Week 8 - Class 1 Flashcards
What does being well calibrated mean?
Being perfectly calibrated means that you are correct x amount of time when you at the same level of confidence.
So if you’re 90% confident, you should be right 90% of the time.
What does a calibration curve show?
That people are overconfident in their answers.
Overconfidence does not go away when…
Task is one of prediction rather than general knowledge.
Subjects are warned about phenomenon.
Incentives are provided for accuracy.
To overcome overconfidence, was it better to ask what why the answer might be correct, or why it might be wrong?
Why it might be wrong.
We are better calibrated when we think of why our answer might be wrong. What does this suggest?
We typically think of reasons why we’re right, not wrong.
Why is the National Weather Service well calibrated?
- Repetitive task
- No self-fulfilling prophesies
- Formal models for combining info.
What is hindsight bias?
We perceive outcomes after the fact as more likely than before the fact.
How did the participants’ that did not know the outcome differ from those that did in the Gurka victory test?
Those that knew the correct answer perceived it as being much more likely than those that didn’t know the answer.
When the subject who knew the outcome of the event in the Gurkha Battle test, what did they say of those who did not know the “true” outcome?
They assumed they’d rate their probabilities higher.
What does the China and USSR test show us about Hindsight Bias?
Outcome knowledge influences what we think we DID know in foresight.
Outcome knowledge influences…
- What we think we would have known in foresight
- What we think others know in foresight
- What we think we DID know in foresight.
How do you reduce hindsight bias?
Subjects write down explanations as to why the “false” outcome might be true.
What are some of the commonalities between hindsight bias and overconfidence?
They both are reduced when asked to consider an alternative possibility.
