How Much Fraud Will Americans Accept to Get Benefits to Eligible People?
How much fraud will Americans accept to make sure eligible people actually receive benefits? Bell and Jilke (2026) put that question to 2,250 U.S. adults in a survey experiment asking them to vote on cutting paperwork in programs such as SNAP and unemployment insurance. The typical respondent would accept about 192 ineligible people receiving benefits so that 1,000 eligible people would no longer be shut out, or roughly one improper recipient for every five people helped. Support fell from about 64% to about 47% as the number of ineligible recipients rose from 1 to 1,000. About one in five respondents opposed cutting paperwork even when it added no fraud at all.
Why This Article Was Selected for The Policy Scientist
Means-tested programs such as SNAP, Medicaid, and unemployment insurance reach tens of millions of Americans, and the procedures governing access to them (documentation, verification, interviews, recertification) often shape who is served as much as eligibility rules do. Those procedures are again central to federal policy: the 2025 budget reconciliation law expanded work requirements in SNAP and Medicaid and required more frequent Medicaid eligibility checks, typically justified as protecting program integrity. Yet these debates seldom state how many eligible people the public will accept excluding to keep ineligible people out. This article quantifies that trade-off, extending Herd and Moynihan's work on administrative burden and recent burden-tolerance research. Both authors have published extensively on administrative burden. The preregistered randomized survey experiment is a strong design for measuring stated preferences, though its opt-in online sample and hypothetical choices limit generalization to actual votes or other countries. It appears in the Journal of Policy Analysis and Management, the flagship journal of the Association for Public Policy Analysis and Management.
Full Citation and Link to Article
Bell, E., & Jilke, S. (2026). Administrative burden and the access–fraud trade-off. Journal of Policy Analysis and Management, 45(4), e70128.
Central Research Question
Policymakers frequently justify documentation requirements, verification procedures, and recertification rules in safety-net programs as defenses against fraud. Those same procedures, however, keep some eligible people from receiving benefits. The study asks how the public weighs these two errors: how many ineligible recipients (false positives) people are willing to accept in exchange for 1,000 eligible people no longer being excluded by paperwork (false negatives). It also asks whether that tolerance depends on how a program is described. Drawing on construal level theory, the authors hypothesize that the abstract label "welfare" evokes stereotyped, decontextualized images of recipients and therefore lowers tolerance for fraud relative to descriptions of specific programs such as SNAP or unemployment insurance.
Previous Literature
The study builds on the administrative burden framework of Moynihan, Herd, and Harvey (2015) and Herd and Moynihan (2018), which divides burdens into learning, compliance, and psychological costs and treats them as political choices embedded in program design. Field experiments show that these costs reduce participation: Finkelstein and Notowidigdo (2019) found that reducing learning and compliance costs raised SNAP enrollment among older adults by 5 and 7 percentage points, respectively, and Giannella et al. (2024) found that flexible interview scheduling raised SNAP approvals by 6 percentage points. Work on welfare attitudes, including Gilens (2000) and Hancock (2004), documents how racialized "welfare queen" narratives shape public opinion, while Schneider and Ingram (1993) show how social constructions of target populations affect policy design. Most directly, a growing literature on burden tolerance (Halling et al., 2022; Bækgaard et al., 2024; Keiser and Miller, 2020) finds that conservatives and people without program experience tolerate more burden, but it measures tolerance with binary or ordinal scales rather than an explicit exchange rate between access and fraud.
Data
The authors fielded a nationally representative online survey through Prolific on October 11, 2024, recruiting 2,250 adults matched to U.S. Census benchmarks on age, gender, race and ethnicity, and political affiliation. About 70% of respondents were White, 48% were female, 38% identified as Republican, and 48% reported that they or a family member had used a welfare program. A pretest of 490 respondents set the ranges for the experimental values, the analysis was preregistered, and 97.5% of respondents correctly identified their assigned program. Respondents were randomly assigned to one of five program descriptions: general "welfare," TANF, SNAP, SSI, or unemployment insurance, each with roughly 450 respondents.
