Since the CPS Food Security Supplement ended, that sentence is everywhere: state agencies, food banks, funders. It deserves an honest ledger. I build these models myself. Advantages first.

A model can produce a number for every place. The flagship survey gave most states one number; a model fills the map, county by county, and in finer detail. It makes small groups visible where direct surveys leave blank cells. It is cheap, fast, hard to cancel: built from data already collected, it does not vanish when a federal survey does.

Now, the other side, and one picture carries most of it.

In 2014, USDA-ERS published a model that predicts national food insecurity from three numbers: unemployment, inflation, and the relative price of food. Over 2001–2012, it was excellent — R-squared 0.94, average miss a third of a point. As good as this kind of model gets.

Freeze its coefficients and run it forward. It drifts a point high before anything dramatic happens. For 2020, it says 18.5% — worse than any year on record. The survey measured 10.5%, the lowest in nearly two decades. Eight points apart. The model knew that unemployment like 2020’s had always meant food insecurity like 2009’s. It could not know the rules had changed: expanded benefits and emergency programs broke the relationship it had learned. By 2022–2024, the error changes character: the model runs high on its inflation term while actual food insecurity climbs for reasons it cannot see. One equation, wrong two ways inside four years. That is what a structural shift does to frozen parameters.

We only know the model missed because the survey kept score. From 2025 on, no official observed line. The dashed line keeps going, and nothing checks it.

The quieter problems compound. Every number a model produces is a prediction of food insecurity, not a report of it. The prediction leans on the measurement it learned from — already two years stale. A yes/no modeled rate erases the severity tiers in which food insecurity goals are written. Margins widen in small places; look-alike areas get look-alike numbers.

Worst is the assistance trap: feed a model SNAP enrollment or pantry visits, and a benefit cut reads as food insecurity going down.

The honest conclusion is not that models are useless. They are the reach layer, not the anchor: they extend a measurement, they cannot replace one. The government works this way elsewhere: the Census’s small-area poverty estimates and the CDC’s neighborhood health estimates both rely on surveys that still exist. Any state funding a modeled map should fund the direct measure that disciplines it — and demand it in writing: what goes in, what is assumed, how wide the uncertainty runs, how it is validated.

A model without a survey is a forecast without a scoreboard.

Five questions for any vendor of modeled food-insecurity estimates:

(1) What direct measurement is the model calibrated to, from what years, and what is the recalibration and validation plan now that the CPS-FSS is gone?

(2) Will the full input list, data vintages, and model documentation be published, so outsiders can evaluate how it was built?

(3) What does the model assume about relationships staying stable, about which places resemble which, and what kind of shock would break those assumptions?

(4) What are the margins of error for the smallest geography you will publish, and will they be printed next to every estimate?

(5) Will every number be labeled as a modeled prediction, not a measured rate of food insecurity?

A good answer to (1) exists: put the USDA food-security module on a survey that the state already runs — several states field a food-hardship item on the CDC’s BRFSS today — and recalibrate to that.

A vendor with good answers is selling the reach layer. A vendor without them is selling the dashed line.

Still on pause.

The revisions are nearly where they need to be, and the time away was worth it.

One more Friday. The series returns June 12.

Thanks for your patience. See you then.

No myth this week. None next week either.

I’m pausing Food Security Fridays for two Fridays. A few manuscripts are in revision, and the deadlines are real. I also want some unhurried time with my family before the summer fills up. Both deserve full attention. Neither gets it if I’m also publishing every Friday.

So I’m holding the series until June 12.

The work matters. So does the rest.

If you read along here, take the same permission for yourself. Step away from the desk a little early today. Call someone you’ve been meaning to call. The inbox will keep until Monday.

One thing to read while the series is dark: Urban Institute, “Food Insecurity Remained High in 2025, As Safety Net Cuts Loom” — a clear-eyed look at where food insecurity stood heading into this year’s safety-net cuts.

Back June 12 with the next myth.

No myth to dismantle this week.

But two pieces I’ve been working on just went live. One is about the survey this series depends on. One is about the cost of eating well. Here’s why each is worth a look.

Every Food Security Fridays post is built on one survey: the CPS Food Security Supplement. Last September, USDA announced it would end it, calling three decades of measurement “redundant, costly, politicized, and extraneous.”

I spent more than a decade at the USDA’s Economic Research Service producing those statistics. My new essay on Metrics on the Margin takes those four words apart, one at a time.

The survey began under Reagan, not Clinton. It costs about $1.06 million a year, compared with the SNAP program, which spent $99.8 billion in 2024. And the survey’s own data contradict the USDA’s claim that food insecurity has barely moved.

Read it: USDA Ends Food Insecurity Reporting — A Response, on Metrics on the Margin.

The model USDA uses to set the maximum SNAP benefit can do something else, too. A new article in Applied Economic Perspectives and Policy applies it, for the first time, to food-purchase data from U.S. military dining halls.

