First you reproduce the number. Then you get to argue with it.

USDA’s Economic Research Service recently published its fruit-consumption trends report (ERR-341).¹ It uses the National Cancer Institute’s usual-intake method — the right tool. It models habitual intake instead of naively averaging two recall days, and it respects the survey design. For 2017–March 2020 it reports 14.7% of adults and 23.2% of children meeting the fruit recommendation.

That ERS number is where I started. Building on our previous work², before I’d trust our own pipeline to say anything ERS didn’t, I wanted it to agree where ERS already had an answer. So I estimated the same quantity on NHANES three ways: the NCI two-part method (13.8% overall), a multilevel small-area model — the kind political scientists use to estimate opinion in all 50 states from one national poll (14.6%) — and a small-area model of per-day compliance (12.1%).

Source: FRESH dietary surveillance and equity pipeline (NCI two-part · Box-Cox MRP), NHANES 2017–18; USDA-ERS column from ERR-341.

Same neighborhood, three engines, matching USDA. The agreement is the least interesting thing in this post. It’s the price of admission.

The estimator isn’t the problem. The resolution is.

I want to be precise here, because it’s easy to strawman: USDA is not doing this wrong. They use usual intake. They split children from adults. The public tables are honest.

The problem is that the unit of reporting — “children,” “adults” — is far coarser than the unit of action. Nobody designs a fruit campaign for “children.” They design it for 4-year-olds, or for teenagers, or for a specific income tier in a specific community. And the moment you look at that resolution, the single number dissolves into something far more useful.

The age cliff dominates the figure, but the gap keeps splitting below the child-vs-adult line that usually gets reported: girls clear the bar more than boys⁵, Hispanic and multiracial kids run nearly double non-Hispanic White and Black kids, and adherence climbs monotonically with income.

None of that is visible in “23%.” All of it is in the same data.

“Fruit is expensive” explains less than you’d think

The standard explanation for a gap like this is cost, and it deserves to be taken seriously. A widely cited meta-analysis put the healthiest diets at about $1.50 a day more than the least healthy ones⁶, and affordability anchors much of the ultra-processed-food debate: lower-income families, the argument goes, can’t afford fresh produce, so you can’t simply tell them to trade chips for fruit⁷. There’s real evidence behind the worry.

And there is an income gradient here — lower-income kids at 18%, higher-income at 25%. The story isn’t wrong. It’s just small.

Put it next to the age cliff and it nearly disappears. The age gap is 55 points; income is 7. The income gradient is about 12 cents on the dollar of the age gradient, and that holds whether you measure in points or in ratios (≈10× across age, ≈1.4× across income).

It also can’t explain who’s actually winning. The highest-adherence kids in the country aren’t the richest. Non-Hispanic White kids, higher-income on average, sit at 18% — behind Mexican-American kids at 31%. Income isn’t even the through-line within the gradient: non-Hispanic Black kids are low (consistent with the cost story), but Hispanic kids — cutting the opposite way — are the highest in the country. One variable doesn’t move both directions.

That a single cost variable can’t carry the gap is exactly what USDA’s own report finds from the other side. In their model, household income and fruit prices have less influence on whether someone meets the recommendation than behaviors that signal concern for health and nutrition knowledge.⁸

None of this says cost doesn’t matter — Rao’s $1.50 is real, food deserts are real, and a 37% relative gap by income is worth closing. It says that for this gap, in fruit, cost is a second-order lever. Reach for “they can’t afford it” first and you’re tuning the smallest dial on the board — and stepping over a 55-point age collapse.

But “age” isn’t an answer either

So if it isn’t price, is it age? Not really. “Age” is a container, not a cause — knowing the gap lives in teenagers tells you where to look, not why it’s there. And part of the cliff isn’t behavior at all: the recommendation itself climbs from about one cup for a toddler to two for a teenager, so the bar rises at the very moment intake falls. The raw age effect quietly mixes a moving target with a real decline.⁹

Pull those apart and the explanations that are easy to reach for are social ones. The literature (a quick, non-systematic read — this is an essay, not a manuscript) has a familiar list for why fruit fades from childhood to adolescence: parents stop cutting it up and handing it over, kids gain autonomy over their own plates, home availability drops, peers and the market move in.¹⁰ Toddlers are handed fruit; somewhere in adolescence they take the plate over, and fruit slides off it. It’s a plausible story — I’m just not sure it’s the whole one. (It isn’t simply a switch off juice, at least: juice and whole fruit fall together. And the companion note turns up a longer-range pattern the tidy autonomy-and-market story doesn’t obviously survive.)

If the cliff is social rather than biological, “age” is the wrong name for it — so what kind of variable is it? Part of the rising bar is plainly biological — bigger bodies need more food. But the fruit falling off the plate seems to be something else. Nothing in a teenager’s physiology rejects an apple; what changes looks social — who controls the plate, what’s in the vending machine, what a snack signals among friends, what the market is selling. If that read is right, the cliff is a social transition wearing a biological variable’s clothing.

