In a second study from the same 1994 paper that put the planning fallacy under a stopwatch, undergraduates named a project they meant to finish in the coming week. For academic work, essays and lab reports and the like, they predicted they would be done in 5.8 days on average. They later reported finishing in 10.7 days. They completed 37.1 percent of those projects inside the predicted window. They felt, on average, 74.1 percent certain they would hit the forecast. Certainty and reality were related, but only modestly. Feeling sure was not the same as being done.
Nonacademic errands in the same study, things like fixing a bicycle or writing a letter, were predicted at 5.0 days and reported at 9.2. About 42.5 percent finished inside the guess. Average certainty was 69.9 percent. The bias was not a homework quirk.
Deadlines did not cure the early guess
A subset of 62 students had an external deadline for the academic project. Most of them, 80.6 percent, finished in time to meet that deadline. The work was due in 12.9 days on average and they reported finishing in 11.0. They had predicted they would finish in 5.9 days. Only 38.7 percent finished when they said they would. They were early in their own story and merely on time in the institution's story. The deadline saved the grade. It did not make the forecast honest. If you only track whether you hit the official date, you will keep believing the 5.9-day film, because the 11.0-day reality still counted as a win.
Roger Buehler, Dale Griffin, and Michael Ross ran that study with University of Waterloo students and published it in the Journal of Personality and Social Psychology in 1994, alongside the honors-thesis results discussed on the dog-years page. Read the limits with the means. Follow-up relied on what students reported. A few people could not be reached. The tasks were self-chosen and mostly short. None of that erases the direction of the error. It does forbid a universal "double your estimate" tattoo. Doubling 5.8 gets you near 11.6, close to the 10.7 they reported, and that coincidence is not a law. A different task will have a different ratio. The paper's own thesis sample was a different ratio again.
Recall is not a connection
The result that should change how you estimate is from their fourth study, a computer assignment with a deadline the experimenters controlled. Students who were asked to remember similar past assignments, and who admitted they typically finished about a day before the deadline, still produced optimistic forecasts when merely told to keep those memories in mind. The bias dropped when the instructions forced a link: how will those past episodes change the plan for this assignment?
The authors are blunt that people could acknowledge a history and then disassociate it from the prediction in front of them. They also note that the "connect it" group did not simply copy their typical lateness into the new guess. Typical completion was about a day before the deadline. The connected forecasts averaged earlier than that, around four days before, in the discussion of that study. Students used the past as a warning about unforeseen snags, not as a stencil. The average optimistic bias went away because long guesses and short guesses balanced. Absolute error did not shrink. You get less self-deception, not clairvoyance.
So the adult habit is not "add a fudge factor you cannot explain." The habit is a sentence that ties a named previous job to this one. "The last client memo took three evenings, not the one evening I promised, because the numbers arrived late. This memo uses the same source. I will not forecast one evening." If you cannot name the previous job, you do not yet have an estimate. You have a hope with a unit attached.
Two glossary tools, used without costume
The PMI Lexicon of Project Management Terms, version 5.0 (January 2026), names two techniques that match that habit, minus the ceremony.
Analogous estimating is estimating duration, cost, or resources from historical data on a similar activity or project. "Similar" is the load-bearing word. A memo is not similar to a migration because both involve writing. Similar means the constraints rhyme: same source of delay, same reviewer, same kind of unknown. If the historical data is "I felt busy," it is not data.
PERT, the program evaluation and review technique, estimates duration as a weighted average of optimistic, pessimistic, and most likely activity durations when the individual estimates are uncertain. The lexicon does not hand you the weights in that definition. The point it does lock in is that a single number is a bad container for uncertainty. Three numbers force the spread into the open. Your optimistic figure is the 5.9-day film. Your most likely figure should be pulled toward the connected history, not toward the film. Your pessimistic figure is the version where the late numbers happen again and the reviewer adds a pass. If you cannot stand to write the pessimistic number, you are still protecting the film.
A worked sketch, clearly invented: a recurring monthly close that last took 6 hours, 9 hours, and 7 hours, with the 9-hour month caused by a late bank file that has happened twice. Optimistic 6, most likely 8, pessimistic 12. You do not need the classic PERT weights to see that "it is a 6-hour close" is the optimistic line pretending to be the likely one. Record the three. Schedule the likely one. Tell the person waiting what the pessimistic one would do to their date. That conversation is the adjustment scope creep refuses to make. The page on scope creep is about work added in silence. This page is about hours omitted in silence.
What to track so the next estimate can see you
Buehler and colleagues found that predictions still correlated with outcomes. In the week-long projects, predicted and reported times correlated at r = .36 for academic work and r = .48 for nonacademic work. People who thought they would finish sooner than their peers tended to finish sooner than their peers, and they were still late relative to their own guess. A log is worth keeping because the rank order contains information the bias does not destroy.
Keep the log boring. For each finished piece: what you guessed, what the clock said, and one sentence on the snag you had labeled one-off. After five entries of the same kind of work, the analogous estimate is the median of the actuals, not the average of the guesses. Medians ignore a single disaster without letting you throw the disaster away. If every entry has a different "one-off," the one-off is the work.
Do not log in order to scold yourself. The 74 percent certainty in the academic sample is a warning about how conviction feels. Conviction is cheap. A median from five actuals is slightly less cheap, and it is the thing you can say out loud to someone else without acting.
Where the method stops
These studies do not say every human underestimates every task by the same factor. They do not cover surgery schedules, flight turnarounds, or a novel you have never written. A first-time task has no analogous actual. In that case the honest estimate is a range with the ignorance labeled, plus a check-in date, not a fake precision. PERT's three numbers are a way to label ignorance. They are not a spell.
If your work is billed, the estimate is also a commercial statement. This page is not legal or financial advice about what you may charge or what a contract requires you to deliver. Read the agreement. Use the log to stop promising the optimistic line as if it were the likely one.
The next time you type a single duration into a plan, can you point at the finished job that justifies it, or are you narrating the version where nothing waits on you?
Studies and definitions
- 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. doi.org/10.1037/0022-3514.67.3.366
- Project Management Institute. (2026). PMI Lexicon of Project Management Terms (Version 5.0). Entries for analogous estimating and program evaluation and review technique (PERT). PMI lexicon PDF