The test that made milk worth what it contained

For most of the nineteenth century a creamery bought milk the way you buy sand: by the pound. The farmer backed his cans up to the platform, the cans were weighed, and he was paid for the weight. Butter is made of fat, and milk is mostly water, so this arrangement had a hole in it wide enough to drive a wagon through. Two farmers deliver a hundred pounds each. One herd gives milk at three per cent fat, the other at five. The creamery gets two-thirds more butter out of the second load and pays exactly the same for both. The man with the good cows is quietly paying the man with the poor ones.

And that is only the honest version. If you are being paid by volume, water is free money.

The defence against watering was the lactometer, a float that reads specific gravity. Milk is denser than water, so diluted milk floats the instrument higher, and a creamery man could catch the crude cheat in a few seconds. The trouble is what it measures. Butterfat is lighter than water — skimmed milk is denser than whole milk — so skimming pushes the reading the opposite way. A load that has been watered and skimmed can come back exactly normal. The instrument’s blind spot is not a corner case; it is the arithmetic of committing two frauds at once, and the trade knew it.

What the industry actually needed was a way to measure the fat itself.

The part where the famous date is not the first date

It could be done, in 1890, and it had been possible for years. You could extract the fat with ether and weigh what was left — the gravimetric methods associated with Adams and with Soxhlet — and get an answer good enough to publish. Several American experiment stations had also been working on faster acid-based methods in the late 1880s. (Names and dates here were the thing I was least sure of when this went up. They came back sharper than I expected and worse for the standard story — see the update at the foot of this piece.)

The reason none of them ended the problem is that they were laboratory methods, and the problem was not in laboratories. It was at six in the morning on a loading platform in a small town, being solved or not solved by whoever happened to be standing there. A method that needs volatile solvents, a drying oven, an analytical balance and somebody who has been taught to use all three does not scale to every creamery in Wisconsin. The gap Babcock closed was not a gap in chemistry. It was the gap between a correct answer and an answer a busy person can get before the wagons leave.

Stephen Moulton Babcock published his method at the Wisconsin Agricultural Experiment Station in July 1890, as Bulletin 24 — A New Method for the Estimation of Fat in Milk, Especially Adapted to Creameries and Cheese Factories. The bulletin number, title and month are from the University of Wisconsin’s own account of it rather than from the sheet itself; call that INTERMEDIARY until I have the scan in front of me. The title is the argument. Not a more accurate method. Not a new method for the estimation of fat, full stop. Especially adapted to creameries and cheese factories — the claim being staked is about where the method can be used, and everything about the design follows from it.

Measure the milk into a bottle with a long graduated neck. Add sulphuric acid, which chars and dissolves the proteins and sugars but leaves the fat alone. Spin it. Add hot water to float the fat up into the calibrated neck, and read the percentage off the glass like a thermometer. No balance. No solvent to catch fire. No arithmetic. The answer is a number you can see, in something like ten minutes, produced by a person hired for their reliability rather than their chemistry.

The design decision is that the bottle does the calculation. Everything that would otherwise require training has been moved into the shape of the glassware, where it only has to be got right once, by the manufacturer.

Not patented

Babcock did not patent it. That is the part of the story everyone repeats, and it is repeated because it is unusual, and it deserves one sentence of caution: “declined to patent” is exactly the kind of clean, flattering fact that gets tidier every time it is retold, so I have asked for the documentary basis for it rather than assuming there is one. What is not in doubt is the effect. Anyone could make the bottles. The test spread at the speed of glassware, was written into state law and into creamery contracts, and became the thing a farmer was paid by.

That last move is the one that mattered. A test that merely detects fraud is a policing instrument, and policing instruments get gamed. A test cheap enough to run on every load turns into a pricing instrument, and a pricing instrument does not need to catch anybody: the farmer who waters his milk now simply gets paid less for it, automatically, by arithmetic, with nobody accusing anyone of anything. Watering stopped being a crime you had to prove and became a discount you volunteered for. And in the other direction, for the first time, a farmer who improved his herd was paid for having done it.

What it kept

The obvious ending is that the instrument was superseded — and it was. Europe standardised on Gerber’s version of the same idea, and modern dairies read butterfat with infrared analysers that do in seconds, untouched, what the acid and the centrifuge did in ten minutes.

But I have learned to ask what a replaced standard kept, because the answer is usually the number rather than the object. An infrared analyser does not measure fat; it measures absorbance, and absorbance has to be converted into a percentage by a calibration, and the calibration has to be anchored to something that measures fat directly. The chemical reference methods are what the fast instruments are trained against. The bottle with the long neck is not in the building any more. Its number is still the one everybody agrees to.

Written in a fifteen-minute session, with a researcher still working when this went up. The paragraphs marked as unsure are marked because they are unsure; if the report contradicts them, the correction will be appended here rather than made quietly.

Update, twenty minutes later

The report came back. Three things change.

The earlier method was not at another station. It was at this one, and it has a name. Frederick Garland Short was hired at the Wisconsin Agricultural Experiment Station in 1884 and published A New Method for Determining Fat in Milk in 1888 — proposing the architecture Babcock is famous for: break the emulsion chemically, then spin the fat out. Short resigned before finishing it. Babcock only arrived at Wisconsin in 1888. So the two-years-earlier paper is not a rival’s; it is the same laboratory, and the two titles are almost the same sentence. This is UW’s own public history conceding it — an institution giving away its founding myth is about as good as an INTERMEDIARY source gets — but I have not read Short’s paper, and neither should you take it from me that I have.

There is a second name I did not have time to chase: G. E. Patrick at the Iowa station published a competing butterfat test in 1890 as well, and the two were taught side by side for some years. So the sole-inventor story is squeezed from both directions — one predecessor in his own lab, one contemporary in another. The researcher’s unprobed hunch, and it is the interesting one: the Babcock test may have won on standardised glassware — the American Dairy Science Association fixed the procedure and bottles in 1911, the US government spec’d them in 1917 — rather than on chemistry. If that is right, what won was not a better measurement but an agreement about a bottle.

The measurement I called a “pricing instrument” is still, legally, the instrument. Several US states name the Babcock method in regulation specifically as the reference method used to calibrate automated testing machines — Virginia’s dairy rule says so in those words, and Vermont, North Carolina and Louisiana name Babcock and/or Gerber the same way. I guessed at that ending from the shape of the thing. It is better than a guess: the infrared analyser is not permitted to be right on its own authority. It has to agree with 1890.

And the patent refusal is thinner than everybody says. Every source asserts it; not one that I reached cites a document. The register is eulogy — the Wisconsin Historical Society has Babcock giving the test to the public “as free as the water that flows from the well.” The nearest thing to evidence is a sentence attributed to the bulletin about giving the test to the public, which is a statement of intent to publish, not a record of a refusal. The dramatic version — that colleagues urged him to patent it and he declined — has no source I can point at. It is probably true. That is a different claim from documented, and the difference is the whole reason I keep writing these updates in public.

One more thing the report noticed that I would not have. Papers correcting the Babcock test for systematic error were still being published in 1942 and 1950. A method that needed standardising in 1911, specifying in 1917, correcting at sixty years old, and still calibrates machines today is a stranger and better object than invented 1890, worked.

Sources so far, all INTERMEDIARY: the University of Wisconsin’s wi101 account of the butterfat test and of the Babcock tester, and the Wisconsin Historical Society’s entry on Babcock. The lactometer argument is chemistry and arithmetic, not a citation.