Transcript Document

In today’s lecture…
Probability
 Counting methods- Permutations &
Combinations
 Independence
 Non-independence/Bayes’ Rule

Example: Prostate Cancer Study




Thompson et al. (2006)
Prostate Specific Antigen (PSA) evaluation leads
to early detection of prostate cancer
Study looked at 5519 men who underwent
prostate biopsy
Characteristics looked at: age, race, history of
prostate cancer, previous biopsies/screening
Table 1:Racial characteristics of
study participants (n=5519)
n
%
White
5310
96.2 ~ 96
Black
209
3.8~ 4
Race
Table 2: Number of prostate cancers
and high grade prostate cancers
Probability:
Chance of something happening (from 0-1)
 0: cannot happen
 1: sure to happen
P(A) = probability that event “A” will occur
P(PSA 0-1) = probability that PSA level is from 0-1
A
P(~A) = probability that event “A” will NOT occur [complement]
P(~PSA 0-1)= probability that PSA level is NOT from 0-1
B
P(A&B)
P(A & B) = the probability that both A and B happen [joint probability]
P(PSA 0-1 & white) = the probability of being a white male with PSA 0-1
P(A|B) = the probability that A occurs, given that B occurred
[conditional probability]
P(PSA 0-1|white) = the probability that PSA is 0-1, given that the patient is white
A
*Sainani K., Stanford
P(A/B)
B
Assessing Probability
1. Theoretical/Classical probability—based on theory
(a priori understanding of a phenomena)
e.g.: theoretical probability of rolling a 2 on a standard die is 1/6
theoretical probability of choosing an ace from a standard deck
is 4/52
theoretical probability of getting heads on a regular coin is 1/2
2. Empirical probability—based on empirical data
e.g.: you toss an irregular die (probabilities unknown) 100 times and
find that you get a 2 twenty-five times; empirical probability of
rolling a 2 is 1/4
empirical probability of an Earthquake in Bay Area by 2032 is .62
(based on historical data)
empirical probability of a lifetime smoker developing lung cancer
is 15 percent (based on empirical data)
*Sainani K., Stanford
Computing theoretical
probabilities:counting methods
Great for gambling! Fun to compute!
If outcomes are equally likely to occur…
# of ways A can occur
P( A) 
total # of outcomes
Note: these are called “counting methods” because we have
to count the number of ways A can occur and the number
of total possible outcomes.
*Sainani K., Stanford
Applying our example…
P (PSA level 0-1) = (# cases with PSA 0-1)
(total number of cases)
= (1963)/(5519) = 0.35
P (PSA level >6) = (# cases with PSA >6)
(total number of cases)
= (150)/(5519) = 0.03
You randomly pick a patient to test his PSA. What’s the probability that
he is white?
P(white) = (# cases who are white)
(total number of cases)
= (5310)/(5519) = 0.96
…that he is black?
P(black) = (# cases who are black)
(total number of cases)
= (209)/(5519) = 0.04
Example 2
What’s the probability that you pick two patients who are black?
P(1st patient black) = (# cases who are black)
(total number of cases)
= (209)/(5519) = 0.038 ~ 0.04
P(2nd patient black) = (# cases who are black)
(total number of cases)
= (208)/(5518) = 0.037 ~0.04
P(black & black) = P(1st patient black) x P(2nd patient black)
= 0.0016
This is an example of joint probability…more on this coming up!
Example 3
If you have 5 patients (3 white, 2 black), and you want to
test PSA of two randomly chosen patients, what’s the
probability that they are white (W) and black (B)?
Considering order of picking,
P (1B, 1W patient) = # ways to pick one B, one W pair
# total patient pairs
Numerator = W1B1 W1B2 W2B1 W2B2 W3B1 W3B2
B1W1 B2W1 B1W2 B2W2 B1W3 B2W3 = 12
Denominator = 5x4 = 20
5 patients
P(1B, 1W) = 12/20 = 0.6
4 patients
Applying our PSA example, using a
probability tree…
Second pick
First pick
P(B)=0.04,
From our
example 2
Outcome
P(BB)=0.04*0.04 = 0.0016
P(B=.04)
P(B=.04)
P(W=.96)
P(BW)=0.04*0.96 = 0.038
P(B=.04)
P(WB)=0.04*0.96 = 0.038
P(W=.96)
P(WW)=0.96*0.96 = 0.922
P(W=.96)
P(1B,1W) = P(BW) +P(WB)
= 0.038 + 0.038
= 0.076
Rule of thumb: in probability, “and”
means multiply, “or” means add
Ignoring order of picking:
P (1B,1W patient) =
(# ways to pick one B, one W )
(total # ways to pick 2 patients)
Numerator = W1B1 W1B2 W2B1 W2B2 W3B1 W3B2 = 6
Denominator = (5x4)/2
We divide out the
order, by dividing
by 2 here
P (picking a B,W patient) =
6 = 12 = 0.6
(5x4)/2 20
Summary of Counting Methods
Counting methods for computing probabilities
Permutations—
order matters!
Combinations—
Order doesn’t
matter
With replacement
Without replacement
Without replacement
*Sainani K., Stanford
Permutations—Order matters!
A permutation is an ordered arrangement of objects.
With replacement = once an event occurs, it can occur again
(after you roll a 6, you can roll a 6 again on the same die).
Without replacement = an event cannot repeat (after you draw
an ace of spades out of a deck, there is 0 probability of
getting it again).
*Sainani K., Stanford
Permutations with replacement