Which makes the way we usually handle age a little strange. Much of geroscience treats it as the thing to measure — epigenetic clocks, biological-age scores, age read off the body to a decimal. A lot of health-inequality research does closer to the opposite, age-standardizing to net age out so the “real” social variables can show through. (Nutrition and life-course researchers do take the age decline seriously — to be fair — so this isn’t “nobody looks”; the narrower point is how age gets handled once the frame is inequality.) Either way, the fruit cliff is a social gradient that one tradition over-measures as biology and the other adjusts away — and, at least in these data, it may be the steepest one in the picture. That seems like a variable to study, not to adjust away. And it isn’t only fruit, or only children: the same lens would change how we read income, race, even old age — each of which we tend to treat as fixed when it’s at least partly built.

Design the estimate for the decision, not the press release

The aggregate “14%” or “23%” isn’t really a measurement. It’s a measurement collapsed to one number so it fits in a sentence. That’s fine for a headline and useless for a campaign.

So if you’re testing a message, you don’t need the average. You need to know the gap sits in the teen years, barely moves with income, and doesn’t follow the affordability script at all — and you need the estimate built at that resolution, with honest uncertainty, on a survey that was never designed to be sliced this thin. That’s a design problem: build the estimate to answer the targeting question you actually have, not the one that fits the press release — or the one that fits your prior about why poor people eat less fruit.

That’s the discipline behind reproducing USDA’s number in the first place: we earned the right to say it isn’t the number you want. The number you want — and the place you should actually be aiming, even if you don’t yet know why it’s there — was inside it the whole time.

Information can be health — but only at the resolution where someone can act on it.

Analysis: FRESH dietary surveillance and equity pipeline.

If you want to go deeper:

• USDA ERS, Trends in U.S. Fruit Consumption Relative to Recommendations in the Dietary Guidelines for Americans (ERR-341): https://www.ers.usda.gov/publications/pub-details?pubid=110657
• The companion Amber Waves piece, “Peeling Open U.S. Fruit Consumption Trends”: https://www.ers.usda.gov/amber-waves/2025/february/peeling-open-us-fruit-consumption-trends
• The National Cancer Institute’s usual-intake method: https://epi.grants.cancer.gov/diet/usualintakes/
• Our prior framework paper (the 100% orange juice case study), where this fruit-adherence pipeline started: https://doi.org/10.1080/09637486.2023.2241672

The subgroup numbers in full

The main essay shows the habitual estimates as a figure. Here is the full table — both estimands (habitual usual-intake and per-day compliance), every subgroup within children 2–19.

Per-day or habitual: which question is the guideline even asking?

The table above carries two columns — habitual and per-day — and they answer different questions.

Start with the daily reading. The Dietary Guidelines are written as a daily target — this many cups per day. Taken at its word, that’s a question about days: on a given day, did you hit the mark? The field usually doesn’t answer it that way. To keep a single lucky or unlucky recall day from speaking for a person, the convention — USDA’s included — is to model habitual intake, the long-run average, and ask whether that clears the bar.

Both are defensible, and they’re different questions. For an episodic food, the two need not agree — here they land within a few points of each other, but that isn’t guaranteed. And the guideline’s own wording arguably fits the daily reading at least as well as the habitual one — yet the single number you usually see has quietly picked one without telling you which. That’s why we report both, and lead with the habitual one (it’s the convention, and what USDA-ERS uses).

Is the age cliff just the moving recommendation?

The obvious objection: the fruit recommendation isn’t fixed — it climbs from about one cup for a toddler to two for a teenager, because energy needs climb. So is the adherence cliff just the bar getting harder, not kids eating less? We pulled the actual usual-intake estimates to find out.

Both things are true at once. Absolute intake nearly halves (1.29 → 0.66 cup-eq), and the bar rises 76%. Decomposing the collapse in intake-relative-to-target on the log scale, it’s roughly 54% real intake decline, 46% rising bar. So the moving recommendation is a big part of the story — but it can’t carry it alone. Even against a frozen toddler-level bar, teens would still fall short. Toddlers genuinely eat about twice the fruit that teenagers do — frozen bar or not.

This is exactly why the main essay treats age as where the gap lives, not why — and it’s the honest answer to “isn’t this just the bar?”: partly, but not mostly.

The boys-vs-girls gap is mostly the bar

Here the check changed our reading. In the main table, girls clear the recommendation more often than boys (25% vs 19%). The natural inference is that girls eat more fruit. They don’t.

