2. Risk
Uncertainty
Cardinals versus Ordinals
Probability Estimation Accuracy
Suppressing Uncertainty
Foxes and Hedgehogs
Uncertainty Calibration
3. A state of uncertainty
Set of possible ?negative? outcomes
A set of possibilities with quantified
probabilities and consequences
4. Lack of complete certainty
More than one possibility
True outcome not known
A set of probabilities assigned to a set of
possibilities
5. Cardinal -> Size
Cardinal -> A number or %
Ordinal -> Position in Set
Ordinal -> High/Medium/Low or 1/2/3/4/5
Ordinals often have different meanings
depending on the person
6. An estimate of reality
The usefulness of an estimate has two
measures
Calibration
▪ How close to reality is the estimate
Discrimination
▪ The range of the estimate
7. We’re the ‘experts’
Client doesn’t want doubt
Client pushes the risk of the decision onto
you
Hitting targets for risk reduction focuses on
goals not outcomes
8. ‘The fox knows many things but the
hedgehog knows one big thing’ – Isiah Berlin
Foxes Hedgehogs
• Probability is an ad-hoc • Probability is a deductive
estimation exercise
• Low confidence in their • Highly confident in their
estimates estimates
• Tolerant of uncertainty • Intolerant of uncertainty
9. Foxes are generally better calibrated and
discriminated
Less ‘foxy’ foxes are slightly worse at
discrimination
Best overall at making short term predictions
in field of expertise
Hedgehogs worst overall when making long
term predictions in field of expertise
Hedgehogs win less but win big
10. 90% Confidence Interval (CI)
‘You are 90% certain the probability of occurrence is
between 60% and 50%’
95% sure below the upper bound
95% sure above the lower bound
Ask yourself 20 general knowledge questions asking for
numerical answers
Assign lower and upper bounds to your 90% CI of being correct
Ask yourself 20 true/false questions
Set a value to how certain you are 50%, 60%, 70%, 80% 90%, 100%
11. General Knowledge
You expected to get 9 out of 10 right (90% CI)
What did you get?
You probably got less than that right
True / False
Convert % to decimal (50% = 0.5)
Add them up
Compare to number of questions you got right
If number of correct answers was 2.5 or more less
then you are overconfident
12. Take multiple tests in succession
Equivalence Bets (next slide)
Pros and Cons
Consider two pros why you should be confident
and two cons why you could be wrong
Avoid Anchoring
95% certain below upper bound
95% certain above lower bound
13. Two choices:
1. Win £1000 if answer falls within your upper and lower
bounds.
2. Spin a dial split into 90% and 10% slices. You win £1000 if
it finishes on the 90% slice
Most people (80%) prefer to spin the dial
Ask yourself which you would prefer
Preferring 1 is over confident
Preferring 2 is under confident
Just pretending to bet money makes you better
14. After 3 tests 70% of people are perfectly
calibrated
After 3 tests 20% of people are significantly
improved
10% of people never really get it
Testing has shown that most people relied on
to make estimates are in the first two groups
already