Bayes’ Theorem
P(C | A) P(A) P(C | A) P(A) + P(C | B) P(B) = P(A | C) = P(E | B) P(B)P(E | B) P(B) + P(E | A) P(A)
= P(B | E) = P(D | A) P(A) P(D | A) P(A) + P(D | B) P(B) = P(A | D) =
Bayes’ Theorem-
Example
Evaluation of Medical Screening ProcedureCost of procedure is $1,000,000
Data regarding accuracy of the procedure is: Prob (+ test result | patient has diabetes) = .90 Prob (+ test result | patient has leukemia) = .95 Prob (+ test result | patient has neither) = .07
Bayes’ Theorem-
Example
Evaluation of Medical Screening ProcedureProb (Patient has diabetes) = .02 Prob (Patient has leukemia) = .01 Prob (Patient has neither) = .97 We also know that in the general population:
Bayes’ Theorem-
Example
Evaluation of Medical Screening ProcedureWhat probabilities do we really need to know?
Prob (patient has diabetes | + test result) = ??? Prob (patient has leukemia | + test result) = ??? Prob (patient has neither | + test result) = ???
How can we find these probabilities?
Bayes’ Theorem-
Example
Evaluation of Medical Screening ProcedureProb (patient has diabetes | + test result) =
P (+ test result | diabetes) P(diabetes) P (+ test result | diabetes) P(diabetes) + P (+ test result | leukemia) P(leukemia) + P (+ test result | neither) P(neither)
Bayes’ Theorem-
Example
Evaluation of Medical Screening ProcedureProb (D | + ) = P (+| D) P(D) P (+ | D) P(D) + P (+ | L) P(L) + P (+ | N) P(N) (.90) (.02) (.90) (.02) + (.95) (.01) + (.07) (.97) = .19
Bayes’ Theorem-
Example
Evaluation of Medical Screening ProcedureProb (L | + ) = P (+| L) P(L) P (+ | L) P(L) + P (+ | D) P(D) + P (+ | N) P(N) (.95) (.01) (.95) (.01) + (.90) (.02) + (.07) (.97) = .10
Bayes’ Theorem-
Example
Evaluation of Medical Screening ProcedureProb (N | + ) = P (+| N) P(N) P (+ | N) P(N) + P (+ | L) P(L) + P (+ | D) P(D) (.07) (.97) (.07) (.97) + (.95) (.01) + (.90) (.02) = .71
Bayes’ Theorem-
Example
Evaluation of Medical Screening ProcedureTo demonstrate that these probs. are correct, look at a random sample of 100,000 people
2000 1000 97,000 1800 950 6790 200 50 90,210 100,000 9,540 90,460 D L N 90% 95% 7%
Bayes’ Theorem-
Example
Prob (+ test result | patient has diabetes) = .90 Prob (+ test result | patient has leukemia) = .95 Prob (+ test result | patient has neither) = .07
Prob (Patient has diabetes) = .02 Prob (Patient has leukemia) = .01 Prob (Patient has neither) = .97 Remember that:
and
Bayes’ Theorem-
Example
May also want to be concerned about false negatives:
Prob (Patient has diabetes | - ) = 200/90460 Prob (Patient has leukemia | - ) = 50/90460
Bayes’ Theorem-
Example
Evaluation of Medical Screening ProcedureCost of procedure is $1,000,000
Data regarding accuracy of the procedure is: Prob (+ test result | patient has diabetes) = .90 Prob (+ test result | patient has leukemia) = .95 Prob (+ test result | patient has neither) = .07
Bayes’
Theorem-Practice problem
Product Manager ProblemThere are two types of probabilities: Actual: P(superior product)
Survey results: P(Survey says superior product)
Bayes’
Theorem-Practice problem
Product manager Problem - p. 21Find P(superior product|survey says not superior) Part a) First survey is negative
Part b) Second survey is negative
Find P(superior product|survey says not superior) Using results from (a) as starting point in (b) – use condit. prob. as new marginals
Quick Review of Discrete and
Continuous Distributions
Discrete Probability Distribution
Continuous Probability Distribution Random Variable
Discrete Continuous
Binomial
Normal (and Normal Approximation to Binomial)
Random Variable
Variable that takes on a set of unique numerical values which represent every possible outcome of a random process
Discrete
Continuous
--Values are usually integers (finite number of values)
Values are real #s (infinite number of values within interval)
Probability Distribution
Discrete – Collection of probabilities that describe frequency with which random variable takes on each of its possible values
1. 0 <= P(x) <= 1 2. ΣP(xi) = 1
i
Called probability mass function
Cumulative distribution
function
Probability that x takes on values <= X F(x) = P( x <= X) =
Σ
P(xi)xi<=X
1. 0 <= F(x) <= 1 2. F(a) <= F(b) if a < b
Discrete Prob. Distribution
example
There are 20 students in a class, where the ages are distributed as follows:
Age 18 19 20 21 24 Frequency 5 6 4 4 1
Discrete Prob. Distribution
example
There are 20 students in a class, where the ages are distributed as follows:
