Summary

  • The CDC’s National Center for Health Statistics documented a steep income gradient in US adult produce consumption between August 2021 and August 2023, with 56 percent of adults below 130 percent of the federal poverty level eating any fruit on a given day versus 76 percent at 350 percent of FPL.
  • The NCHS brief predates the SNAP provisions of the legislation the substrate labels the “Big Beautiful Bill,” whose enrollment effects a separately attributed Associated Press analysis places at a 13 percent national decline between May 2025 and May 2026.
  • The USDA’s September 2025 termination of its Food Security Supplement sponsorship ended three decades of annual federal household-hunger data collection, removing the principal measurement instrument for tracking post-policy food-insecurity consequences just as those policy changes take effect.
  • Four mechanism hypotheses — direct affordability, policy amplification, time poverty, and structural baseline — remain discriminable by varying degrees in the substrate, with the Arizona 47 percent enrollee reduction the load-bearing subnational evidence for administrative-barrier attrition over earnings-driven exit.

A data brief released by the CDC’s National Center for Health Statistics documents a relationship between household income and US adults’ consumption of fresh produce that the brief itself describes as “deep divides between the low-income and middle class.” The brief draws on data collected between August 2021 and August 2023 from adults older than 20 and finds that 56 percent of adults living below 130 percent of the federal poverty level — $20,748 for a single adult, per the brief — ate any fruit on a given day, compared with 76 percent of adults at 350 percent of FPL, or $55,860. The brief reports that more than nine in 10 Americans ate any vegetable on a given day and about seven in 10 ate any fruit. The gradient widens for berries, citrus, and melons — 19.6 percent of the lowest-income adults ate those foods on a given day versus 38 percent of those at 350 percent of FPL, “nearly twice the rate.” For dark green vegetables, 29 percent of the lowest-income adults reported consumption on a given day, compared with 42 percent at the highest income level in the brief. The same gradient held for red and orange vegetables and other vegetable categories.

What the brief documents and what it does not

The NCHS brief reports the gradient without discriminating among the mechanisms producing it. The substrate supplies four candidate explanations the brief does not directly test.

Direct affordability. The substrate reports a University of Washington study finding that foods aligned with the dietary recommendations issued under Health Secretary Robert F. Kennedy Jr. cost a mean of $15.82 per day; the average SNAP household consists of two people and receives $177 per month, a figure the substrate characterizes as “about $3 per day per person.” The substrate also reports that a “food pyramid” released in January as part of US dietary guidelines changes led by Kennedy “emphasized animal protein, fresh fruit, and vegetables, all of which cost more, according to studies.” Priya Fielding-Singh, director of policy and programs at the George Washington University Global Food Institute, is quoted in the substrate: “Particularly for people with lower incomes this is really infeasible, especially when millions of people — and what will be probably millions of children — are going to lose access to a really critical benefit that keeps people food secure.” The implied gap between recommended-diet cost and SNAP benefit is the type of evidence more probable under an affordability reading than under explanations that do not invoke price.

Policy environment amplification. The substrate describes “Republican cuts to food assistance programs in the ‘Big Beautiful Bill’” — a colloquial the substrate supplies without identifying the legislation’s formal name. The substrate reports that the legislation required states for the first time to share the cost of SNAP benefits, barred certain legal immigrants including asylum seekers and certain trafficking survivors from food assistance, and raised the age at which older adults must document 20 hours of weekly work from 54 to 64. The Center for Budget and Policy Priorities, cited in the substrate, reported that the state cost-share requirement “could result in hundreds of millions of dollars of new costs in some states.” The NCHS brief predates these provisions; the hypothesis predicts gradient widening in the post-brief window — a prediction the substrate cannot measure.

Time poverty and labor instability. The substrate carries no direct measurement of time scarcity in the brief. Pamela Lachman, chief strategy officer for the Center for Employment Opportunities, which the substrate identifies as working with people recently released from jail and prison, is quoted: “They have requirements to do their parole supervision, they’re trying to find a place to live — all of those things mean your work might be inconsistent. If you immediately need to start proving you’re working 20 hours a week, if not more, that’s very challenging to do.” Lachman’s account is the substrate’s evidence for a conditional structure in which low-wage employment is irregular in ways that work-requirement documentation does not accommodate.

Structural baseline. The substrate cites a 1935 UK study of diet and income finding that “nearly everyone ate potatoes and bread, while the poorest bought far less milk, eggs, vegetables, and fruit.” Sir John Boyd Orr, writing in that period, is quoted in the substrate: “A review of the state of health of the people of the different groups suggests that, as income increases, disease and death-rate decrease, children grow more quickly, adult stature is greater and general health and physique improve.” The same historical evidence weakens the “contemporary US prices alone” reading and strengthens the structural-baseline reading; it updates both in opposite directions.

The enrollment evidence and what distinguishes its readings

A 13 percent national SNAP enrollment decline between May 2025 and May 2026 is reported in the substrate as drawn from “an Associated Press analysis”; the substrate does not name the methodology, the specific report, or its publication date. Two readings compete for that decline.