I coauthored it with colleagues at USDA, the U.S. Department of War, Penn State, Tufts, and Cornell; Linlin Fan led the work. The question behind it is the same as that behind food insecurity itself: what does it cost to eat a nutritious diet within a fixed budget, and whom does a single budget leave short?

The paper identifies a modeled diet that meets military nutrition standards at a cost consistent with the current allowance. It shows why one food budget can’t serve men and women equally. And it documents how far current military dining sits from federal dietary guidance: in 2019, a male soldier’s tray averaged about 189 grams of added sugar a day, nearly four times the recommended limit.

Open access: The Cost of a Nutritious Diet in Cafeterias, in Applied Economic Perspectives and Policy.

Myth-busting resumes next Friday.

8 percent of US adults said their family sometimes or often did not have enough to eat in the past 30 days. Another 26 percent had enough, but not always the kinds of foods they wanted. Roughly 21 million and 69 million US adults, respectively.

Those numbers come from the Federal Reserve Board Survey of Household Economics and Decisionmaking (SHED) 2025, fielded online October 17–28, 2025. The single-item food sufficiency question SHED uses is a separate measurement framework from the 18-item Household Food Security Survey Module (HFSSM). Per the USDA Economic Research Service, food insufficiency is a more severe condition than food insecurity. The 8% falls at the severe end of the food hardship spectrum, corresponding to very low food security severity.

On the surface, food deprivation is unchanged. The 8% rate has held flat across 2023, 2024, and 2025 — even as rent arrears rose 6 pp since 2021 (23% of renters now behind), layoffs ticked up to 7%, and 15% of adults under 30 who weren’t working said they couldn’t find work (+5 pp since 2023). Why food deprivation moved less than rent and labor market measures is the open question this stability poses: SNAP and other food assistance buffering pressure, substitution behavior, measurement-window effects in a 30-day reference period, or genuine resilience in food access.

The stability sits inside a hardship pathway, not outside it. 41% of adults who experienced a major unexpected expense in the prior year reported at least one of three hardships (bills, food, medical), compared with 28% of those without an expense shock. Because SHED measures food sufficiency alongside bills, medical care, rent, and financial well-being in the same wave, the co-occurrence is observable directly: food deprivation tracks expense shocks even when the national rate doesn’t move.

The distributions underneath the flat headline tell a different story. Black adults’ food insufficiency rose 3 pp to 15% (matching the disability subgroup), alongside a 3 pp rise in Black layoffs to 13%. Among adults with $25K–$50K incomes, food insufficiency rose 4 pp to 14%; 66% of adults under $50K called price increases a major concern. Asian adults’ food insufficiency fell 3 pp to 2%, with the highest financial well-being of any group (82% doing okay). The Black-Asian gap widened to 13 pp. The headline is stable. The distributions underneath are not.

The 2025 SHED is a pre-implementation baseline for the OBBBA SNAP changes phasing in across 2025–2028: ABAWD (able-bodied adults without dependents) age-cap expansion is already binding, and state cost-share starts in FY 2028. The 2026 SHED, fielded next October, will be the first federal read after more of those changes take effect. If food assistance has been buffering pressure, that buffer is now being adjusted downward.

No post this week.

Back next Friday with the detailed Week 17 post.

“Food banks will pick up the slack.”

The substitution test is already in the December 2024 CPS Food Security Supplement — the final wave before USDA cancelled the series, and the most recent national data on charitable food access. The comparison is cross-sectional, with both pantry use and SNAP receipt measured over a 12-month reference period. Among food-insecure households who lost SNAP during 2024 (N = 179) — they received SNAP in an earlier month but not in the survey month — 37% used a food pantry. That rate is statistically indistinguishable from the 43% rate for current SNAP recipients (p = 0.21) and significantly above the 25% rate for households who never had SNAP (p < 0.01). If charitable food were filling the gap when SNAP is lost, lost-SNAP households would exceed current recipients’ pantry use. They don’t. Pantries do not scale up to replace SNAP. (The lost-SNAP cell is small, and the 12-month reference period for both pantry use and SNAP receipt means the temporal ordering of pantry use vs SNAP loss is not pinned down by the CPS-FSS; the directional argument requires the lost-SNAP rate to exceed current recipients to evidence scaling, and it doesn’t.)

The One Big Beautiful Bill Act (OBBBA; signed July 4, 2025) cut SNAP by $186 billion over 2025–2034 — $18.6 billion per year (see Week 10 for the SNAP-cuts baseline). For pantries to absorb that cut, the Feeding America network would have to roughly double in size. The annual cut alone (translating to roughly 6–9 billion meals/year at SNAP’s typical per-meal cost) matches FA’s 2024 distribution of approximately 6 billion meals through 200 food banks and 60,000 partner agencies. The broader U.S. charitable food system (including non-FA pantries, Catholic Charities, Salvation Army, and TEFAP commodities) would have to grow by roughly 80–130%.