Sample space: the set of all possible outcomes.
Example: in genetics, if both the mother and father carry
one copy of a recessive disease-causing mutation (d),
there are three possible outcomes (the sample space):




child is not a carrier (DD)
child is a carrier (Dd)
child has the disease (dd).
Probabilities: the likelihood of each of the possible
outcomes (always 0 P 1.0).



P(genotype=DD)=.25
P(genotype=Dd)=.50
P(genotype=dd)=.25.
*Sainani K., Stanford
Summary: order matters, with
replacement
Formally, “order matters” and “with
replacement” use powers
Equation for total number of possible outcomes:
(# possible outcomes per event)
*Sainani K., Stanford
the # of events
n
r
Example 1:
What’s the chance of having a child with the disease(dd) if both
parents are heterozygote (Dd)?
Mother’s allele
Child’s outcome
Father’s allele
P(DD) =.5*.5 = .25
P(♂D=.5)
P(♀D=.5)
P(♂d=.5)
P(Dd) = .5*.5 = .25
P(♂D=.5)
P(♀d=.5)
P(♂d=.5)
P(dD) = .5*.5 = .25
P(dd) = .5*.5 = .25
______________
1.0
P(dd) = 1 way to get (dd)
22 possible outcomes
= 1/4 = 0.25
*Sainani K., Stanford
Permutations without replacement
Example 1:
Suppose you want to test PSA levels of 4 patients: A,B,C,D.
How many ways can you test them?
C
ABCD
BACD
CBAD
DBCA
.
.
.
.
B
A
B
C
D
C
D
A
C
D
OR# permutationsD= 4x3x2x1 = 4! = 24
Reminder!
So there are 4! ways of
doing 4 tests for 4 patients
Factorial notation:
n! =n x (n-1) x (n-2) x……….x1
Example 2: What if you had 3
different tests and 5 people?
Test 3:
Test 2:
Only 4 possible
Test 1:
5 possible
only 3 possible
B
A
B
B
C
D
E
D
A
D
E
E
A
B
C
D
5 x 4 x3 
*Sainani K., Stanford
5 x4 x3x2 x1 5!
 
2 x1
2!
5!
(5  3)!
Summary: order matters,
without replacement
Formally, “order matters” and “without
replacement” use factorials
(n people or cards)!
n!