Methods
The design is a contingent valuation experiment presented as an advisory referendum. Respondents read that an agency was considering reducing identity and income documentation, which would allow 1,000 additional qualified people to receive benefits. They first voted on the change when it produced no additional ineligible recipients. They then voted again after being told it would also admit a randomly drawn number of ineligible recipients, ranging uniformly from 1 to 2,000. A second exercise asked whether respondents would pay a randomly assigned annual tax increase of $1 to $600 for five years to fund their preferred option. Because the trade-off values were randomized, the decline in support as the number of false positives rises can be read as a causal effect on stated preferences. The authors estimate logit models of vote choice on the logged number of false positives and compute cost-adjusted median willingness to accept by simulation, checking robustness to certainty restrictions, covariates, and removing the cost adjustment. The main weaknesses are that choices are hypothetical, the sample is an opt-in online panel, the "welfare" label varies stereotype salience and social construction as well as abstraction, and the partisan, burden-tolerance, and welfare-experience analyses are exploratory and not corrected for multiple comparisons.
Findings/Size Effects
The central estimate is that respondents would accept about 192 ineligible recipients (191.85) in exchange for 1,000 eligible people gaining access, a ratio of roughly 1 to 5. The logit coefficient on the logged number of false positives is −0.49 (p < 0.0001). In a linear probability model, each additional 500 false positives reduced support for cutting paperwork by 8.8 percentage points, so support fell from 64.1% with one false positive to 46.5% with 1,000. Once the ratio of ineligible to newly eligible recipients reached one to one, a majority opposed the change. About 20% of respondents rejected reducing documentation even when it produced no fraud at all.
Tolerance varied widely by program. Median willingness to accept was lowest for SNAP (97 ineligible recipients per 1,000) and the general "welfare" description (126), followed by SSI (151) and TANF (162), and highest for unemployment insurance (357). The hypothesis that the "welfare" label would yield the lowest tolerance was therefore not supported in the full sample. In the tax exercise, respondents who favored reducing paperwork would pay a median of $55.53 a year (SNAP) to $594.60 (SSI), while status-quo supporters would pay only $7.95 (SNAP) to $82.46 (unemployment insurance). At a $1 tax, 87.6% of paperwork reducers kept their position, compared with 58.3% of status-quo supporters.
The exploratory subgroup results are where the framing hypothesis gains support. Republicans would accept only about 3 ineligible recipients per 1,000 under the "welfare" label, compared with about 107 for unemployment insurance and 16 to 49 for the other specific programs. Democrats and Independents would accept 289 under "welfare" and as many as 914 for unemployment insurance. Formal interaction tests showed a statistically significant partisan difference only in the "welfare" condition. Baseline support with one false positive was 47% among Republicans and 75.4% among Democrats and Independents, a gap of 28.4 percentage points that narrowed as false positives increased. Respondents with welfare experience showed higher baseline support (68.6% versus 60.5%) and a similar, significant difference only under the "welfare" label. Even respondents with high burden tolerance supported reducing paperwork at 57.7% when it produced one false positive, versus 72.5% for those with low tolerance. The main results held across robustness checks.
Conclusion
The study offers the first quantitative estimate of how the public weighs access against fraud in U.S. safety-net programs: the median respondent would accept about one ineligible recipient for every five eligible people helped, and majority support disappears when the ratio reaches one to one. Tolerance is lowest for SNAP and highest for unemployment insurance, and the abstract "welfare" label sharply reduces tolerance among Republicans and among people without program experience, although it has no detectable effect in the sample as a whole. These are stated preferences from a single online survey conducted before the 2024 election, so they do not establish how voters or legislators would behave when facing actual program rules, and the subgroup findings should be treated as hypotheses rather than confirmed effects. The design also cannot isolate whether the "welfare" effect operates through abstraction, stereotypes, or partisan cues. Future research could replicate the experiment in other countries and periods, measure psychological distance directly, test burden-increasing as well as burden-reducing frames, and compare the public's preferred ratio with programs' actual error rates.
Comments