The CPS Food Security Supplement asks two questions about charitable food. One is regularly reported; the other is almost never analyzed.

Use: Among food-insecure households, 31% used a pantry in the last 12 months. Among those with very low food security — actual hunger — 36% did.

Availability: Among FI households that DON’T use a pantry, 29% say there isn’t one available. Roughly 3.2 million food-insecure households with no pantry available in their community. Severity doesn’t close the gap — the rate is statistically indistinguishable for very low FS non-users. For free meal services, 38% of FI non-users report none available.

The Midwest stands apart. No-pantry rates among FI non-users in the South (33%), Northeast (31%), and West (30%) are statistically indistinguishable from each other (all pairwise p > 0.29). The Midwest is alone at 20% — significantly better access than each of the other three regions (p < 0.01). (Weighted shares of FI households, summing to 100% via largest-remainder rounding: South 43%, West 21%, Midwest 21%, Northeast 15%.) Metro/nonmetro flips the same way: 31% no-pantry in metro, 18% in nonmetro.

Pantry visits are infrequent for most users. Only 35% of users go almost every month. Even among 12-month users, 40% haven’t visited in the last 30 days. SNAP averages $187 per person per month (USDA-ERS, EIB-291, FY 2024) on a fixed schedule; a pantry visit is valued at roughly $40–$60 (Byrne and Just 2023). Multiplying through, only ~11% of all FI households receive charitable food on the monthly cadence SNAP delivers to 100% of its participants.

OBBBA expands ABAWD work requirements to a bucket that already uses pantries below average. About 30% of FI households are working-age adults without children or disability. The 23% already subject (ages 18–54) use pantries at 21% — significantly below the 31% FI average (p < 0.01) and the lowest pantry-use rate of any FI subgroup. The 7% newly subject under the age-cap expansion (55–64) use pantries at 27% — statistically indistinguishable from the FI average. The headline gap is concentrated in the already-subject group; the marginal change OBBBA creates targets a subgroup at average pantry-use levels.

Absorbing the OBBBA cut would require an unprecedented expansion of the charitable food system. SNAP’s role in the safety net is uniquely counter-cyclical. The 2013 ARRA SNAP cliff is the broader historical analog: benefits fell overnight for 47.6 million recipients, and charitable food did not fill the gap.

USDA cancelled the CPS Food Security Supplement three months after OBBBA was signed. The supplement was the only nationally representative instrument that could measure household substitution between SNAP and charitable food at scale. The cuts will land. Whether pantries scale up to meet them will not be measurable at the household level. We turned off the instrument right before we needed it.

This week, the series looks a little different.

No myth to dismantle today. I spent this week heads-down on research — three working papers I’m preparing to release, plus follow-up from a talk I gave to students at the University of Washington.

Here is a quick look at what’s on the desk:

1. A BRFSS crosswalk for state food insecurity monitoring. With the CPS Food Security Supplement cancelled after December 2024, states lose the only federal instrument built for allocation, needs assessment, and disparity monitoring. This paper delivers four conditional probabilities that translate one BRFSS item into comparable state-level food insecurity estimates — a partial, temporary remedy for state-level monitoring, not a replacement for the 18-item CPS-FSS measurement system that distinguished severity, identified child food insecurity, and anchored federal food assistance policy.

2. rasch_estimate: a Stata command for Rasch measurement. The psychometric engine behind nearly every food security analysis I publish. Conditional maximum likelihood, polytomous items, screening blocks, item and person fit, classification reliability — in one command. Built to audit the quality of new food security data collections stepping in after the CPS-FSS, and to calibrate and diagnose any Rasch-scaled instrument beyond food security.

3. Classification accuracy and consistency for IRT-based measures. When we classify a household as food secure, low, or very low food secure — or any category built on a latent-variable estimate — we are classifying a noisy measure. This paper presents a Stata implementation of the Rudner (2001, 2005) methods that quantify how often those classifications are right, and how often they would replicate on a retest — essential for any decision resting on IRT-based measures.

The UW talk

This week I also joined the University of Washington’s Food & Nutrition Security Seminar — a virtual, cross-disciplinary series that convenes students and faculty around current research in food and nutrition security. My session drew 200+ undergraduate and graduate students for a talk followed by an open Q&A. Title: “How Food Security Has Been Measured Over Time and Why It Matters.” I worked through the full arc: how food security is defined as a continuum, how the 18-item HFSSM and Rasch scaling turn yes/no answers into a severity measure, what 30 years of prevalence data reveal, and where measurement goes now that the CPS-FSS has been cancelled.

A recording will be posted to the UW series YouTube channel: youtube.com/@uwfsnh. I’ll share the direct link here as soon as it is up.

Myth-busting resumes next Friday.

