The planning fallacy: why everything takes longer than you think

The planning fallacy is the habit of underestimating how long your own tasks will take, even when you know that similar tasks have run late before. Daniel Kahneman and Amos Tversky named it in a 1979 paper. In the best-known test of it, final-year students predicted they would hand in their theses in 33.9 days on average, and took 55.5.

The cause is where the estimate comes from. You picture this task going to plan, step by step, instead of asking how long tasks like it have actually taken. The fix the research supports is to start from that second question, even though it rarely comes to mind on its own.

Where the name comes from (and Hofstadter’s law)

Kahneman and Tversky’s paper was about forecasting in general. Its argument was that people lean on what they know about the case in front of them and neglect how cases like it usually turn out. Planning was their example: scientists and writers, they noted, are known for underestimating how long a project will take, even after plenty of experience of missing their own schedules. They called this the planning fallacy, and pointed out that it persists even when underestimating is penalised, so wishful thinking can’t be the whole story. Part of the cause is arithmetic. A building is finished on time only if the materials arrive, nobody strikes and the weather behaves. Each problem is unlikely on its own. The chance that at least one turns up is not, and intuition doesn’t add those chances together.

The same year, Douglas Hofstadter put it in a single line in Gödel, Escher, Bach, after noting that a world-champion chess computer, once forecast to be 10 years away, still seemed more than 10 years away a decade later:

“Hofstadter’s Law: It always takes longer than you expect, even when you take into account Hofstadter’s Law.”

It’s a joke, and also a fair summary of what the research went on to find: knowing about the bias doesn’t switch it off.

The thesis study

Roger Buehler, Dale Griffin and Michael Ross’s 1994 paper opens with a plain test. They phoned 37 psychology students in the last semester of the honours thesis course at the University of Waterloo and asked when, as accurately as possible, they would hand in their thesis. They also asked for a date if everything went as well as it possibly could, and one if everything went as badly. The course coordinator recorded when each thesis actually arrived.

Which estimate	Predicted, on average	Finished by that date
Best guess	33.9 days	29.7%
If everything went as well as possible	27.4 days	10.8%
If everything went as badly as possible	48.6 days	48.7%
What happened	55.5 days

The averages cover the 33 students who had finished; the other 4 still hadn’t submitted when the records stopped, 2 semesters later. Look at the third row. Asked to imagine the worst, more than half still pictured something better than what happened.

The estimates weren’t random, though: students who expected to take longer did take longer (the correlation was .77). They were shifted early, all in the same direction, which makes this a bias rather than noise.

The same paper repeated the result with smaller jobs. Students named a school task and a task outside school that they meant to finish within a week, such as fixing a bike. The school tasks were predicted at 5.8 days and took 10.7; the others were predicted at 5.0 and took 9.2. Telling students their accuracy would be checked made no difference. When another group thought aloud while estimating, 71% of what they said was about the plan, 3% was about what might get in the way, and 7% was about their own past experience.

The limits are the usual ones. These were psychology students estimating school work and chores over days or weeks, and in the everyday-task study they reported their own finishing times. Later researchers describe the effect as robust, at least for familiar tasks.

Are you better at predicting other people’s deadlines?

You’ll often read that people in this research forecast others’ timelines accurately while getting their own wrong. That isn’t quite what the paper found, and the real finding is more useful.

In its fifth study, 123 students acted as observers. Each was paired with a student from the fourth study, who had been given an hour-long computer assignment with a deadline 1 or 2 weeks away and asked when they’d finish. Observers saw that student’s deadline and prediction, plus their thoughts while planning, their record on past assignments, or both, and then made their own forecast.

The students doing the work predicted 5.5 days on average and took 6.8. The observers predicted 8.5. So the people doing the task were optimistic by about 1.3 days and the observers pessimistic by about 1.7, and neither group was more accurate overall.

What differed was where they looked. Almost half the observers (48%) mentioned the student’s past record; 2.4% of the students themselves did. Observers were also about 4 times as likely to mention things that might get in the way.

An earlier study in the paper suggests one reason. When students explained a time they had missed their own estimate, their reasons were more temporary and more specific to that project than the reasons they gave for a friend’s lateness. Your own late projects get filed as one-offs: the flu, the broken laptop, the week the in-laws stayed. A friend’s look like a pattern.

The optimism, then, belongs to the planner. Someone reading the same facts from outside reaches for the track record and the obstacles, and you can make that move about your own work.

Why tasks take longer than expected: the inside view

In a 1993 paper, Kahneman and Dan Lovallo called the two ways of forecasting the inside view and the outside view. They began with a story from 1976, when Kahneman was on a small team writing a high-school curriculum on judgement and decision-making in Israel. About a year in, he asked everyone to estimate how many months it would take to reach a complete draft. The answers ran from 18 to 30 months.

Then he asked the team’s curriculum expert to think of other teams that had built a curriculum from nothing, at the same stage: how long did they take? After a long silence, the expert said about 40% never finished. Of the rest, none that he could think of took less than 7 years or more than 10, and he didn’t think this team was any better than those. The project was finished about 8 years later.

The inside view is what everyone at that table did first: focus on the case at hand, its plan and its obstacles, and build a scenario of how it will go. The outside view sets the details aside and asks how a class of similar cases turned out. The telling part is the expert. He knew the base rate, and his first estimate still looked like everyone else’s until someone asked him the outside question.

