Conditional Probability
What is a conditional probability?
•
•
•
•
•
It is the probability of an event in a subset of the sample space
Example: Roll a die twice, win if total ≥ 9
Sample space S = set of outcomes
= {11, 12, 13, 14, 15, 16, 21, 22, …, 65, 66}
Event W = pairs that sum to ≥ 9
= {36, 45, 46, 54, 55, 56, 63, 64, 65, 66}
Pr(W) = 10/36
What is a conditional probability?
•
•
•
•
•
Now suppose we know that the first roll is 4 or 5. What is now the
probability that the sum of the two rolls will be ≥ 9?
Let B = first roll is 4 or 5
= {41, 42, …, 46, 51, 52, …, 56}
Event W∩B = {45, 46, 54, 55, 56}
Pr(W | B) = |W∩B|/|B| = 5/12
“Probability of W given B”
Conditional probability
•
But since the sample space is the same,
| Ω ∩ Β | | Ω ∩ Β | / | Σ | Πρ(Ω ∩ Β)
Pr(W | B) =
=
=
| Β|
| Β| / | Σ|
Πρ( Β)
•
In general, the conditional probability of event A given event B is
defined as
Πρ( Α ∩ Β)
Pr(A | B) =
Πρ( Β)
What is the difference between
Pr(A|B) and Pr(B|A)?
•
Pr(A|B) is the proportion of B that is also within A, that is, Pr(A|B) is |
A∩B| as a proportion of |B|
A
•
B
Pr(A|B) is close to 1 but Pr(B|A) is close to 0
A∩B
CS20
•
•
•
This class has 42 students, 13 freshmen, 17 women, and 5 women
freshmen
So if a student is selected at random,
– Pr(Freshman) = 13/42,
– Pr(Woman) = 17/42
– Pr(Woman freshman) = 5/42.
If a random selection chooses a woman, what is the probability she is
a freshman?
– Simple way: #women freshmen/#women = 5/17
– Using probability:
Πρ(Ω ∩ Φ) 5 / 42
5
Pr(F | W ) =
=
=
Πρ(Ω )
17 / 42 17
Conditional Probability and Independence
•
•
Fact: A and B are independent events iff Pr(A|B) = Pr(A).
•
Proof:
That is, knowing whether B is the case gives no information that
would help determine the probability of A.
A and B independent iff Pr(A)∙Pr(B) = Pr(A∩B)
Pr(A∩B) = Pr(A|B)∙Pr(B)
So as long as Pr(B) is nonzero,
Pr(A)∙Pr(B) = Pr(A|B)∙Pr(B) iff Pr(A) = Pr(A|B)
Total Probability
•
Suppose (hypothetically!):
– Rick Santorum has a 5% probability of getting enough delegates to
become the Republican nominee, unless the voting goes beyond the
first ballot and there is a brokered convention
– In a brokered convention, Santorum has a 65% probability of winning
the nomination
•
– There is a 7% probability of a brokered convention (cf. Intrade.com)
What is the probability that Santorum will be the Republican
nominee?
Total Probability
Simple version: For any events A and B whose probability is neither
0 nor 1:
Pr(A) = Πρ( Α | Β)⋅ Πρ( Β) + Πρ( Α | Β)⋅ Πρ( Β)
That is, Pr(A) is the weighted average of the probability of A
conditional on B happening, and the probability of A conditional
on B not happening.
_
B
B
A
S
“Total probability” = weighted average of
probabilities
Pr(S) = Πρ(Σ | Β)⋅ Πρ( Β) + Πρ(Σ | Β)⋅ Πρ( Β)
•
•
•
•
Pr(Santorum|Brokered) = .65
Pr(Santorum|¬Brokered) = .05
Pr(Brokered) = .07
Then Pr(Santorum) =
.65∙.07 + .05∙.93 = .092
FINIS