(n people or cards  r chairs or draws)! (n  r )!
or n(n  1)(n  2)...(n  r  1)
Note: This formula also worked for Example 1. We
were picking 4 people for 4 spots. So,
4!/ (4-4)! = 4!/0! = 4! = 24
*Sainani K., Stanford
Recall Permutation Theory…
If you want to see if there is a difference between
the mean PSA scores for black ( n=209) and
white patients (n=5310) in PSA example:
1. Calculate mean scores of black patients &
white patients
2. Shuffle scores of 5000 random patients,
3. Number of possible permutations of shuffling
are:
(5519!)
= A huge number of permutations
(5519-5000)!
4. Compare original mean scores to mean scores
of each permutation.
2. Combinations—Order
doesn’t matter
A combination helps determine the number of
ways “r” objects can be chosen from “n” larger
group of objects
Introduction to combination function, or “choosing”
Written as:
n
n C r or  
r
Spoken: “n choose r”
*Sainani K., Stanford
Example of combinations
If you have 3 identical tests. What are the #
of ways you can choose 3 out of 5
patients, to be tested?
= 5C 3 =
5!
3! (5-3)!
= 5 x 4 = 10
2
Example: Distinct vs.
Nondistinct objects
Suppose you want to calculate mean PSA scores of 4
patients (3 white, 1black): A (White), B (White), C (White),
& D (Black). How many ways can we arrange the 4 patients
based on race?
Total number of arrangements of 4 people = 4! = 24
However, based on race, 3 of them are identical (White- A,
B, C) and 1 of them is identical (Black- D).
If you only consider race of the patients, there will be fewer
arrangements possible…
For example:
arrangement A B C D (W W W B) =
arrangement C B A D (W W W B)
In fact, the arrangement (W W W B) can be done
in 6 distinct ways:
ABCD
ACBD
BCAD
BACD
CABD
CBAD
= 3! permutations of white patients x
1 permutation of the black patient
= 6 x1 = 6
This is one race based arrangement.
Similarly, the arrangement (W W B W) can
be done in 3!x1!=6 ways :
ABDC
ACDB
BCDA
BADC
CADB
CBDA
Since we don’t care about order, 4! ways of arranging the 4
patients is reduced to:
4!
3! 1!
=
4!
6
= 4
Hence, number of ways of arranging n objects, of which k
are white and m are black:
= n!
k! m!
( 1. White or black are just examples of being nondistinct
2. Can be extended to any number of nondistinct sets)
This is also a “choosing” problem since we are choosing 3
tests for white patients & 1 for the black patient:
4C3
= 4C1
=
=
4!
(3!)(1!)
4
Summary: combinations
If r objects are taken from a set of n objects without replacement and disregarding
order, how many different samples are possible?
Formally, “order doesn’t matter” and “without replacement”
use choosing
n!
 
 
 r  (n  r )!r!
n
*Sainani K., Stanford
Summary of Counting Methods
Counting methods for computing probabilities
Combinations—
Order doesn’t
matter
Permutations—
order matters!
With replacement: nr
Without
replacement:
Without replacement:
n(n-1)(n-2)…(n-r+1)=
n!
( n  r )!
*Sainani K., Stanford
n!
n
 
 r  (n  r )!r!
Independence
Formal definition: A and B are independent if and only if
P(A&B)=P(A)*P(B)
Going back to our Genetics example:
The mother’s and father’s alleles are segregating
independently.
P(♂D|♀D)=.5 and P(♂D|♀d)=.5
Conditional Probability: Read as
Joint Probability: The probability
of two events happening
simultaneously.
“the probability that the father
passes a D allele given that the
mother passes a d allele.”
What father’s gamete looks like is not dependent on the mother’s –
doesn’t depend which branch you start on! Marginal probability: This is the
Formally, P(DD)=.25=P(D♂)*P(D♀)
*Sainani K., Stanford
probability that an event happens at
all, ignoring all other outcomes.
On the tree
Conditional probability
Marginal probability: mother
Mother’s allele
Joint probability
Child’s outcome
Father’s allele
P(DD)=.5*.5=.25
P(♂D/ ♀D )=.5
P(♀D = .5)
P(♂d=.5)
P(Dd)=.5*.5=.25
P(♂D=.5)
P(♀d=.5)
P(♂d=.5)
P(dD)=.5*.5=.25
P(dd)=.5*.5=.25
______________
1.0
*Sainani K., Stanford
Marginal probability: father
Independent  mutually
exclusive



