Conditional Probability and Bayes Theorem

CONDITIONAL PROBABILITY AND BAYES THEOREM
HOT NOTES FOR STATISTICS
Abstract. Define conditional probability and the multiplication rule, and
show how Bayes Theorem works.
1. Conditional Probability
Condition probability, written Pr(A | B) is the probability of event A, given the
knowledge that event B has occurred. For example, in this compilation of Indiana
bicycle accident injuries, where H indicates wearing a helmet and I indicates a
head injury,
head injury
yes
no
total
wearing helmet
yes no total
17 218
235
130 428
558
147 646
793
235
, while the conditional
the marginal probability of a head injury is
793
probability of a head injury, given the fact that the subject was not wearing
218
a protective helmet, is
In the first case, all cyclists are considered (the
646
denominator is the grand total); in the second case only this without helmets are
considered (a column total).
Mathematically, the conditional probability for this event is defined as
Pr(I | ¬H) =
Pr(I ∩ ¬H
Pr(¬H)
2. Multiplication Rule
Pr(I ∩ ¬H) = Pr(¬H) × Pr(I | ¬H)
1
2
HOT NOTES FOR STATISTICS
2.1. Bayes Theorem. Using the multiplication rule and the Law of Total
Probability, the definition of conditional probability can be expanded to give Bayes
Theorem
Pr(B | A) Pr(A)
Pr(A | B) =
Pr(B | A) Pr(A) + Pr(B | ¬A) Pr(¬A)
For example, from the head injury data, we see that
Pr(¬H | I)
=
=
Pr(I | ¬H) Pr(¬H)
Pr(I | ¬H) Pr(¬H) + Pr(I | H) Pr(H)
218/646 × 646/793
218
=
218/646 × 646/793 + 17/147 × 147/793
235
3. A Diagnostic Example
Consider an HIV test with the following sensitivity, specificity, and prevalence:
sensitivity
Pr(+ | HIV )
0.999
specificity Pr(− | ¬HIV )
0.995
prevalance
Pr(HIV )
0.0025
The possible test outcomes are shown in this decision tree
Using Bayes’ Theorem, we can find the posterior predictive value, Pr(HIV | +)
as
PPV
=
=
sensitivity × prevalence
sensitivity × prevalence + (1 − specificity × (1 − prevalence)
0.999 × 0.0025
= 0.3337
0.999 × 0.0025 + 0.005 × 0.0075
Department of Management Science and Statistics, UTSA
E-mail address: [email protected]