— Matthew

The USDA Economic Research Service annual food security report groups 134 million U.S. households into four race/ethnicity categories when reporting food insecurity:

• White, non-Hispanic: 10.1%
• Black, non-Hispanic: 24.4%
• Hispanic (any race): 20.2%
• Other, non-Hispanic: 11.9%

The 2024 CPS — the final Food Security Supplement before USDA cancelled the series in September 2025 — collected 26 detailed race codes and 8 Hispanic-origin codes. The annual report collapses them into these four lines. The 2024 revision to OMB’s Statistical Policy Directive No. 15 requires disaggregation within each category, and the CPS already publishes the detailed recodes.

Every rate is the share of households (not persons) that are food insecure, classified by the race and Hispanic origin of the CPS reference person — the same convention the annual report uses.

ERS disaggregated these groups once (EIB-269) but never in the annual report. Here’s what it hides:

Inside “Other, non-Hispanic food insecurity” (11.9% on paper):

• American Indian/Alaska Native alone: 30.9% [95% CI 24–38%] — highest point estimate of any single-race group
• Asian alone: 6.6% — lowest rate of any single-race non-Hispanic group
• Native Hawaiian/Pacific Islander alone: 18.4%
• Multiracial (non-Hispanic): 21.8%

A 4.7x range inside one row. AIAN households also face the highest very-low-food-security rate of any single-race group (14.5%), and White-AIAN multiracial households show 27.4% FI — the pattern is not a single small-sample artifact.

Inside “Hispanic” (20.2%): Puerto Rican 24.9%, Dominican 24.7%, Central American 23.7%, Salvadoran 22.3%, South American 20.6%, Mexican 18.9%, Other Spanish 18.5%, Cuban 17.3%. A 7.6-point gap within one row.

Inside “Asian alone” (6.6%): Vietnamese 12.5%, Japanese 8.9%, Other Asian 8.6%, Filipino 7.9%, Korean 7.4%, Chinese 5.0%, Asian Indian 3.2%. A nearly 4x range.

SNAP take-up mirrors the gaps — with a policy twist. Among food-insecure households, 52% of AIAN and 51% of Puerto Rican HHs receive SNAP; only 18% of South American, 21% of Central American, and 24% of Salvadoran HHs do. Puerto Ricans are U.S. citizens by birth; many Central and South American households contain non-citizen adults subject to the 5-year bar or mixed-status rules. Higher food insecurity paired with lower program access is consistent with that policy architecture.

What this means:

1. ERS should publish disaggregated race/ethnicity appendices in ERR-358 using the existing CPS recodes. With the CPS-FSS now cancelled, this is the last chance to put these numbers on the federal record.
2. Any successor collection should adopt SPD 15 from day one, with a MENA minimum category — a rate the CPS-FSS never allowed us to see.

Aggregation is never neutral. With the series closed, the categories on the final ERR-358 are the ones that stay in the historical record.

“Food insecurity is a nuclear family problem.”

That’s the assumption behind almost every food security statistic. Married couples. Single mothers. People living alone. These categories appear in every annual report, every data table.

But 1 in 5 U.S. households don’t fit — multigenerational homes, grandparents raising grandchildren, cohabiting couples, adult siblings, extended families, roommates, doubled-up families.

The CPS Food Security Supplement has always collected the data to identify them. But in 30 years of the USDA annual food security report, these categories have never been separately published.

Here’s what we find in the 2024 data:

The households we overlook:

• Grandfamilies (grandparents raising grandchildren, no middle generation): 23.6% food insecure
• Cohabiting couples with children: 23.9% food insecure — more than double the rate of married couples with children
• Extended-kin households: 18.3% food insecure
• Multigenerational (3+ generation): 18.5% food insecure
• Unrelated co-residents (roommates, boarders): 18.3% food insecure

Compare these with the standard categories: Married couples with children: 10.0% food insecure. Married couples, no children: 5.8% food insecure.

The multigenerational question: About 8.6 million U.S. households (7.7M multigenerational + 0.9M grandfamily) span three or more generations. Multigenerational living is one of the fastest-growing household types in the U.S. — roughly 60 million Americans (Pew, 2022). But the USDA annual report has never published their food insecurity rate.

Three-generation households (grandparent + parent + child) are food insecure at 18.5%. But grandfamilies — where the middle generation is absent and grandparents are raising grandchildren alone — face food insecurity at 23.6% (N=268; interpret with caution given the small sample).

The cohabitation gap: Among two-parent households with children, cohabiting families face food insecurity at 2.3x the rate of married families (23.9% vs. 10.4% in a paired comparison). Cohabiting couples differ from married couples on income, education, and access to legal protections and benefit eligibility — factors associated with higher food insecurity risk. SNAP participation also differs: 46.7% of food-insecure cohabiting families receive SNAP vs. 27.1% of married.

Why this matters: SNAP defines “household” by who purchases and prepares food together — not by who lives under the same roof. For a multigenerational family sharing a kitchen, this can determine whether a grandmother’s Social Security income counts against her grandchild’s SNAP eligibility.

One in four of these food-insecure households — 4.6 million — falls outside the standard categories.