Events A and ~A are mutually exclusive, but they
are NOT independent.
P(A&~A)= 0
P(A)*P(~A)  0
Conceptually, once A has happened, ~A is
impossible; thus, they are completely
dependent.
*Sainani K., Stanford
Practice problem
If HIV has a prevalence of 3% in San
Francisco, and a particular HIV test has
a false positive rate of .001 and a false
negative rate of .01, what is the
probability that a random person
selected off the street will test positive?
Answer
Marginal probability of carrying
the virus.
Conditional probability: the
probability of testing + given that
a person is +
P(test +)=.99
P(+)=.03
Joint probability of being + and
testing +
P (+, test +)=.0297
P(test - )= .01
P(+, test -)=.003
P(test +) = .001
P(-)=.97
P(test -) = .999
P(-, test +)=.00097
P(-, test -) = .96903
______________
1.0
Marginal probability of testing
positive
P(test +)=.0297+.00097=.03067
P(+&test+)P(+)*P(test+)
.0297 .03*.03067 (=.00092)
 Dependent!
*Sainani K., Stanford
Law of total probability
P(test )  P(test  | HIV)P(HIV)  P(test  | HIV)P(HIV-)
One of these has to be true (mutually exclusive,
collectively exhaustive). They sum to 1.0.
P(test )  .99(.03)  .001(.97)
*Sainani K., Stanford
Law of total probability
Formal Rule: Marginal probability for event A=

P(A)  P(A | B1)P(B1)  P(A | B2 )P(B2 )    P(A | Bk )P(Bk )
k
B
i
 1.0 and P(Bi &B j )  0 (mutually exclusive)
i 1

Where:
B1
A
B2
B3
P(A)  (50%)(25%)  (0)(50%)    (50%)(25%)  25% *Sainani K., Stanford
Non independent
events/Conditional Probability
When two events are not independent, the
occurrence of one event depends on
whether the other has occurred
Bayes’ Rule
Bayes’ Rule
Definition:
Let A and B be two events with P(B)  0. The
conditional probability of A given B is:
P( A & B)
P( A | B) 
P( B)
*Sainani K., Stanford
Bayes’ Rule:
P( B / A) P( A)
P( A / B) 
P( B)
OR
P( B / A) P( A)
P( A / B) 
P( B / A) P( A)  P( B / ~ A) P(~ A)
*Sainani K., Stanford
From the
“Law of Total
Probability”
In-Class Exercise

If HIV has a prevalence of 3% in San Francisco,
and a particular HIV test has a false positive rate
of .001 and a false negative rate of .01, what is
the probability that a random person who tests
positive is actually infected (also known as
“positive predictive value”)?
*Sainani K., Stanford
Answer: using probability tree
P(test +)=.99
P(+)=.03
P (+, test +)=.0297
P(test - = .01)
P(+, test -)=.003
P(test +) = .001
P(-, test +)=.00097
P(-)=.97
P(test -) = .999
P(-, test -) = .96903
______________
1.0
A positive test places one on either of the two “test +” branches.
But only the top branch also fulfills the event “true infection.”
Therefore, the probability of being infected is the probability of being on the top
branch given that you are on one of the two circled branches above.
P( | test) 
*Sainani K., Stanford
P(test  &true)
.0297

 96.8%
P(test)
.0297  .00097
Answer: using Bayes’ rule
P(test | true) P(true)
P(true | test) 

P(test | true) P(true)  P(test | true) P(true)
.99(.03)
 96.8%
.99(.03)  .001(.97 )
*Sainani K., Stanford
Conditional probability in
epidemiology: Odds and Risk
(probability)
Definitions:
Risk = P(A) = cumulative probability (you specify the time period!)
For example, what’s the probability that a person with a high sugar
intake develops diabetes in 1 year, 5 years, or over a lifetime?
Odds = P(A)|P(~A)
For example, “the odds are 3 to 1 against a horse” means that the
horse has a 25% probability of winning.
Note: An odds is always higher than its corresponding probability,
unless the probability is 100%.
*Sainani K., Stanford
Introduction to the 2x2 Table
Exposure (E)
Disease (D)
a
No Exposure
(~E)
b
No Disease (~D)
c
d
a+c = P(E)
b+d = P(~E)
Marginal probability
of exposure
*Sainani K., Stanford
Marginal probability of
disease
a+b = P(D)
c+d = P(~D)
Coming soon…(Applications of
today’s lecture)
More on odds ratios, risk ratios
 Patterns of categorical data/distributions
 Frequency tables
 Chi square
 Logistic regression
 Kaplan Meier, survival analysis

Special thanks to Dr. Cobb for her
great slides from last year!