Week 8 - Class 2 Flashcards
What does it mean to be well calibrated?
That your confidence levels align closely with actual outcomes.
What is overconfidence?
Thinking you’re much better at making a judgment than you actually are.
What happens to calibration when people write down a reason supporting their selected correct answer?
Their calibration goes down.
What happens to calibration when people write
down a reason contradicting their selected
correct answer?
Their calibration goes up.
People will change their answer if they give a reason as to why their answer might be wrong. What does this tell us about how people report confidence?
They initially think of reasons why their answer is correct, not why they might be wrong.
Did Gigerenzer attack Kahneman & Tversky on normative or descriptive grounds?
Both!
What is Gigerenzer’s normative issue with Kahneman and Tversky?
Most staticians would not even view K&T’s tasks as dealing with probability. How can subjects be said to be violation probability theory?
What is the Gigerenzer’s descriptive issue with K & T?
Perhaps the mind is a frequentist. We should be able to make the errors of the tasks go away through repeat tests.
What is with Gigerenzer’s issue with the conjunction fallacy?
For a frequentist, the Linda problem is not about probability because it asks about a unique event, not a relative frequency.
What happened when they changed the Linda problem into the frequentist version? What were the results?
Relatively few conjunction fallacies to when they hadn’t framed it as a frequency.
What was Gigerenzer’s first problem with the research on overconfidence? (Hint: apples and oranges)
Because they’re comparing confidence in unique cases with frequencies. It’s like comparing apples to oranges.
In Gigerenzer’s study on overconfidence, what happened when subjects were asked after each set of 50 questions how many they thought they got right?
Overconfidence disappeared.
In Gigerenzer’s study on overconfidence, he asked how many questions they thought they got right at each confidence level. What did they say?
Subjects responded with systematically smaller values than their reported confidence, resulting in good calibration.
What was Gigerenzer’s second problem with overconfidence? (Hint: New York or Rome)
That the questions were too complicated.
What happened to calibration when the questions in an overconfidence test were “representative?”
Subjects were much better calibrated for the “representative” set than for the “selected.”