Our measurement system captures these households. Our reporting does not. If we only count the households we expect, we miss the ones that need us most.

“Even at full employment, jobs couldn’t fix food insecurity.”

In 2024, the unemployment rate averaged 4.0% — near historic lows. Food insecurity hit 13.7% — 18.3 million households. About two-thirds of them — 12.2 million — had at least one employed adult. Even at near-full employment, a job wasn’t enough.

The labor market added 178,000 jobs in March 2026, yet unemployment stands at 4.3%. If employment couldn’t prevent food insecurity at 4.0% unemployment, a tight labor market alone won’t fix it.

The data from 2024 — captured at near-full employment — reveals a key part of the problem. Since 2019, food-at-home prices and median weekly earnings each rose about 25%. In real food purchasing power, individual workers are flat — earning more dollars but buying the same amount of food.

The timeline shows the disconnect. In 2020, unemployment surged to 8.1% but food insecurity held at 10.5% — emergency benefits cushioned the blow, though pandemic nonresponse may have understated the true rate. By 2022, unemployment fell to a 50-year low of 3.6% while food insecurity jumped to 12.8%. By 2024: 4.0% unemployment, 13.7% food insecurity.

Among households with a full-time worker in 2024, 11.8% are food insecure — and 4.3% have very low food security, reporting disrupted eating patterns and reduced food intake despite full-time work. 78% of working food-insecure households have at least one member working 40+ hours per week. Evidence from separate studies suggests the path isn’t symmetrical — job loss from firm closures cut food spending 15% and reduced food sufficiency immediately (Restrepo et al. 2021), but the path back is slower.

The working food-insecure are structurally different. 63% are single-earner households. They earn more (median income-to-poverty ratio 1.99 vs. 0.91) and have more children (46% vs. 19%). They report spending $160/week on food — roughly $53 per person in a median 3-person household — yet 53.5% say they need to spend more, with a median shortfall of $100/week. And their self-reported SNAP participation rate is less than half that of non-working food-insecure households (24% vs. 57%).

This is the gap the unemployment rate can’t see. Those 12.2 million working food-insecure households aren’t jobless — their wages haven’t kept pace with food costs. But wages are only half the equation. When emergency SNAP benefits ended in March 2023, food insufficiency among recipients increased substantially — even as unemployment held at a 50-year low (Wells et al. 2024). Higher minimum wages are linked to modestly lower food insecurity among working families (Winkler et al. 2025). Without adequate wages and an adequate safety net, the gap remains.

A paycheck is necessary. A safety net is necessary. Neither alone is sufficient.

“People who really need SNAP will still get it.”

That was the promise. Here is the reality, from the last official data we may ever have.

In December 2024, the CPS Food Security Supplement — the federal government’s only nationally representative comprehensive measure of food insecurity — surveyed 12.6 million U.S. households receiving SNAP. It captured who they are and how they were doing.

Half were already food insecure. That is more than 5x the rate among non-SNAP households (49.8% vs. 9.7%).

21.1% experienced very low food security — meaning reduced food intake, skipped meals, and hunger. Not risk. Not anxiety. Actual disruption to eating. Only 30% of SNAP households were fully food secure.

Who SNAP served in 2024:

• 42% of SNAP households had children
• 42% included someone age 60 or older
• 45% included an adult with a disability
• 49% had at least one employed adult

These are not marginal cases. They are the households that food and nutrition assistance was designed to reach — and they were already struggling before anything was cut.

Who loses under H.R. 1:

The One Big Beautiful Bill Act, signed July 4, 2025, cuts approximately $187 billion from SNAP over 10 years. CBPP estimates, based on CBO projections, that roughly 4 million people will lose or have benefits substantially reduced — including 800,000 adults aged 55–64 facing new work requirements, 300,000 parents of children 14 and older, roughly 1 million people in areas where jobs are scarce, and 300,000+ veterans, people experiencing homelessness, and former foster youth.

In the 2024 CPS-FSS, SNAP households with adults 55–64 (no children) were already food insecure at 49%. Parents with only teenagers were food insecure at 55%. These are not households gaming the system. They are households at the boundary of very low food security — and they are the ones being targeted.

The measurement gap:

The December 2024 CPS-FSS may be the last fielding. The administration discontinued the survey before the cuts take full effect. For nearly 30 years, this instrument has been the basis for the annual USDA food security report, congressional testimony, program evaluation, and state-level planning. Without it, there is no nationally representative way to track what happens next.

The 2024 data is the last clean baseline. Food insecurity among SNAP households was already at crisis levels — with the program at full strength. Now benefits are being cut, food prices are up 3.1% year-over-year, the economy lost 92,000 jobs in February 2026, and the survey that would document the consequences has been eliminated.

If you are a researcher, this is the last benchmark. If you are an advocate, this is the evidence base. If you are a policymaker, these are the people whose outcomes you can no longer measure.

These numbers are the “before.” We may never get to measure the “after.”

