In 1994, Roger Buehler, Dale Griffin, and Michael Ross asked 37 psychology students a question with an obvious answer: how long will your senior thesis take? The students were not naive. They knew their own history of late nights and missed self-imposed deadlines, and they had watched classmates grind through the same process a year ahead of them. They still produced an average estimate of 33.9 days. Then the researchers pushed once more, asking for two edge cases that would turn out to be the most interesting part of the study.
🧠 The Worst Case They Could Imagine
If everything went as well as it possibly could, the students said, the thesis would be done in 27.4 days. If everything went as badly as it possibly could — every source delayed, every draft returned, every week lost — 48.6 days.
The average thesis took 55.5 days.
Read that again with the two numbers side by side. The students were not merely optimistic about the good scenario. They overshot even the catastrophe they had summoned for the sake of the question, the version where everything goes wrong. Roughly 30 percent of them finished inside the time they originally predicted.
A follow-up version of the work made the miscalibration harder to wave off. Students were asked to name the dates by which they were 50 percent, 75 percent, and 99 percent certain their academic projects would be finished. By the 50 percent date, 13 percent were done. By the 75 percent date, 19 percent were done. By the date they attached 99 percent certainty to, fewer than half — 45 percent — had finished.
🤔 The Version That Survives Knowing
Here is the part that makes this more than a story about students being students. A survey of Canadian taxpayers, published in 1997, found that people mailed their returns about a week later than they predicted. They had no illusions about their own record. They remembered, accurately and without prompting, that they file late every single year. They simply expected that this year would be different, and this expectation held while the memory of the past sat right next to it.
That is the defining feature of the planning fallacy, and it is why the usual advice slides off. Knowing about a bias and being able to correct for it are two separate skills, and this one is unusually resistant to the first. People acknowledge that their past predictions ran optimistic, and simultaneously insist that the current one is realistic.
It also runs in a strange direction. The bias applies to your own tasks. Ask someone to estimate how long a colleague’s project will take and the error flips sign: outsiders are reliably pessimistic, guessing longer than reality. Same task, same information, opposite error, depending on whose hands are on it.
Then there is an experiment that reframes the whole thing. When people made their time predictions anonymously, the optimistic bias disappeared. The estimates got realistic the moment no one was watching. Whatever else is happening here, part of it is a performance put on for an audience. The schedule you present is a social object, and social objects are supposed to look competent.
The cognitive explanations do the rest of the work. People plan by simulating the task and skip the interruptions, so the simulation is mostly the task’s own logic with no traffic, no illness, no bad week. When the plan does run late, the delay gets attributed to circumstance while the easy parts get claimed as skill, which quietly discounts the past as evidence. People also misremember how long old tasks took, in the direction of shorter. Every input into the estimate leans one way.
And there is a colder mechanism that has nothing to do with cognition. Much project planning happens inside an approval process, where the planner needs a budget signed off and knows a realistic number might not survive the meeting. Underestimating on purpose is a strategy, and it works, because it is easier to get forgiveness for an overrun than permission to start. The planning fallacy in that room is not a bug in anyone’s brain. It is an incentive.
Which is why the fixes are less satisfying than they should be. Breaking a task into subtasks and estimating each one does help, measurably, in three separate experiments. It also demands a lot of effort, more than most people can spend on an ordinary Tuesday. Writing out exactly when and where you will do the work has a curious two-stage effect: it initially makes your predictions even more optimistic, then reduces the bias over time, mostly because you start earlier and get interrupted less. Nothing here is a switch. The bias is built into how the future gets imagined, and the imagining is not optional.
🔗 Stop Estimating, Start Checking
The practical version is dull and it works. Do not ask how long this will take. Ask how long the last one took, and look it up.
That distinction matters more for creative work than for anything else, because a chapter is not a task with a duration. A chapter is a task plus everything that happens between sessions of it. The scene that will not land, the note from a reader that turns out to require rebuilding an earlier chapter, the afternoon with the wrong kind of attention. None of that shows up in the estimate, because when you picture the work ahead you picture the work, not the week.
So the log matters more than the plan. Written-down actuals are honest in a way that written-down intentions never are, and a personal log of how long things really took is the closest thing to a calibration instrument most people will ever build. Anyone running several projects at once — a site, a product, a novel, a day job — has likely already noticed that the deadline that slipped was always the one estimated from the task rather than from the record.
There is a second-order version worth watching, too. Products that promise order do the same work as the optimistic estimate: they take a messy week and represent it as a shape. That is useful and often kind. It becomes a problem only when the representation is treated as a commitment, and the person holding it concludes that the slippage is theirs to explain.
🎲 The Opera House and the Law
The Sydney Opera House was budgeted at seven million dollars and scheduled to open in 1963. A scaled-down version opened in 1973, ten years late, at a cost of 102 million. The Big Dig in Boston was approved in 1988 with a 2.8 billion dollar budget and finished at 8.08 billion, seven years behind. Denver’s airport opened sixteen months late. Berlin’s new airport was set to open in October 2011 and opened on October 31, 2020. The James Webb Space Telescope launched fourteen years after its original date, roughly nine billion dollars over budget. When these are the outliers, planners can argue about competence. When nearly every one of them runs long, the pattern is not in the projects.
Which brings in the small law worth keeping: it always takes longer than you expect, even when you take into account that it always takes longer than you expect.
Here is a test you can run today. Pick one small task you are certain you can finish today. Write down your estimate, then write down a time you would bet everything on. Do the task. Compare the three numbers. The distance between your estimate and your actual is your coefficient, and it will be much easier to remember than this essay the next time you plan something.