experimental-design
The first randomized trial
It is 1948, and Britain's Medical Research Council has only a scarce supply of streptomycin and a deadly disease, tuberculosis, to spend it on — a shortage that, for the first time, makes it ethical and necessary to assign patients to treatment or control by a concealed random schedule rather than a physician's judgment. Design that trial: how randomization removes the selection bias that had made every previous cure claim untrustworthy, how allocation is concealed, and what the control group licenses you to conclude. Get it wrong and the drug's effect is confounded with which patients doctors chose to save, and medicine keeps enshrining treatments on impression — this template becomes the standard every later therapy must pass.
Who this problem belongs to
The two figures whose methods fit it best, out of 35 in contention.
The 1948 MRC streptomycin trial, designed by Austin Bradford Hill, applied Fisher's own randomization principle — laid out in The Design of Experiments in 1935 — to human patients for the first time, using a table of random numbers to assign treatment or control precisely so that selection bias could not confound the drug's effect with which patients doctors happened to favor. Fisher's insistence that randomization, not merely careful matching, is what licenses causal inference is the entire logic Bradford Hill borrowed from agricultural trials into medicine. Fisher himself was not part of the medical trial, which is why the score falls short of a perfect mark, but the trial is his method transplanted wholesale into a new domain.
Rubin's potential-outcomes framework, developed from the 1970s onward, gives the rigorous modern language for exactly what the 1948 streptomycin trial accomplished: defining each patient's outcome under treatment and under control as fixed but only one observable, and showing that randomization is what makes the average of observed outcomes a valid estimate of the causal effect rather than a confounded comparison. His formalization retroactively explains with total precision why concealed random allocation, not physician judgment, licenses the trial's conclusion. He was not present for the trial itself and worked decades later, which is why he sits just below Fisher despite supplying the deepest theoretical justification for what the trial actually proved.
In the mind map
The same ideas, as concepts rather than history — in John's ML knowledge map.
35 figures are scored on this problem. Draw it in a battle to see where you land.