Some weeks, life and work just get away from you. This was one of those weeks for me.

I have several genuinely interesting posts in the pipeline — topics I think you’ll find worth the wait — but none of them are ready for release today. Rather than rushing something out the door, I’m choosing to hold.

So instead of a myth-versus-data or methodology breakdown this Friday, I want to offer a different kind of invitation.

If you follow this series, take this one for yourself:

• Grab coffee with a colleague you’ve been meaning to catch up with. The kind of conversation that doesn’t happen over email.
• Read the new Urban Institute brief: Food Insecurity Remained High in 2025, As Safety Net Cuts Loom. It dropped this week and it’s timely — nearly 1 in 4 working-age adults reported food insecurity in 2025, and the analysis provides a critical baseline as OBBBA-era SNAP cuts begin rolling out. Their team does rigorous, accessible work, and this one belongs in your reading stack.
• Or — and I mean this sincerely — leave a little early today. Spend that extra hour with the people who matter most to you.
• And if you just need a laugh, my Lego-inspired dismal brick working paper — a formal economic model of parental Lego pain (if you know, you KNOW), complete with the Invisible Caltrops Theorem and a Rasch measurement model for pedal vulnerability — is on my website: matthewpatrickrabbitt.com/lego_pain_economics.pdf. Science has never been more applied.

The work matters. So does the rest. See you next Friday.

— Matthew

“Food-insecure families just need to budget better.”

It’s a persistent idea — in op-eds, legislative debates, and the design of nutrition education requirements tied to benefit programs. The CPS Food Security Supplement tests it directly, asking every household: How much do you usually spend on food each week? Would you need to spend more, less, or the same? If more — how much more?

Among food-insecure households that report needing to spend more — and 56% do — the median shortfall is $100 per week. If sustained, that’s $5,200 per year. A specific dollar amount that households themselves identify as the distance between what they spend and what they need.

They’re not overspending. The median food-insecure household spends $150 per week — above the USDA’s Thrifty Food Plan ($133), widely considered a lower bound on the cost of a nutritionally adequate diet. More than half (53%) already spend at or above that benchmark. The gap isn’t between actual spending and smarter spending. It’s between actual spending and enough spending.

They’re already making the efficient choices. Food-insecure households allocate 78% of food spending to supermarkets vs. 68% for food-secure households. Those reporting a gap have the highest supermarket shares (80%) and spend 0% on restaurants. They’re buying where food costs least — and have already cut what costs most.

Three numbers: usual weekly spending ~$150, self-reported adequate spending ~$200, USDA Thrifty Food Plan ~$133. When actual spending already exceeds the government’s own adequacy floor, the problem isn’t allocation. It’s resources.

But what about the 44% who don’t report a gap? About 35% say spending is adequate; 9% say they could spend less. Their circumstances are measurably different — higher incomes (median IPR 1.78 vs. 1.45), less severe food insecurity (32% VLFS vs. 45%), and larger SNAP benefits ($62/week vs. $55). For some, the absence of a reported gap may reflect adapted expectations rather than adequate diets: 32% of this group has very low food security.

Even SNAP doesn’t fully close it. Among food-insecure SNAP recipients who report a gap, the median benefit covers 66% of the shortfall. Only 35% receive enough to fully close the gap. SNAP was designed as a supplement — but when spending is already at the TFP floor, supplementing an inadequate base still leaves families short.

For households below the poverty line, the gap represents 42% of annual income. For families with children, $100 on $200 in weekly spending. No reallocation within the food budget closes a gap that large.

These households can quantify how much more they need. The 56% who report a gap and the 44% who don’t both point in the same direction: the issue was never that they didn’t know how to spend. It’s that they didn’t have enough to spend.

We measure how bad food insecurity gets. We rarely measure how often it happens. That’s a dimension we’ve been overlooking — and the data to change that already exist.

In 2024, 78% of food insecure households experienced it episodically. The other 22% faced it persistently, month after month. This methodology has been available for over a decade yet remains underutilized.

A household that worried about food once after a layoff and one that worried every month look identical under the current measure.

From severity to severity × frequency. USDA classifies food insecure households by severity: low food security or very low food security (VLFS). This tells us how bad it got over the past 12 months — but not how often.

Take an item from the 10-item adult food security scale: “I worried whether our food would run out before we got money to buy more.” The standard scale codes “often true” and “sometimes true” as affirmative. The Frequent Food Insecurity Scale counts only “often true.”

Over a decade ago, USDA researcher Mark Nord showed how to score 7 of those items on frequency. The raw score ranges from 0 to 7; households scoring 3 or more are classified as frequently or persistently food insecure — applying the same threshold logic as the standard scale. Cross-classifying severity with frequency expands 2 categories into 4 subgroups within the same 12-month window. No new data collection needed.

What the 2024 data show. I applied this methodology to the December 2024 CPS-FSS.

