Sit in a boardroom, a Whitehall ministry, or a tech company product review for more than 15 minutes and somebody will explain to you that human beings are irrational. People are primitive monkeys running on a defective “System 1” brain, you’ll be told. We panic over small losses, we can’t weigh probabilities, we get subconsciously “primed” by stray words on a page, and we need enlightened experts to arrange our “choice architecture” and “nudge” us into eating our vegetables and saving for retirement.
For 20 years, this catechism from Daniel Kahneman’s Thinking, Fast and Slow, Richard Thaler’s Nudge, and Dan Ariely’s Predictably Irrational has been the state religion of the Western managerial class. Permanent secretaries, management consultants, and corporate executives still quote those books as if they were quoting Isaac Newton.
Nobody seems to have told them that over the past decade, the entire intellectual edifice of behavioural economics has been completely dismantled.
Its famous laboratory experiments failed to replicate across dozens of universities. Its celebrity studies on honesty nudges turned out to rely on fabricated Excel spreadsheets. Its signature proofs that ordinary people can’t do probability turned out to be basic maths errors by the psychologists themselves. Even the field’s foundational theorists have started publishing mea culpas admitting that behavioural public policy was a colossal mistake that distracted governments from fixing real problems. The science is dead, but the people running our institutions are unfortunately still governing with its corpse.
Start with the laboratory experiments that turned behavioural science into an airport-bookshop empire.
In Chapter 4 of Thinking, Fast and Slow, Kahneman marvelled at “social priming”, the claim that subtle unconscious cues hijack human decisions. The centrepiece of the chapter was a 1996 experiment by the Yale psychologist John Bargh, claiming that 15 New York University students who unscrambled words associated with old age, such as “Florida” and “wrinkle”, walked more slowly down the hallway when leaving the room. Kahneman assured his millions of readers that they had “no choice but to accept that the major conclusions of these studies are true”.
Within one year, Stéphane Doyen and his colleagues ran the experiment again. Instead of relying on graduate students holding manual stopwatches in the corridor, they installed infrared timing sensors. The effect vanished completely, reappearing only when the researchers tricked the stopwatch-holders into expecting it. Young people don’t walk like pensioners because they read the word “Florida”; graduate students with stopwatches simply click what their professor wants them to click.
When the Many Labs replication project tested another textbook favourite, “money priming” (the claim that seeing faint pictures of dollar bills makes people selfish and right-wing), across 36 independent laboratories, 35 of the 36 labs found nothing at all, yielding a combined effect size of 0.03. Kahneman privately wrote to priming researchers warning of a “train wreck looming”, and shortly before his death he admitted that priming research was dead.
Defenders of the field tried to quarantine the damage by arguing that priming was merely social psychology, whereas hard behavioural economics rested on sterner stuff. Then the sterner stuff collapsed too.
Dan Ariely and Harvard Business School professor Francesca Gino became the 2 biggest stars in applied behavioural science for a 2012 study showing that tax authorities and insurers could “nudge” people into telling the truth on forms simply by moving the honesty signature box from the bottom of the page to the top. The White House and the UK Cabinet Office rushed to test the trick on real citizens. In 2023, forensic investigators at Data Colada proved that the data in the underlying Excel spreadsheets had been falsified; the paper was retracted and Harvard stripped Gino of her tenure.
Even Amos Tversky’s famous 1985 “hot hand fallacy”, taught in economics departments for 30 years as proof that human beings hallucinate patterns in random noise, turned out to be a maths mistake by the psychologists. Tversky and Thomas Gilovich had recorded basketball players shooting baskets, calculated how often a hit followed a hit, and concluded that because the rate was no higher than 50%, streaks were a cognitive illusion. In 2018, the economists Joshua Miller and Adam Sanjurjo proved in Econometrica that in any short sequence of coin flips, sampling without replacement pushes the expected fraction of heads following a head well below 50% (down to 40.5% in a 4-flip sequence). Once you correct the psychologists’ arithmetic, the basketball players really did have hot streaks. The sports fans had better probabilistic intuition than the Stanford professors who spent 3 decades mocking them.
The replication crisis was only half the scandal. Even when a laboratory effect is real, the theoretical premise of behavioural economics gets the maths of human survival backwards.
In 1954, when the mathematician Leonard Savage published The Foundations of Statistics and formalised modern expected-utility theory, he included a warning that economists promptly ignored. Savage pointed out that his probability equations only define rationality inside a “small world”, a closed game like roulette where every possible future state, payoff, and probability is known in advance. Applying those equations to the “large world” of real human life, with changing rules and unmeasured tail risks, was, in Savage’s words, “utterly ridiculous”.
Behavioural economists ignored that boundary. Whenever they found a gap between their small-world casino equations and how humans behave in the large world, they kept the equations and asserted that humans have a mental defect.