Age 18 19 20 21 24 Frequency 5 6 4 4 1 P(x) .25 .30 .20 .20 .05 20 1.0 F(x) .25 .55 .75 .95 1.00
Discrete Prob. Distribution
example
Find
1. P(x = 19) = .30 2. P(x <= 21) = .95 3. P(19 < x <= 22) = .40 4. The mean age of students
in the class
µ= Σ xiP(xi) = 19.6
Discrete Prob. Distribution
example
To find variance: σ2 = Σ (x i-µ) 2 P(xi) i = E (xi2 ) -µ2 Where E (xi2 ) = Σx i2P(xi) (Easier to compute) iDiscrete Prob. Distribution
example
σ2 = Σ (x i-µ) 2 P(xi) σ2 =.25(18-19.6 )2 + .3(19-19.6)2+.2(20-19.6)2 +.2(21-19.6)2+.05(24-19.6)2 = 2.14 = E (xi2 ) -µ2 = 386.3-384.16 = 2.14 E (xi2 ) = Σx i2P(xi) = 182(.25) + 192(.3) + 202(.2) + 212(.2) + 242(.05) σ2 1. 2.Discrete Prob. Distribution
example
√
To find standard deviation: σ = σ2 = 2.14 = 1.46√
µ= Σ xiP(xi) = 19.6 = E (xi2 ) -µ2 = 2.14
σ2
Binomial distribution
1) Only 2 possible outcomes, success or failure 2) There are n independent trials
3) Prob. of S or F remains constant
4) Random variable represents # of successes in n trials
Binomial distribution
r = # of successes n = # of trials p = probability of success P(x=r) = Cn(p)r (1-p)n-r Where Cn= n! r!(n-r)! r r µ= np σ2= np(1-p)Binomial distribution
Quiz Problem example:r = # of successes n = # of trials = 9
p = probability of success = 0.4
P(x=3) = C9(0.4)3 (1- 0.4)9-3
3
Quiz Problem example:
Binomial distribution
Use Table C from textbook - look under n=9 and prob. = 0.4 p=0.40 r 1 .9899 2 .9295 3 .7682 4 .5174 5 .2666 6 .0994 7 .0250 8 .0003 prob. of x=> 1 prob. of x=> 2 prob. of x=> 3 prob. of x=> 4 prob. of x=> 5 prob. of x=> 6 prob. of x=> 7 prob. of x=> 8
Binomial distribution
Quiz Problem example:p=0.40 r 1 .9899 2 .9295 3 .7682 4 .5174 5 .2666 6 .0994 7 .0250 8 .0003 1. P(x=3) = P(x=>3) - P(x=>4) = .7682 - .5174 = .2508 2. P(x<5) = 1 - P(x=>5) = 1 -.2666 = .7334
Binomial distribution
Quiz Problem example:p=0.40 r 1 .9899 2 .9295 3 .7682 4 .5174 5 .2666 6 .0994 7 .0250 8 .0003 3. P(3<x<8) = P(x=>4)-P(x=>8) = .5174 - .0003 = .5171 4. µ= np = 9(.4) = 3.6
5. Most likely = value with highest probability
Normal Distribution
(Continuous)P(a <= x <= b) = F(b) - F(a) for b>a
Infinite number of values for x within an interval; values within a range
Normal Distribution: Z = x -σµ 0 1 2 3 4 -1 -2 -3 -4 N (µ, σ)
Normal Distribution
(Continuous)Production process fills 46-oz. fruit juice cans Juice filling cans is N(46.5 oz., 0.2 oz.) a) State law requires that no more than 1 in 20 cans have less than 46 oz. Does this process meet the state law?
P(x<46) <= 1/20 = .05 ????
Normal Distribution
(Continuous) Normal Distribution: Z = x -σµ 0 1 2 3 4 -1 -2 -3 -4 = 46 - 46.5 .2 = - 2.5 46.5 Look up -2.5 in Normal Table Z = 1 - .9938 = .0062 < .05Normal Distribution
(Continuous)Production process fills 46-oz. fruit juice cans Juice filling cans is N(46.5 oz., 0.2 oz.)
b) P(46.3<=x<=46.6) = ??? c) P(x=>46.8) = ???
Normal Distribution
(Continuous)Production process fills 46-oz. fruit juice cans Juice filling cans is N(46.5 oz., 0.2 oz.)
b) P(46.3<=x<=46.6) = ??? Z = x -σµ 46.3 - 46.5 .2 = -1.0 = Z = x -σµ 46.6 - 46.5 .2 = .5 =
Normal Distribution
(Continuous)Production process fills 46-oz. fruit juice cans Juice filling cans is N(46.5 oz., 0.2 oz.)
b) P(46.3<=x<=46.6) = P(-1.0<=x<=.5) = .6915 - (1-.8413) = .5328 0 1 2 3 4 -1 -2 -3 -4
Normal Distribution
(Continuous)Production process fills 46-oz. fruit juice cans Juice filling cans is N(46.5 oz., 0.2 oz.)
c) P(x=>46.8) = ??? Z = x -σµ 46.8 - 46.5
.2 = 1.5
=
P(x=>1.5) = 1 - .9332 = .0668
Normal Approximation to the
Binomial
This applies when you have a Binomial setting but for very large n values. Normal Distribution: 0 1 2 3 4 -1 -2 -3 -4 N (µ, σ) Binomial Distribution: B (n, p ) Normal Approximation: µ= np σ2= np(1-p) σ = np(1-p)
Normal Approximation to the
Binomial
Out of graduating Seniors, 80% have jobs by graduation. Take a sample of 200. a) What is the likelihood that at least 125
have jobs?
b) That between 170 and 185 have jobs? Binomial setting (S vs. F), but very large n
Normal Approximation to the
Binomial
n = 200 p = .80 µ= np = 200 (.8) = 160 σ2= np(1-p) = 200 (.8) (.2)= 32 σ= np(1-p) = 32 = 5.66 a) P(x=>125) = P(Z=> -6.27) = 1.0 Z = x -σµ 124.5 - 160 5.66 = - 6.27 =Normal Approximation to the
Binomial
This applies when you have a Binomial setting but for very large n values. Normal Distribution: 0 1 2 3 4 -1 -2 -3 -4 N (µ, σ) Binomial Distribution: B (n, p ) Normal Approximation: µ= np σ2= np(1-p) σ = np(1-p)