The standard classification finds 13.7% food insecure, including 5.4% VLFS. Adding the frequency dimension reveals:

• 78% experienced it episodically; 22% persistently
• Among low food security households, only 6.5% were persistent
• Among VLFS households — the most severe — 45% faced it month after month
• Persistence is not uniform: it concentrates heavily at the VLFS level

Expanding research possibilities. Short-run severity × frequency classification enables new analytical pathways:

For researchers: Even within a single survey year, studies can now distinguish persistent from episodic exposure — examining how frequency correlates with stress, mental health, coping strategies, and program participation.

For advocates: “3.3 million households faced very low food security month after month” is more precise than aggregating all 7.2 million very low food secure households.

For policymakers: Episodic food insecurity may respond to emergency food assistance and short-term transfers. Persistent food insecurity demands structural solutions — living wages, affordable housing, and sustained nutrition support.

Because this builds on the 10-item adult food security scale, any dataset with it can implement this approach — not just the CPS-FSS.

The data are already being collected. We just need to use them more completely.

A quick note for those following along with #FoodSecurityFridays — tomorrow’s post will be postponed. Week 7 will go up next Friday instead, and the series will continue on schedule from there.

In the meantime, I’d love to hear from you: what topics would you like to see covered in this series?

Over the first five weeks, we’ve built from the ground up:

• Week 1: A job doesn’t guarantee food security — 58% of food-insecure households have at least one employed adult
• Week 2: What food insecurity actually looks like — a severity spectrum, from worrying about running out of food (92.7%) to not eating for a whole day (13.2%)
• Week 3: Higher-income households can’t be food insecure — 6.2 million households above 185% of the poverty line say otherwise
• Week 4: Food insecurity is a permanent state — it’s not; most spells are episodic, averaging 7 months per year, and over half resolve within 2 years
• Week 5: If parents are food insecure, the kids must be too — in about half of food-insecure households with children, parents absorb the hardship and keep children’s eating patterns near-normal. But when severity escalates, that protective buffer collapses

Every post challenges a common assumption with what the data actually show. And there’s a lot more to come — older adults, college students, military families, veterans, SNAP, measurement, and more.

But I want this series to reflect the questions people are asking, not just the ones I’ve planned. If there’s a myth you’ve heard, a population you think deserves attention, a method you’re curious about, or a topic where you’d like to see what the data say — drop it in the comments or send me a message. I read every one.

Catch up on the full series: matthewpatrickrabbitt.com/food-security-fridays.html

See you next Friday.

In 2024, 18.4% of U.S. households with children — 6.7 million households — were food insecure. But the top-line number obscures a critical pattern: parents absorb the impact.

In about half of food-insecure households with children (9.3%), only the adults were food insecure. Parents cut their own meals, worried about running out of food, and couldn't afford balanced diets — but kept their kids' eating patterns near-normal. This is parental shielding.

In the other half — 9.1% of all households with children, or 3.3 million households — that protective buffer broke down. Both adults and children were food insecure. And in 318,000 of those households, children experienced the most severe form: skipped meals, hunger, or not eating for an entire day.

These top-line numbers raise a question: what distinguishes the half where shielding holds from the half where it breaks? My analysis of the December 2024 CPS microdata points to severity. Parents who successfully shield their children don't report higher food insecurity — they report lower.

The pattern is visible across the full severity gradient. At the anxiety level, shielding parents look nearly identical to every other food-insecure adult — worry about food running out is almost universal. But gaps open at the diet-quality level — only 73.9% of shielding parents couldn't afford balanced meals, vs. 89.6% of food-insecure adults without children — and widen sharply when food intake is reduced. Among shielding parents, 22.9% went hungry but didn't eat, vs. 37.5% without children. Only 6.8% didn't eat for a whole day, vs. 15.0%.

When shielding fails, the picture looks very different. Parents in households where children are also food insecure report rates far higher than shielding parents: 35.0% went hungry (vs. 22.9% where shielding held), and 14.4% didn't eat for a whole day (vs. 6.8%). When adult severity reaches the level of reduced intake, children are exposed too.

So no — if parents are food insecure, the kids are not necessarily food insecure too. Shielding parents absorb the hardship through anxiety and diet-quality compromises, largely avoiding the most severe forms. But when severity escalates past that threshold, the protective buffer collapses. Programs that prevent adult food insecurity from deepening — SNAP, WIC, school meals, the Child Tax Credit — don't just help adults. They preserve the parental buffer that keeps children fed.

What policy levers do you think are most effective at preventing that escalation?

When people hear that 18.3 million U.S. households were food insecure in 2024, many assume it's permanent. That assumption shapes policy — if food insecurity looks fixed, the response defaults to long-term assistance alone. But the data tell a different story — one with a beginning, middle, and end.

How a spell begins. Food insecurity is typically triggered by a sudden change in circumstances. A job loss. A medical bill. A car repair that crowds out the grocery budget. The monthly SNAP benefit cycle, where many households exhaust benefits before month's end. Or the expiration of policy supports like emergency allotments and the expanded Child Tax Credit.

