Suppose you must choose between action A and Action B
when the probability of state S is p and the probability of state T is
1-p.
| State S
Pr(S)=p |
State T
Pr(T)=1-p |
Expected
Utility |
![]() |
|
| Action A | AS | AT | p*AS +(1-p)*AT | |
| Action B | BS | BT | p*BS +(1-p)*BT |
Without loss of generality, assume outcome AS is better
than BS,
the only nontrivial case is then when outcome BT is better
than AT.
| Action A better than
Action B
p*AS +(1-p)*AT > p*BS +(1-p)*BT
p > BT
- AT
|
Action B better than
Action A
pAS +(1-p)AT < pBS +(1-p)BT
p <BT
- AT
|
| Let W = | BT - AT | = | advantage of B given T | = | regret of A given T |
| AS - BS | advantage of A given S | regret of B given S |
W= "Critical Odds"
Note that W is a pure value judgment, compltely idepndent of any consideration
of the probability of stae S ot T
The more regrettable A would be if T happens, the higher the probability of S needs to be to justify choosing A.
CONVERTING FROM CRITICAL ODDS TO CRITICAL PROBABILITY
| p/ (1-p)
> W
P > (1-P)W P > W - PW P + PW > W p > W/(1+W)
|
p/ (1-p)
< W
P < (1-P)W P < W - PW P + PW < W p < W/(1+W)
|
,