Take “loss aversion”, the cornerstone of Kahneman and Tversky’s Nobel Prize-winning prospect theory, which says humans are irrational because we fear a £1,000 loss roughly 2 times as much as we welcome a £1,000 gain.
To see why our brains are wired that way, imagine an economist offers 100 people a coin-toss game. Heads increases your net worth by 50%; tails reduces it by 40%.
If 100 different people each flip the coin one time, 50 of them make 50% and 50 of them lose 40%. Across the room as a whole, average wealth goes up by 5%. An economist looking at that room-wide average declares the game a winner, and diagnoses anyone who walks away with “loss aversion”.
Now play that exact same game as one real person living sequentially through time, flipping the coin once a week for 100 weeks. In 50 of those weeks your wealth is multiplied by 1.5, and in 50 of those weeks your wealth is multiplied by 0.6. Each pair of one head and one tail multiplies your money by 1.5 times 0.6, which equals 0.9, a 10% loss. After 50 heads and 50 tails, your starting wealth has been multiplied by 0.9 for 50 steps in a row:
1.550 × 0.650 = 0.950 ≈ 0.005
You lose 99.5% of everything you own.
As the physicist Ole Peters demonstrated in Nature Physics in 2019, the average outcome across a crowd of 100 people at 1 snapshot in time (the ensemble average) bears no relation to the outcome for 1 person living sequentially through 100 steps in time (the time-series average). In the real world, wealth is multiplicative and bankruptcy is a trapdoor you can’t climb out of. Because human beings evolved to survive in a multiplicative world over time, we carry a deep, hard-wired instinct to demand roughly 2-to-1 odds before taking symmetric downside risks. Bringing that survival instinct into a professor’s £10 parlour game is like leaving your seatbelt on while the car is parked: it looks redundant inside the laboratory, but it keeps you alive everywhere else.
The same blindness explains Richard Thaler’s crusade against pension under-saving. In the behavioural textbook, a 27-year-old who declines to lock cash into a pension until age 68 suffers from “present bias”. In reality, a pension is a 40-year illiquid contract with a government that rewrites the tax rules every 18 months. Between 2010 and 2026, British Chancellors slashed the pension Lifetime Allowance from £1,800,000 to £1,073,100, abolished it, threatened to reinstate it, legislated to raise the minimum access age from 50 to 55 and then 57, and dragged unused pension pots into the 40% inheritance tax net. Meanwhile, £10,000 of liquid cash at age 27 carries huge option value for a house deposit, starting a business, or surviving a redundancy, whereas £10,000 locked in a pension has zero liquidity for 4 decades. Applying a steep discount rate to an illiquid 40-year promise from a broke politician is sound risk management.
Why did governments, universities, and corporations fall so hard for a discipline built on stopwatch errors, faked spreadsheets, and bad probability maths?
Because behavioural economics offered a failing governing class two irresistible gifts: policy on the cheap, and an alibi when policy failed.
In the 2010s, politicians discovered that building nuclear power stations, reforming planning laws to build houses, or fixing marginal tax traps was expensive and politically difficult. Behavioural scientists walked into Downing Street and the White House with a magical alternative, assuring ministers that poverty, obesity, energy shortages, and underfunded retirements were caused by citizens having “System 1 cognitive bugs” that could be patched for pennies by changing the wording on a government letter.
When the economists Stefano DellaVigna and Elizabeth Linos finally audited the entire output of the two largest government Nudge Units in the United States, covering 126 randomised controlled trials and 23,000,000 citizens, the miracle evaporated. Academic journals had cherry-picked tiny studies to claim an average nudge effect of 8.7 percentage points. Across all government trials, including the 87% that went unpublished, the average impact was 1.4 percentage points. Even the one policy that defenders always point to, pension auto-enrolment, relies on a massive traditional financial subsidy (mandatory employer contributions and tax relief) combined with administrative friction, rather than any psychological cure. When Maximilian Maier and his co-authors re-analysed the global nudge literature in PNAS in 2022 and corrected for publication bias, the evidence that choice-architecture nudges work at all dropped to zero.
By 2022, the failure was so undeniable that George Loewenstein, one of the founding fathers of behavioural economics, and Nick Chater, a longtime adviser to the UK’s Behavioural Insights Team, published a landmark recantation in Behavioral and Brain Sciences. Focusing on individual mental quirks, they admitted, had led public policy astray and played directly into the hands of vested interests that wanted to blame citizens for structural failures rather than fix the rules of the system.
That confession is the epitaph for a 40-year detour in modern thought. Behavioural economics flourished because it united the worst incentives of modern academia with the worst incentives of modern government. Universities rewarded researchers who p-hacked cute, counterintuitive stories about human stupidity; bureaucracies rewarded officials who blamed the public’s brain whenever a bad policy failed. The next time an expert in a lanyard tells you that the public is acting irrationally because of a cognitive bias, you should thoroughly review their spreadsheet. The human brain works fine; it was our institutions that once again failed us.