That is why tasks take longer than expected. An inside-view estimate is a story about this task going right. Any single way of slipping may be less likely than the plan running smoothly, but there are so many ways to slip that together they can easily be the likelier outcome.

How to estimate how long something will take

The research points to an approach that feels odd at first: estimate your own task the way an outsider would. The observers overshot, so copy their method rather than their pessimism. Start from the record, then adjust for what you genuinely know about this job. In practice that’s 3 moves.

Look up the last 3 jobs like it. Find out how long the last 3 similar jobs actually took, using dates from your calendar, sent mail or file history rather than memory. The number 3 doesn’t come from a study; it’s enough to show a pattern and few enough that you’ll actually check. With no history of your own, borrow someone else’s. When it isn’t clear which jobs count as similar, Kahneman and Lovallo’s advice was to try more than one comparison.

Then make the record count. In the fourth study, students who recalled their past assignments before predicting, and were told to keep them in mind, did not do significantly better than students who just predicted: 38.1% and 29.3% finished by their predicted time. A third group had to connect the past to this task, by writing the date they’d finish if they finished as far ahead of the deadline as usual and describing a plausible way that could happen. In that group 60% finished by their predicted time, and the optimistic bias disappeared. Remembering the past wasn’t enough. Writing it into this forecast was.

So write the sentence out in full: the last 3 reports took 9, 12 and 14 days, so this one lands around day 12, probably because the review round drags again.

Unpack the steps. Next, break the task into its parts before you put a number on it. Justin Kruger and Matt Evans tested this in 2004. In one study, people estimated how many days their Christmas shopping would take, some after first listing everyone they planned to buy for. Those who made the list gave longer estimates, and that is the paper’s finding in short: unpacking reduces the planning fallacy. A list reminds you of the parts a quick mental picture skips: the setup, the waiting, the second round of edits, the end where you send it off and fix what comes back.

What you list matters, though. Constantinos Hadjichristidis and colleagues later found that spelling out unusually long parts of a task pushed estimates up, and spelling out unusually quick parts pulled them down. List all of it, especially the slow, dull parts.

Put the buffer in the plan, not the promise. The usual repair for a hopeful estimate is to pad it, but Kahneman and Tversky warned that padding rarely goes far enough: the adjusted figure stays anchored to the first one. Start from the outside-view number instead, then make the slack real. Promise a padded date while planning your days around the quick version, and the spare time gets spent elsewhere before the overrun arrives. Leave it empty in your schedule, straight after the task, where you can see it. Keep two kinds of date apart, too: the day you’ve planned to do the work is not the date it’s due, and they fail differently. A weekly review is a good place to note how long things actually took, so next time the last 3 are easy to find.

A fair worry is that a generous estimate gives you permission to drift. In the fourth study it didn’t: the students who made more realistic predictions finished at about the same time as the others. What moved finishing times was the deadline. Students given 2 weeks finished later than those given 1.

Realistic estimates are still estimates. Across the paper, the methods that removed the optimistic lean didn’t make predictions more accurate; some people now overshot, balancing those who still undershot. That trade is worth taking. A lean that always points the same way piles up across a week of tasks, while misses in both directions tend to cancel out.

When you’re late anyway

Even with good estimates, some things will run past the date you had in mind. In the second study, 80.6% of students whose school work had a deadline met it, yet only 38.7% finished by the date they’d predicted. They expected to take 5.9 days, took 11.0, and the work was due in 12.9. Missing your own estimate is the normal case, and it isn’t the same as missing the deadline.

When something does run past its date, finishing it late still counts as finishing. That’s the argument of Late is not a failure state. And the next time you catch yourself thinking this should take a week, look up the last 3 first.

Sources

• Buehler, R., Griffin, D., & Ross, M. (1994). Exploring the “planning fallacy”: Why people underestimate their task completion times. Journal of Personality and Social Psychology, 67(3), 366–381. https://doi.org/10.1037/0022-3514.67.3.366
• Hadjichristidis, C., Summers, B., & Thomas, K. (2014). Unpacking estimates of task duration: The role of typicality and temporality. Journal of Experimental Social Psychology, 51, 45–50. https://doi.org/10.1016/j.jesp.2013.10.009
• Hofstadter, D. R. (1979). Gödel, Escher, Bach: An Eternal Golden Braid. Basic Books. Quoted from the first edition, p. 152.
• Kahneman, D., & Lovallo, D. (1993). Timid choices and bold forecasts: A cognitive perspective on risk taking. Management Science, 39(1), 17–31. https://doi.org/10.1287/mnsc.39.1.17
• Kahneman, D., & Tversky, A. (1979). Intuitive prediction: Biases and corrective procedures. In S. Makridakis & S. C. Wheelwright (Eds.), Forecasting. TIMS Studies in the Management Sciences, 12, 313–327. North-Holland. Earlier version: Decision Research technical report PTR-1042-77-6 (1977), https://archive.org/details/DTIC_ADA047747
• Kruger, J., & Evans, M. (2004). If you don’t want to be late, enumerate: Unpacking reduces the planning fallacy. Journal of Experimental Social Psychology, 40(5), 586–598. https://doi.org/10.1016/j.jesp.2003.11.001