The first, framed in the substrate as offered by “proponents of welfare cuts,” attributes the decline to higher earnings lifting households above eligibility thresholds. The second attributes it to administrative barriers introduced by the Big Beautiful Bill provisions. The conditional dependency between the two is consequential. Under an administrative-burden reading, even households whose earnings remain below eligibility thresholds can exit the rolls when documentation requirements exceed the administrative capacity of either applicants or state agencies.

The substrate identifies four jurisdictions — Florida, Georgia, Louisiana, and Nevada — as having steeper-than-national declines without providing percentages. The substrate’s strongest available quantitative subnational evidence for the administrative-barrier reading is Arizona, where “new documentation requirements paired with severe staff shortages” are attributed by the substrate to unnamed “analysts” as contributing to a 47 percent reduction in enrollees. The substrate names no institution or report for that figure. The Arizona case is the load-bearing observation distinguishing the two explanations; if removed from the evidence set, the administrative-barrier reading weakens substantially and the earnings-recovery reading gains relative weight.

The likelihood ratio between the two readings on the available evidence is moderate. A national enrollment decline of that magnitude occurring within roughly a year, concentrated in jurisdictions described in the substrate in documentation-burden and state-cost-share terms, is more probable under a policy-amplification reading than under a baseline-gradient reading. The earnings-recovery reading generates the same surface observation through a different mechanism; the Arizona sub-case is the specific evidence that moves the posterior toward policy-driven attrition.

Measurement infrastructure withdrawn

In September 2025, the substrate reports, the US Department of Agriculture announced it would no longer sponsor the Food Security Supplement in the US Census Bureau’s annual survey, ending data collection that had occurred annually since 1995. The timing places measurement of food-insecurity consequences beyond direct observation as the Big Beautiful Bill’s SNAP provisions take effect. About 39 percent of SNAP beneficiaries are children, per the substrate. The supplement’s suspension has an asymmetric effect on the hypothesis space: under the policy-amplification reading, that infrastructure would be most needed precisely when it has been removed — an observation the substrate supports but does not itself evaluate.

What the substrate supports and what it does not

The most defensible reading the substrate supports is that direct affordability is well-supported as the within-brief mechanism — where “within-brief” refers to the August 2021–August 2023 NCHS observation window; that the structural-baseline hypothesis supplies the historical prior explaining why the gradient was a documented phenomenon in 1935; that policy amplification is moderately supported by the parallel enrollment evidence from the May 2025–May 2026 AP enrollment analysis and predicts post-brief widening the substrate cannot measure; and that time poverty is supported by qualitative practitioner testimony but is not directly measurable from the brief.

The substrate supplies several figures it does not independently verify. The 13 percent national figure, the Arizona 47 percent figure, and the $15.82 per day figure are sourced in the substrate to parties the substrate does not name beyond “an Associated Press analysis,” “analysts,” and “A University of Washington study.” The arithmetic of “about $3 per day per person” is internally consistent with the substrate’s $177 per month for a two-person household ($177 divided by 30 divided by 2 equals approximately $2.95 per person per day). The Boyd Orr quotation is reproduced by the substrate; the 1935 study itself was not retrieved for independent verification. The phrase “deep divides between the low-income and middle class” is quoted by the substrate from the NCHS brief; the brief itself was not retrieved for independent verification. “Big Beautiful Bill” is the substrate’s colloquial for legislation whose formal name the substrate does not supply.

Future claims about the food-security consequences of recent food-assistance policy changes will be made against a prior in which the principal federal measurement instrument no longer exists. The CDC’s August 2021–August 2023 produce-income data is the most recent available federal evidence on the gradient itself; the AP enrollment analysis is the most recent available federal-context evidence on participation; and the USDA’s September 2025 withdrawal of the Food Security Supplement ends the run of comparable household-level food-insecurity data that has accumulated annually since 1995. The substrate does not contain a measurement instrument capable of tracing what happens to the gradient between the brief’s observation window and any subsequent federal data collection effort.

Additional considerations

The substrate’s note that a “food pyramid” released in January 2026 emphasized animal protein, fresh fruit, and vegetables — categories the substrate reports as costing more — establishes that the dietary-recommendation environment, like the SNAP-eligibility environment, has shifted toward categories that carry higher per-calorie cost. Fielding-Singh’s quoted observation that the recommended pattern is “really infeasible” for lower-income households, paired with the $15.82 per day University of Washington figure and the approximately $3 per day per person SNAP arithmetic, frames the affordability constraint as operating against the recommended pattern rather than against a cheaper alternative. The substrate carries no direct test of whether households substitute toward the energy-dense packaged foods the substrate describes as filling and shelf-stable, and no direct test of whether the gradient observed in 2021–2023 has steepened in the post-brief window. The principal federal instruments that would have tested both questions have been altered or withdrawn within the period the substrate describes.

Analytical techniques used in this piece

This analysis applies the methods below. Each links to a short, plain-English explainer you can read and reuse.

Bayesian Hypothesis Network
Updates the probabilities of competing hypotheses as evidence accumulates.