Chicken passed beef in 1992. And in 2003. And in 2010.

This data system does not measure eating. It is a supply-and-use balance sheet: total supply — production plus imports plus beginning stocks — minus feed, seed, exports, industrial use and ending stocks. What is left is the residual that moved into U.S. food marketing channels. ERS calls that availability, and so does every number here.

That residual still carries non-edible portions, spoilage, plate waste and ingredients later exported inside processed goods. ERS states directly that the calories in the food supply run well above what the population needs. So the levels are supply, biased upward by an unquantified amount — and the defensible object is thetrend, and the year one line crosses another. Nothing on this page should be read as “Americans ate this much”. The one exception is the loss-adjusted series, which is named as such where it appears.

The per-capita denominator is the U.S. resident population plus Armed Forces overseas, July 1, taken from the population column embedded in each commodity file.

ERS measures every commodity wherever its source data actually exist — farm gate for fruit and vegetables, the processing plant for margarine, the slaughter plant for meat — and then republishes the same meat supply at two further weights. A beef carcass carries far more bone and separable fat than a chicken does, so every step toward the plate takes proportionally more weight off beef.

None of the three adjusts for cooking loss, plate waste or spoilage. Boneless weight is the closest of them to what reaches a plate, and it is still not that. The shaded wedge on the opening mark spans 1992–2010, the 18 years over which the three bases disagree about who is ahead. Each band plots beef minus chicken for every annual observation from 1909 to 2021: no smoothing, no interpolation, no fitted curve. “Chicken” means chicken — turkey is excluded. “Beef” is beef alone, not total red meat.

• Dairy. Fluid milk and cheese are product pounds; the total is ERS's own milkfat-equivalent aggregate column, not a sum of the parts. The mark draws all three against their own 1970 level so a halving and a tripling are the same distance.
• Butter. The added-fats table rounds butter to one decimal and puts its trough in 1981; the unrounded dairy table puts it at 4.19 lb in 1997. The unrounded figure is authoritative and is the one used. Margarine cannot be carried past 2010 — the series stops there, and the mark stops with it.
• Calories. The supply line is the archived nutrient-availability table, published rounded to two significant figures. The eaten line is the loss-adjusted series, which rests on loss coefficients ERS set in the mid-1990s and holds constant — so it cannot support any claim about waste changing over time. Neither series continues past 2010: the Census Bureau discontinued the source reports.
• The 2000 seam. The largest single-year rise in the eaten line, +89 kcal, coincides with more firms reporting vegetable-oil production to the Census Bureau. ERS says so in a footnote. The mark draws the join open rather than smoothing over it.
• First-counted years. Restricted to the two per-commodity aggregate tables. Every other file restates the same food at a second weight, and a restatement's first year is a publishing fact, not a national one. A start year is not evidence that consumption was zero before it.
• Donated dairy. The column is all USDA-donated dairy on a milkfat-equivalent basis, not cheese specifically. “Government cheese” is the cultural name for it.
• Reversals. The scan covers per-capita quantity series in the food-availability files with at least 25 years, at least 1 lb at the turning point, and a 20% move in each direction. It returns 553 rows, deduplicated to 253 — one per commodity per turn. Loss-adjusted files are excluded because they restate each commodity at seven weights.
• Fish. Deliberately absent. ERS has announced a revision that will cut coverage to 1990–2022, withdraw the supply/use tables, and produce values “not comparable to previously published data”.

• No cause. This data system contains no price, no policy and no income variable. It can show that lines crossed; it cannot say why. Any explanation would come from outside the file, and none is offered here.
• No one's own row. No state, regional, demographic or socioeconomic breakout exists. National aggregate only.
• No finer grain than a year. Calendar year is the smallest unit published, across the whole span. There is no seasonal or weekly cut.
• No fixed vintage. ERS revises released tables on a per-table schedule of its own. The meat tables held here are stamped 2022-11-16; a future vintage can move these values.

The site imports pipeline/out/hero.json and pipeline/out/findings.json and draws them. Both are produced by the pipeline, which starts from the published bytes:

The download step records each file's byte length and SHA-256 before anything is decoded, so every number traces back to specific bytes. Raw downloads are not committed; step 1 refetches them and the manifest proves they match. Every figure quoted in the prose above is read out of the same JSON the drawings use, so the words and the marks cannot disagree.

Source: USDA Economic Research Service, Food Availability (Per Capita) Data System, retrieved 2026-08-27 by direct HTTPS download. Calculated by ERS/USDA based on data from various sources. ERS publishes these tables as U.S. Government work — public domain under 17 U.S.C. §105 — though no licence statement appears on the product page, so that status is inferred rather than quoted.

One drawing sits beside each headline. Every one was generated with OpenAI'sgpt-image-1 from a written description of a single object, then cut out and trimmed to its own ink. They are decoration and orientation only: a picture tells you which food a stop is about, and nothing else.

No illustration asserts or implies a quantity. The size of a drawing is the same on every stop, the number of objects in it is never a count, and nothing in it is scaled to a number in the data. If a picture and a mark ever seem to disagree, the mark is the claim and the picture is a label. Every quantity on this page comes from the marks and the JSON behind them.