What a spell looks like. Once it starts, food insecurity tends to be recurrent but not constant. On average, affected households experienced the condition in 7 months of the year — not all 12. And 56% were also food insecure during the 30 days before the December survey — the rest experienced it only in earlier months of the year (Rabbitt et al., 2025). Of those food insecure in the 30 days before the survey, most experienced hardship on only 1 to 7 days.

How a spell ends. More than half of new food insecurity spells resolve within 2 years (Lee, Barrett, & Hoddinott, 2024 — based on an expenditure-based measure, not the standard survey scale). Among households food insecure at some point over a 5-year period, half experienced it in just a single year, and only 6% in all five (Wilde, Nord, & Zager, 2010). But recurrence is common — 43% experienced it in 2 to 4 of those years, suggesting that while individual spells end, many households cycle back in.

Why this matters for policy. If food insecurity has a lifecycle — entry, duration, and exit — then interventions should target each stage. Prevention and rapid response at the front end. Bridge programs and benefit continuity during a spell. And sustained support to reduce recurrence after a spell ends. The burden is not equally shared: households headed by women, people of color, or those with less education face longer, more persistent spells.

Food insecurity is not a fixed identity. It's a fluid economic condition — and understanding its arc changes how we respond.

If a family earns $60,000 a year, they can't be food insecure. Right?

Wrong.

In 2024, 6.2 million households with incomes above 185% of the federal poverty line — roughly $59,000 for a family of four — were food insecure. That's nearly 1 in 12 households in that income range.

And these aren't borderline cases. Among food-insecure households above 185% FPL:

• 91% worried food would run out before they had money to buy more
• 85% couldn't afford balanced meals
• 63% cut the size of or skipped meals
• 35% went hungry because they couldn't afford food

The natural question: are they just confused by the survey? Nord & Brent (2002) investigated exactly this. They found that a small proportion, at most, of measured food insecurity among higher-income households could be attributed to misunderstanding or erratic responses. About 90% of these households explicitly cite not having enough money as the reason — the same resource constraint reported by lower-income food-insecure households.

So what explains it?

Income volatility is a major driver. Bartfeld & Collins (2017) found that both income shocks and expense shocks independently predict food insecurity — a job loss, reduced hours, or a sudden medical bill can push a household into food hardship even when their annual earnings look adequate. Loopstra & Tarasuk (2013) showed that changes in income and employment status directly affect the severity of food insecurity, with even modest income losses translating into measurably worse food security outcomes.

There's also a measurement timing gap. Food insecurity captures the worst period over the past 12 months, while income is averaged over the same year. A household that earned $65,000 annually might have spent three months after a layoff earning nothing. The food insecurity measure catches that crisis. The income measure smooths over it.

These households aren't gaming the system. They're caught between a bad few months and an annual statistic that makes them look fine.

185% FPL is where most program eligibility ends. Food insecurity doesn't.

In 2024, 13.7% of U.S. households—18.3 million—were food insecure at some point during the year. But food insecurity is a spectrum, not a single experience. The conditions households face vary widely in severity.

Among food-insecure households:

• 92.7% worried food would run out before they had money to buy more
• 87.5% reported the food they bought didn't last
• 86.5% couldn't afford to eat balanced meals
• 60.5% cut meal sizes or skipped meals
• 34.3% were hungry but didn't eat because they couldn't afford food
• 13.2% didn't eat for a whole day

Notice the pattern: nearly all food-insecure households experience anxiety and uncertainty about their food supply. But the share reporting more severe conditions—actual reductions in food intake—drops off substantially.

This is because food insecurity has two tiers:

Low food security (8.3% of U.S. households): These households report anxiety about food, reduced diet quality, and reliance on cheaper foods—but they generally avoid substantial reductions in food intake.

Very low food security (5.4% of U.S. households, or 7.2 million): These households experience disrupted eating patterns. Adults skip meals, go hungry, or don't eat for entire days because there isn't enough money for food.

For households with children, the picture is more complex. Parents often shield children from the worst effects, but 9.1% of households with children had food-insecure children in 2024.

Why does this matter for policy?

Different levels of severity may call for different responses. Households with low food security may benefit most from programs that improve diet quality and reduce food-related stress—like SNAP benefits that stretch further or initiatives that lower the cost of nutritious foods. Households with very low food security need more immediate interventions to prevent hunger and its health consequences.

Understanding where households fall on this spectrum helps target resources more effectively and design policies that address the full range of food insecurity experiences.

That's more than half.

In 2023, 18 million U.S. households faced food insecurity—13.5% of all households nationwide. And the majority of them had someone working.

Wages, hours, and rising costs all play a role in whether working families can put food on the table. This is why food assistance programs like SNAP and WIC remain critical even for employed households.

Food insecurity isn't just an unemployment problem. It's a wages problem, a cost-of-living problem, and a policy problem.