How do we know what works? In an uncertain world, full of untested treatments, disputed policies and untried inventions, it should be a pressing question. For an impoverished young tailor named Franz Reichelt, the answer was: we find out the hard way.
In February 1912, on a cold Sunday morning, Reichelt persuaded journalists, a camera crew and a small crowd to watch him demonstrate his new wearable parachute. The official story was that Reichelt would throw a dummy from the first floor of the Eiffel Tower, more than 50 metres up. Reichelt, however, had tipped off journalists that he would make the jump himself. Otherwise, one suspects, there would have been fewer spectators.
The parachute is, as it happens, something of a running joke in evidence-based policymaking. In the Christmas 2003 edition of the British Medical Journal, obstetrician Gordon CS Smith and public health researcher Jill Pell published a systematic review of “Parachute use to prevent death and major trauma related to gravitational challenge”. They noted, with mock sorrow, that “we were unable to identify any randomised controlled trials of parachute intervention” and called for “the most radical protagonists of evidence-based medicine” to organise and participate in a rigorous randomised trial of the parachute.
In case the joke has passed you by, the argument being lampooned is a complaint by advocates of evidence-based policymaking. Many well-established medical practices — along with ideas in policing, criminal justice, education, economics and beyond — have never been seriously tested by the gold standard for evidence: a double-blind, randomised controlled trial (RCT). RCT advocates suggest that this is a scandal. Is it?
Leave that question hanging for a moment, because 15 years later — again in the BMJ Christmas issue — a team of researchers detonated an evidential bombshell. They had run the long-missing randomised trial of the parachute — and found that parachutes were no better at preventing major injury or death than an empty backpack. Anyone surprised at this result need only consult the small print: nobody in the trial had suffered any noteworthy injury at all after jumping from an aircraft, and this is because they did so “at significantly lower altitude . . . and lower velocity”. To be more specific, they jumped out of “small stationary aircraft on the ground”. Another hilarious satire from the BMJ.
There are serious lessons here, for those who can stop sniggering. The first is that, yes, there is no need for a RCT of a technology — such as the parachute — whose benefits are indisputable. The second is that RCTs themselves do not necessarily prove anything: the researchers suggest “cautious extrapolation to high-altitude jumps”.
It is striking how incautious extrapolation can be, particularly in press releases and breathless media summaries. A small study of nursery school children drawing pictures is generalised to be a fully fledged theory of human motivation. Big claims about the benefits or risks of a food additive teeter on the questionable foundation of research on laboratory mice.
Randomisation is neither the only route to knowledge nor a guarantee of eternal truth. But it is a very good idea — which may help to explain why there have been serious randomised trials, even of parachutes. As Andrew Leigh explains in Randomistas (2018), “experiments on parachute efficacy and safety are widespread”. Some of them use crash dummies, while others investigate details of technique or equipment: paratroopers are six times more likely to suffer an ankle sprain if they jump without an ankle brace. So there.
One important reason to use some form of randomisation is that it strips away all kinds of potential confounding. Do small class sizes help students to thrive? You won’t find that out by simply comparing an overcrowded, cash-strapped inner-city academy with a lavishly funded public school: their class sizes may differ, but so does everything else. If you want to learn about learning, run a randomised experiment.
The second reason, still under-appreciated, is that experts know less than they like to think. Doctors embraced bloodletting for 2,000 years before belatedly acknowledging that it is rarely an appropriate treatment. (The Lancet is named after a bloodletting tool.) More recently, as Helen Pearson notes in Beyond Belief (2026), “In 1980s Britain . . . it was standard practice to shave pregnant women’s pubic hair when they were admitted to hospital and to give them enemas.” Episiotomies, too — cuts in the opening of the vagina — were “pretty routine”. None of this was evidence-based. It just seemed to doctors that it would be a jolly good idea.
In the summer of 2020, when the Recovery trial revealed that a cheap steroid, dexamethasone, was hugely beneficial to Covid-19 patients undergoing mechanical ventilation, one doctor told me that this so-called discovery was simple common sense. But Sir Martin Landray, one of the leaders of the trial, recalls the letters that experts had sent to him and to the regulators warning that dexamethasone — which suppresses the immune system — was so dangerous to Covid patients that it was unethical even to test it. When some experts think it’s a parachute, and others think it’s bloodletting, it’s a good idea to run the trial.
What happens when you need a decision urgently, and gold-standard evidence does not exist? One approach is to do what Landray and his colleagues did: move fast. By introducing a simple clinical trial alongside routine care, they were able to get the trial running within 10 days and produce life-saving results within 100.
But what if there is no trial, and no prospect of one? Then, argues Pearson, we need “aluminium standard” evidence. Be as clear as possible about your hypothesis, break it down into a chain of causal links, and get the best data you can — for and against — to test each link in the chain. Gold is better, but aluminium is cheap and lightweight and sometimes sufficient for the job.
Franz Reichelt was a better publicist than parachute designer. His leap from the Eiffel Tower was captured for the ages on Pathé-Journal newsreel. I would advise against watching it. He died of a shattered skull and spine, and — perhaps — of a coronary arrest mid-fall. Wikipedia simply lists his cause of death as “parachute failure”. The self-belief of a pioneer is a powerful thing. Normally, when the experts propose untested solutions, they are not the ones who suffer the consequences.
Written for and first published in the Financial Times on 24 June 2026.
Loyal readers might enjoy the book that started it all, The Undercover Economist.
I’ve set up a storefront on Bookshop in the United States and the United Kingdom. Links to Bookshop and Amazon may generate referral fees.