Statistics 101H: Knowledge (part 1)
Knowledge can be represented by sigma-algebras
Example of information
Game: 30 cents to buy in. Win $1 if you get the Ace of hearts. 10 cards in all. Any takers?
Suppose I tell you the denomination of the card? Any takers?
Suppose I tell you the color (red/black) of the card? Any takers?
Detail: The cards are the 4 aces and 6 other clubs.
This is called an option. Payment is max(E(Win|know)-.3,0).
We need to konw what the expected value of this option is to decide.
Three cases:
trivial knowledge: E(Win|know) = E(win). Option worth zero.
know Number: E(Win|know) = .25 or 0. Option worth zero.
know color: E(Win|know) = .5 or 0. Option worth 4 cents.
So what you know matters.
Options of this sort are studied in finance
Example of two players
Suppose 3 cards are in a "deck" (2 black and 1 red)
Each of two players get one card.
Call the remaining card the "kitty"
Computing some probabilities:
What is the expected probability that the kitty is black?
Conditional on first player seeing his card? (100%/50%)
Learning from actions:
Suppose the second player sees the first player bet that the kitty will be black?
What is her conditional probability that the kitty is black?
Suppose the second player sees the first player NOT bet that the kitty will be black?
What is her conditional probability that the kitty is black?
So the second player then KNOWS the color of the card.
Knowledge about actions
What is "her" probability that "he" will bet on black?
Representing all of this
Each players knowledge is represented by a sigma-field
Sample space contains 6 items: draw it.
Draw each sigma-field.
After we see the first player bet, the sigma-field is updated.
Last modified: Mon Dec 2 14:15:04 2002