Why does randomisation matter in field trials?
Soil changes across every field, so with one variety on one side and another on the other, you can’t tell the varieties from the soil. Letting chance decide which plot gets which treatment means no gradient, known or unknown, can line up with it except by luck. It is the only method that deals with confounders nobody has thought of.
Last time's quick check: an infection would have spread to staff as well as inmates. Pellagra didn't, which pointed to the one thing they didn't share, their diet. Today we go to a field in Hertfordshire. In 1843, at Rothamsted in Hertfordshire, John Bennet Lawes and Henry Gilbert began growing wheat on a field called Broadbalk, with different fertilisers on different strips. It is still running, one of the oldest continuous field experiments in the world.
By 1919, Rothamsted had seventy-six years of records and no reliable way to read them. So it hired a young mathematician, Ronald Fisher. The trouble with fields is that they're never even. Soil changes across a field: a damp corner, an old hedge line, a patch where muck was spread years ago. Put variety A on one side and variety B on the other, and you can't tell the varieties apart from the soil.
Fisher's answer was startlingly simple. Let chance decide which plot gets which treatment. Then no gradient, known or unknown, can line up with the treatment except by luck, and statistics can say how big that luck might be. That's randomisation. It's the only method that deals with confounders nobody has thought of.
He added two more ideas. Replication: test each treatment on several plots, so one odd plot can't decide the answer. And blocking: group plots that are alike, say along the same side of the field, and randomise within each group. Blocks soak up the differences you know about; randomisation handles the ones you don't.
Fisher set it out in a short paper in 1926, The Arrangement of Field Experiments, and in 1935 in a book, The Design of Experiments. Around the same time, in Poland, Jerzy Neyman wrote the ghost fields from 1.2 into mathematics, for crop varieties on plots: every plot with a yield under every variety, only one ever seen.
And one of the tools Fisher built on came from a brewery. William Gosset, testing barley for Guinness, published under the name "Student" in 1908 and gave us the t-test. Fisher's design is still how careful farm trials work. In Australia, researchers tested a red seaweed, Asparagopsis, as a cattle feed additive.
Twenty steers were randomly assigned to four diets, within blocks. At the highest dose, methane per kilogram of feed fell by 98 per cent; at half that dose, by 38. Twenty animals on one ration is a small trial, and methane from the animal isn't the whole farm's footprint. But because the diets were randomised, nobody had to wonder which steers had been picked for the seaweed.
Next: a geneticist who drew causes as arrows, and his father, who used butter and flaxseed to measure a cause from outside.
Blocking groups plots that are alike, such as those along the same side of a field, and randomises treatments within each group. Blocks soak up the differences you know about. Randomisation handles the ones you don’t, because chance, not any gradient in the soil, decides which plot gets which treatment.
Twenty steers were randomly assigned to four diets, within blocks. At the highest dose of the red seaweed Asparagopsis, methane per kilogram of feed fell by 98 per cent; at half that dose, by 38. It was a small trial, and methane from the animal isn’t the whole farm’s footprint, but randomisation made the comparison fair.
Causal evidence is changing how farming is measured and rewarded. Leave your email and we will tell you when new films, courses and findings are published.
Prefer to read online? Vitagri also writes on Substack. See our privacy policy for how we use your email.