Practice confidence interval estimation
Estimate ranges for factual quantities and check how often the true value falls within your range.
Why it works
A 90% confidence interval should contain the true value 90% of the time. Most people’s 90% intervals are far too narrow — real values fall outside them 30–50% of the time, reflecting severe overconfidence in precision. Practicing interval estimation with immediate feedback recalibrates the internal sense of "how much I know," which transfers to more honest uncertainty expression in decisions.
How to do it
- Take a calibration quiz: for each factual question, state a range you’re 90% confident contains the true answer.
- After each batch of questions, score how often the true value was inside your range.
- If your 90% intervals contain the right answer less than 90% of the time, widen your ranges on the next round.
- Repeat weekly until your 90% intervals genuinely land 90% of the time — this is calibration.
Evidence
Multiple studies on probability training found that confidence interval exercises with feedback significantly reduce overconfidence in calibration tests, with effects that persist over weeks. This is among the more directly supported forms of debiasing. (observational)
Calibration improvements in training tasks do not fully transfer to novel domains; calibration is partly domain-specific and requires feedback in each domain to be reliable.
Sources
- Lichtenstein, Fischhoff & Phillips (1982), "Calibration of subjective probabilities," in Judgment Under Uncertainty: Heuristics and Biases
- Tetlock & Gardner (2015), Superforecasting — calibration training modules
Common mistake
Widening all ranges uniformly rather than learning to discriminate — a well-calibrated person has narrow ranges when they genuinely know something and wide ones when they don’t, not uniformly wide ranges.
Practice this with IX Coach
7 days free, then $40/month (~$1.30/day).
More practices for Calibration Training
- Maintain a scored prediction log
Record predictions with explicit probabilities and score them when they resolve.
- Identify the specific domains where you are most overconfident
Calibration is domain-specific — find where your confidence most exceeds your accuracy.
- Anchor your confidence level on base rates, not gut feel
Start your confidence estimate from the historical hit rate for this type of prediction, not from how confident you feel.
- Pre-commit to resolution criteria before making a prediction
Define exactly what counts as "I was right" before the outcome happens.
- Build calibration reps with low-stakes trivia and almanac questions
Use factual trivia questions as a practice ground for calibration — outcomes resolve immediately and the stakes are zero.
- Distinguish uncertainty (quantifiable) from ignorance (unquantifiable)
Know when you can assign a probability and when the situation is so novel that a number would be fabricated.