AFDC/TANF (HHS and state agencies), Unemployment Insurance (1099-Gs), Social Security (SSA’s PHUS and MBR), Supplemental Security Income (SSA’s SSR), veterans’ disability compensation (DVA), SNAP (state agencies), public and subsidized housing (HUD PIC/TRACS files), Medicaid (CMS), and Medicare (CMS). All datasets except for SNAP and TANF are at the national level.
For each income source and year, we quantify three key dimensions of measurement error: false negatives (recipients not reporting receipt), false positives (non-recipients erroneously reporting receipt), and dollar misreporting among true reporting recipients. We show how each component contributes to total dollar underreporting. We also examine which subgroups experience the largest changes in underreporting.
These estimates have major implications for assessing how income-based measures of poverty, inequality, and program effectiveness have changed over time. They will also form the basis for imputation models to correct for misreporting in survey data, which we plan to share with the research community.
Coverage Error in Surveys: Evidence from Linked Administrative SNAP and Tax Records
Abstract
The ability of surveys to measure the circumstances of policy-relevant subgroups hinges on whether surveys accurately represent their target populations. This paper is one of the first to examine the extent to which poor individuals (proxied for by SNAP recipients in administrative state records) and the U.S. population more broadly from those with the highest to the lowest incomes (based on those appearing in IRS tax records) are adequately covered in major Census surveys. We focus on the American Community Survey (ACS) and the Current Population Survey Annual Social and Economic Supplement (CPS). Our linked microdata approach provides a simple and powerful way of assessing survey coverage, which accounts not only for unit non-response but also the understudied issues of frame and weighting error.Focusing first on the bottom of the distribution, we find that 93.1% and 94.7% of SNAP recipients are represented in the ACS and CPS, respectively. However, coverage rates differ notably across characteristics. Young children, Black individuals, those with irregular SNAP receipt, and those with less connection to others or work tend to be under-covered among SNAP recipients. Conversely, spouses in multiple-person cases and the elderly tend to be over-covered.
For the broader distribution, we plan to examine how well these surveys capture the tax data population at different income percentiles, using IRS Form 1040 for tax filers and third-party information returns for non-filers. While prior aggregate comparisons showed that 99.8% of the Decennial Census population appeared on a tax or information return in 2010 (Larrimore et al. 2019), our linked microdata analysis can assess the coverage of specific subgroups in surveys.
Our estimates seek to provide key lessons for researchers using household surveys to assess a wide range of topics, ranging from measuring poverty and inequality to evaluating the effects of government transfers.
Long-Term Effects of Universal Free School Meal Policies: Evidence from the Community Eligibility Provision
Abstract
This paper evaluates the short- and long-term effects of the Community Eligibility Provision (CEP), a federal program offering universal school meals, on student academic, behavioral, and economic outcomes. I employ a difference-in-differences research design based on the staggered adoption of CEP in 3,000 Texas schools between 2011 and 2022. To address the limitations of prior research, which has focused on a narrow set of short-term outcomes, I use linked administrative data from Texas that tracks K-12 students through college and into the workforce. The primary outcomes of interest include academic (test scores, ACT/SAT scores, high school graduation, college enrollment), behavioral (meal participation, attendance, suspensions, dropout rates), and economic (employment, earnings) outcomes. Findings from this project will shed light on whether universal free school meals improve student outcomes compared to existing means-tested meal policies.Disentangling Barriers to Welfare Program Participation with a Semiparametric Approach
Abstract
This paper investigates why eligible households do not participate in welfare programs and identifies effective policies to improve takeup rates. Unlike prior studies that use a one-stage approach, we model welfare takeup as a two-stage process: households must first pay attention to the program before deciding whether it is worthwhile to participate. By separating the attention problem from the utility maximization problem, we find substantial subpopulation heterogeneity which offers policy targeting insights.We develop a discrete choice model where fully attentive (FA) households, i.e., those already enrolled last period, and stochastically attentive (SA) households, i.e., those not enrolled in the last period, are analyzed separately. Using FA data, we identify utility parameters through variations in benefit entitlements, recertification costs, and program accessibility. These parameters allow us to impute counterfactual takeup probabilities under full attention for SA households and nonparametrically identify their attention probabilities. We propose a corresponding semiparametric estimation procedure and derive asymptotic normality for the estimator under the condition that FA and SA sample sizes are asymptotically equivalent.
We apply our model to the Special Supplemental Nutrition Program for Women, Infants, and Children (WIC). Empirical findings highlight two key results: (i) households with infants have higher attention probabilities (3.8\% - 7.6\%), making informational campaigns an effective policy for households without infants (their attention probability ranges from 0.7\% to 2.2\%), and (ii) households from high-benefit states, with a two-year college degree or above, without infants, are less attentive but more likely to participate once becoming attentive. This suggests that partnering with parenting student groups at higher education institutions during informational campaigns is an effective strategy for promoting welfare takeup.
JEL Classifications
- H5 - National Government Expenditures and Related Policies