In fact, girls’ mean usual intake (0.88 cup-eq) is slightly lower than boys’ (0.90). The adherence gap exists because girls get a lower recommendation (1.5 vs 2.0 cup in the older bands) — they clear a lower bar with the same fruit. The boys/girls adherence gap is a denominator effect, not a behavioral one.

Race and income are the opposite — real intake differences, because the bar doesn’t vary by race or income. Mexican-American and other Hispanic children eat ~1.13 cup-eq, against ~0.76–0.79 for non-Hispanic White and Black children; intake climbs from 0.80 (lower-income) to 0.98 (higher-income). Those gaps are about fruit, not about the bar.

Which energy do you adjust by? (It matters.)

Because the recommendation is energy-scaled, you can normalize fruit intake two ways: by a reference energy need (a standard moderately-active person of that age and sex — what USDA-ERS does, and what we do), or by each person’s own reported energy from the 24-hour recall.

We use the reference, deliberately. Reported energy is systematically under-reported, and unevenly — under-reporting is worst in adolescents and higher-BMI individuals, while toddlers’ intake is caregiver-reported and more complete. Normalizing by reported energy hands the worst under-reporters an artificially low bar, which inflates their apparent adherence — and since teens under-report most, it would quietly flatten the very cliff we’re trying to measure. Usual-intake modeling cleans up day-to-day noise, but it does not fix this kind of systematic bias; only an external calibration (a biomarker like doubly-labeled water) would, and we don’t have one here.

So the intake ÷ bar column above is an energy-adjusted view that leans only on reference energy — no recall error in the denominator. That’s the robust way to ask the question, and it still shows the cliff.

Is it a juice story? (No — we checked.)

The tempting explanation is juice: toddlers drink it, teenagers switch to soda, and the “fruit” cliff is really a juice cliff. The pipeline already carries the components — F_JUICE alongside total fruit, plus the 100%-orange-juice counterfactual from our published OJ work — so this is checkable, not speculative. It isn’t a juice story.

Juice and whole fruit fall in lockstep — both down about 40% from the early-childhood peak to the late teens — and juice holds a roughly constant third of fruit at every age. So the cliff is a broad, whole-diet decline, not a single beverage being weaned away. (That, if anything, reinforces the social-transition reading: it’s the whole food environment shifting, not one drink.)

One component bucks the trend: 100% orange juice falls only 15% while everything else falls 40%, rising from 4% of toddlers’ fruit to 16% of teenagers’. As the fruit cliff deepens, OJ is disproportionately what’s left — a thread that ties straight back to our OJ paper and is worth a dedicated follow-up (built on usual-intake models, not these raw day-1 means).

The decline isn’t one-way: fruit intake is U-shaped

The main essay stays with children, and for the child decline alone the usual explanations (autonomy, declining parental support, the peer-and-market environment) are hard to rule out — they all move the right way. But widen the lens to the whole lifespan and a wrinkle appears that they don’t obviously survive.

That wrinkle is the shape of the lifespan curve: fruit intake doesn’t keep falling with age. It bottoms out in adolescence and then climbs back through adulthood — by the 70s nearly to early-childhood levels, well above the teenage floor. That’s a problem for the tidy story. Autonomy, market exposure, and the absence of a parent cutting up your fruit all increase with age — and a cause that rises monotonically can’t produce a U. If those were the whole account, 70-year-olds — maximal autonomy, a lifetime of market exposure — would be the worst; instead they’re nearly the best.

So either something specific to the adolescent window is doing work the standard list misses, or the recovery is driven by something else. Two honest possibilities we can’t separate from a single cross-section:

• an age effect — health concern, routine, and means rising into adulthood (the same kind of behaviors USDA’s drivers model flags); or
• a cohort effect — older Americans grew up eating more fruit and kept the habit.

That’s the age–period–cohort confound, and one survey snapshot can’t resolve it. Either way it does the same thing to the easy explanation: it rules out “autonomy and the market” as a complete account — which is why the main essay leaves the why behind the cliff genuinely unsettled rather than closed.

What this note doesn’t settle

• The decomposition is mean-based. Adherence is a tail probability, P(usual ≥ bar), so it also depends on how the spread of usual intake changes with age — not just the mean. A full counterfactual (recompute adherence holding the bar fixed) is the cleaner test.
• Intake estimates here are from the two-part (gamma) engine; the main essay’s adherence uses the Box-Cox MRP. The means are close, but we’d recompute on one engine for a final figure.
• Differential under-reporting cuts both ways. It biases the reported-energy bar (above), but it also touches the fruit numerator itself — teens likely under-report fruit too, which would push the other direction. Neither is fixable without a calibration sub-study.

None of these move the headline: the gap lives in age, age is a container not a cause, and cost is a second-order lever. They just mark where the real work — the why — begins.

Analysis: FRESH dietary surveillance and equity pipeline (NCI two-part · Box-Cox MRP).