Your data never had a vote

In 1954 Paul Meehl showed that a simple formula predicted human outcomes better than seasoned clinicians did — in nearly every study he could find. His profession was furious. Fifty years and 136 studies later, the result held, and practice had barely moved.

That is the pattern. Every few decades a different science discovers something uncomfortable about how evidence meets a decision, and nothing changes. We read the whole record before starting our own study. Here is what it says, without the footnotes.

1. Decisions are not made. They collide.

Herbert Simon won a Nobel for pointing out that nobody optimises: a real mind searches until something is good enough, then stops. His students went inside companies and found worse. Search starts only when a number falls below what people expect, looks near the symptom, and stops at the first acceptable answer.

Then, in 1972, the garbage can model: problems, solutions and people drift through an organisation as separate streams, and a decision happens when they happen to collide. What decides the outcome is who was in the room when the window opened. Not the quality of the argument.

So: evidence that arrives after the window closes doesn't lose. It never played.

2. The person writing the report is making the decision

Kahneman and Tversky spent the seventies proving that the mind takes shortcuts with predictable failures. The one that matters most for business is the plainest: the same programme, described as "saves 200 of 600" or "400 will die", flips the majority's choice. Facts unchanged. Decision reversed.

Losses also hurt about twice as much as equivalent gains — and people who are losing start gambling. Keep that for finding four.

3. "Twenty years of experience" means nothing by itself

Gary Klein went to fires and found that expert commanders don't compare options at all. They recognise the situation and act. Intuition, in his data, was compressed experience, not ego.

Kahneman and Klein eventually wrote a joint paper — subtitled A Failure to Disagree — and settled when a gut call deserves trust. Two conditions. The environment must have learnable regularities: chess and anaesthesia yes, stock picking and long-range hiring no. And the person must have had years of fast, clear feedback. Without feedback, twenty years is one year repeated twenty times.

So: whether a leader's "I just see it this way" is expertise or ego is not a question about the leader. It's a question about whether their world ever told them they were wrong.

4. The real power is never in the fight

Political scientists spent the sixties arguing about who governs, and the answer that won was Bachrach and Baratz's: the deepest power is deciding what reaches the agenda. The dangerous question is simply never asked. They called it a non-decision. A research project that never finds a sponsor because its result might embarrass a vice-president is not a defeat for evidence. It is a non-decision, and nobody will call it a conflict.

Barry Staw supplied the mechanism that makes this sticky. Give people a failing project they personally chose and they invest more, not less. Public commitment binds hardest. Add Kahneman's finding that losers gamble, and a prediction falls out: evidence that lands after a leader has publicly committed doesn't get ignored. It raises the stakes.

5. Smart people are better at fooling themselves

Festinger showed in the fifties that when facts contradict a choice, people rework the facts. Kunda explained why it's invisible from inside: we build the most plausible bridge to the conclusion we want, so it feels like honest analysis.

Then Dan Kahan ran the experiment that should retire a certain kind of optimism. A two-by-two table: does the treatment work? When it was about skin cream, numerate people got it right. When the identical numbers were about gun control, the most numerate got it wrong most often — exactly when the right answer hurt their side. Skill with numbers didn't protect against motivated reasoning. It armed it.

Mercier and Sperber added the reason: reasoning evolved for winning arguments in groups, not for finding truth alone. Individuals are advocates. Groups with real disagreement are where thinking works. Fix the room, not the head.

6. "We are data-driven" is a costume, and it works

Sociologists refused to look inside anyone's head. Meyer and Rowan showed that organisations adopt structures because they look like what a serious organisation has, then quietly decouple the facade from the work. Brunsson went further: what a company says, decides and does are three separate products, and a loud decision to "become data-driven" can replace the action rather than start it.

Theodore Porter asked who reaches for numbers. Not the powerful — they can say "trust me". The people who lack authority. So "show me the data" is a map of who in the room isn't trusted, and the person allowed to decide without numbers is the one with power.

Their prediction for AI is the simplest of all. A new technology of reasoning gets absorbed by the ritual first. AI makes reports, analyses and justifications almost free, so the performance gets richer and the link to the decision need not get stronger. "We are AI-driven" is the next costume, and it is being sewn right now.

What nobody has measured

Meehl's formula has learned to talk. The research on how people treat algorithmic advice is young and split: people abandon an algorithm after one visible mistake yet defer to it on tasks they don't care about, and stop checking any system that is usually right. All of it measured on prediction tasks with a right answer.

Nobody has reconstructed real business decisions to ask: at which stage does a generative model enter — framing the question, or writing up a rationale for a choice already made? Is its output checked as hard when it agrees with the boss as when it disagrees? In a 2025 survey, close to half of three hundred US C-level leaders said they'd override a planned decision on an AI insight. Stated trust is cheap. Every finding above says revealed behaviour will look different.

That is the gap our first study, Who actually decides, walks into. We reconstruct real decisions from the people who made them and the people who produced the evidence, code the timeline, and let the types emerge from the data. Our prior, published so you can hold it against us: AI will show up at the justification stage far more than at the choice, and "data-driven versus ego-driven" will dissolve into properties of decisions rather than types of people. If the data disagree, the data win.

If you were part of a decision this year where budget, people or product direction was at stake — four minutes, one real decision, and you get it back as a mirror.

Four minutes. One real decision. Take the diagnostic →

Institute for Human Reasoning is a research and education organisation strengthening human reasoning for consequential decisions in a technology-shaped world.