Bolt Series · Human Performance Calibration · Article 07
The dangerous part is not being wrong
A student predicts 88%.
The paper returns at 51%.
That is a large error.
But the most important question is not:
“How could you be so overconfident?”
It is:
“What information made 88 feel reasonable to you?”
That question changes the problem.
Overestimation stops being a personality insult.
It becomes a measurement problem.
Research finds overestimation often enough that we should take it seriously
Studies of metacognitive calibration repeatedly find that some students predict substantially better performance than they later produce.
In one study following students across four in-class examinations, lower-performing students continued to overpredict their grades and remained confident in those predictions even after earlier tests had supplied contrary evidence.
Another study in introductory biology found that many of the lowest-performing students continued to overestimate their knowledge even after receiving feedback from practice tests.
And a physical-education study across three experiments found an overconfidence effect when students predicted performance on sport tasks, with higher performers generally showing better calibration.
That pattern is interesting.
But we need to interpret it carefully.
Do not turn calibration research into “stupid people do not know they are stupid”
That popular version is much too crude.
A person can misjudge themselves for many reasons.
- They may not yet know which features of the task matter.
- They may mistake familiarity for mastery.
- They may remember successful examples more readily than failed ones.
- Their previous performance may be highly variable, giving them a noisy baseline.
- They may know a procedure but not recognise when it fails.
- They may judge effort rather than correctness.
- They may receive feedback too vague to update their internal model.
- The task itself may differ substantially from what they practised.
Research on classroom calibration has found that variability in prior performance can itself be associated with poorer prediction accuracy. If your recent evidence about yourself jumps around, building a stable estimate is harder.
So overestimation does not automatically tell us what kind of person we are dealing with.
It tells us that the person’s current internal estimate and the observed performance do not agree.
Then we investigate why.
Familiarity is one of education’s great traps
A student looks at a worked example.
“Yes, yes. I know this.”
The page feels familiar.
The method makes sense while somebody else is doing it.
Then the book closes.
A new question appears.
Nothing happens.
The original feeling was real.
But it was evidence of recognition, not necessarily evidence of independent retrieval, transfer or execution.
This is why performance receipts matter.
You do not discover whether you can independently solve the problem by asking whether the solution looks familiar.
You discover it by attempting the problem without the solution in front of you.
Bolt did not improve his start by believing his start was good
Usain Bolt was capable of extraordinary overall sprint performances while still identifying the beginning of his 100 metres as an area for improvement.
He spoke publicly about the first 30 to 40 metres, his start and acceleration as weaknesses relative to the later part of his race.
He also acknowledged that his coach could see details he could not see himself.
That is exactly the opposite of defensive overconfidence.
Overall greatness did not require pretending every component was already great.
In fact, improvement depended on making the weak component visible enough to work on.
Overestimation becomes expensive when it controls decisions
Imagine a student who thinks a topic is mastered when it is not.
What happens next?
- They stop revising too early.
- They select work that is too difficult or skip foundations they still need.
- They reject feedback because it conflicts with their self-estimate.
- They blame isolated mistakes instead of noticing a repeated mechanism.
- They arrive at the examination with a preparation plan built on a false state estimate.
The problem is not merely emotional.
A poor internal estimate can generate poor actions.
The cure is not humiliation
If a learner predicts 90 and receives 50, saying “See? You are not as good as you think” may win the argument and lose the educational opportunity.
Better:
- Record the prediction before the attempt.
- Record the result.
- Locate where the prediction and result separated.
- Ask which cues the learner relied on.
- Test the same capability again under slightly different conditions.
- Look for a pattern across several receipts.
The child needs enough evidence to update.
And sometimes the adult needs enough evidence to update too.
Remember the earlier Bolt principle: no single observer owns reality.
A surprising benefit of discovering you were overconfident
If handled properly, a large prediction error can be the moment a learner begins to understand themselves at higher resolution.
“I thought recognising the chapter meant I knew it.”
“I thought getting homework right with examples beside me meant I could do it independently.”
“I thought because I understood the easy questions, I understood the topic.”
Those are powerful discoveries.
Now future predictions can use better evidence.
The goal is not to make the learner think less of themselves. The goal is to make their estimate of themselves more useful.
Sometimes that means the estimate must come down.
Not as punishment.
As calibration.
Earlier in the Bolt Series
Evidence and further reading
- Low-Performing Students Confidently Overpredict Their Grade Performance throughout the Semester
- Persistent Miscalibration for Low and High Achievers despite Practice Test Feedback
- Performance Calibration Across Sport Tasks
- Metacognitive Errors in the Classroom: Variability of Past Performance and Exam Prediction Accuracy
- World Athletics — Bolt on his start, acceleration and what his coach could see