Upload
daoud-saleem
View
219
Download
0
Embed Size (px)
Citation preview
8/14/2019 9185l_2013_Lecture2.pdf
1/62
Risk Analysis, Assessment & Management
9185 Chemical & Biochemical Engineering 2013
Housyn Mahmoud
Probability ReviewEvent Tree & Decision Tree Analysis
8/14/2019 9185l_2013_Lecture2.pdf
2/62
ObjectivesBy the end of this presentation you will
know:
Probability Requirements
Probability Approaches
Event Tree Analysis Decision Tree Analysis
8/14/2019 9185l_2013_Lecture2.pdf
3/62
Probabilities
http://www.google.ca/url?sa=i&rct=j&q=cards&source=images&cd=&cad=rja&docid=lmWNao-cJluzyM&tbnid=vf-JX3ATwW6E3M:&ved=0CAUQjRw&url=http://www.angelfire.com/anime3/db1234567/cards.html&ei=o16xUePmJ4bVyAGt9IHQAw&psig=AFQjCNEr2qflS6x61So-c4YLMAfuwOmqbQ&ust=1370664947570882http://www.google.ca/url?sa=i&rct=j&q=dice&source=images&cd=&cad=rja&docid=7c5Jqi8gsoU4wM&tbnid=z11KD5-xFg4N7M:&ved=0CAUQjRw&url=http://commons.wikimedia.org/wiki/File:Dice.png&ei=zVyxUe-vBOr1ygHw3YCwCw&bvm=bv.47534661,d.aWc&psig=AFQjCNEhjGvueoRYdDtvgQs8zKNIbBuonA&ust=13706645182795098/14/2019 9185l_2013_Lecture2.pdf
4/62
Random Experiment a random experimentis an action or
process that leads to one of several possible
outcomes. For example:
Experiment Outcomes
Flip a coin Heads, Tails
Exam Marks Numbers: 0, 1, 2, ..., 100
Assembly Time t > 0 seconds
Course Grades F, D, C, B, A, A+
8/14/2019 9185l_2013_Lecture2.pdf
5/62
ProbabilitiesList the outcomes of a random experimentList: Called the Sample Space
Outcomes: Called the Simple Events
This list must be exhaustive, i.e. ALL possible outcomes included.
Die roll {1,2,3,4,5} Die roll {1,2,3,4,5,6}
The list must be mutually exclusive, i.e. no two outcomes canoccur at the same time:
Die roll {odd number or even number}
Die roll{ number less than 4 or even number}
8/14/2019 9185l_2013_Lecture2.pdf
6/62
Sample Space A list of exhaustive [dont leave anything out]and mutually
exclusive outcomes [impossible for 2 different events tooccur in the same experiment]is called a sample spaceand is denoted by S.
The outcomes are denoted by O1, O2, , Ok
Using notation from set theory, we can represent thesample space and its outcomes as:
S = {O1, O2, , Ok}
8/14/2019 9185l_2013_Lecture2.pdf
7/62
Requirements of ProbabilitiesGiven a sample space S = {O1, O2, , Ok}, theprobabilities
assigned to the outcome must satisfy these requirements:
(1) The probability of any outcome is between 0 and 1
i.e. 0 P(Oi) 1 for each i, and
(2) The sum of the probabilities of all the outcomes equals 1i.e. P(O1) + P(O2) + + P(Ok) = 1
P(Oi) represents the probability of outcome i
8/14/2019 9185l_2013_Lecture2.pdf
8/62
Approaches to AssigningProbabilities
There are three ways to assign a probability, P(Oi), to anoutcome, Oi, namely:
Classical approach: make certain assumptions (such asequally likely, independence) about situation.
Relative frequency: assigning probabilities based onexperimentation or historical data.
Subjective approach: Assigning probabilities based on theassignors judgment. [Bayesian]
8/14/2019 9185l_2013_Lecture2.pdf
9/62
Classical ApproachIf an experiment has n possible outcomes[all equally
likely to occur], this method would assign a probabilityof 1/n to each outcome.
Experiment: Rolling a die
Sample Space: S = {1, 2, 3, 4, 5, 6}
Probabilities: Each sample point has a 1/6 chance of
occurring.What about randomly selecting a student and observing their
gender? S = {Male, Female}
Are these probabilities ?
8/14/2019 9185l_2013_Lecture2.pdf
10/62
Classical ApproachExperiment: Rolling 2 die [dice] and summing 2 numbers
on top.
Sample Space: S = {2, 3, , 12}Probability Examples:
P(2) = 1/36
P(7) = 6/36
P(10) = 3/36
1 2 3 4 5 6
1 2 3 4 5 6 7
2 3 4 5 6 7 8
3 4 5 6 7 8 9
4 5 6 7 8 9 10
5 6 7 8 9 10 11
6 7 8 9 10 11 12
What are the underlying,
unstated assumptions??
8/14/2019 9185l_2013_Lecture2.pdf
11/62
Relative Frequency ApproachBB Computer Shop tracks the number of desktop computer
systems it sells over a month (30 days):
For example,
10 days out of 30
2 desktops were sold.
From this we can construct
theestimated probabilities of an event
(i.e. the # of desktop sold on a given day)
DesktopsSold # of Days
0 1
1 2
2 103 12
4 5
8/14/2019 9185l_2013_Lecture2.pdf
12/62
Relative Frequency Approach
There is a 40% chance BB will sell 3 desktops on anygiven day [Based on estimates obtained from sample of30 days]
Desktops Sold [X] # of Days Desktops Sold
0 1 1/30 = .03= P(X=0)
1 2 2/30 = .07= P(X=1)2 10 10/30 = .33= P(X=2)
3 12 12/30 = .40= P(X=3)
4 5 5/30 =.17 = P(X=4)
= 1.00
8/14/2019 9185l_2013_Lecture2.pdf
13/62
Subjective ApproachIn the subjective approach we define probability as the
degree of belief that we hold in the occurrence of an event
P(you drop this course)
P(NASA successfully land a man on the moon)
P(girl says yes when you ask her to marry you)
8/14/2019 9185l_2013_Lecture2.pdf
14/62
Events & ProbabilitiesAn individual outcome of a sample space is called a simple
event [cannot break it down into several other events],
An eventis a collection or set of one or more simple events
in a sample space.
Roll of a die: S = {1, 2, 3, 4, 5, 6}Simple event:the number 3 will be rolled
Event:an even number (one of 2, 4, or 6) will be rolled
8/14/2019 9185l_2013_Lecture2.pdf
15/62
Events & ProbabilitiesTheprobability of an eventis the sumof the probabilities
of the simple events that constitute the event.
E.g. (assuming a fair die) S = {1, 2, 3, 4, 5, 6} and
P(1) = P(2) = P(3) = P(4) = P(5) = P(6) = 1/6
Then:
P(EVEN) = P(2) + P(4) + P(6) = 1/6 + 1/6 + 1/6 = 3/6 = 1/2
8/14/2019 9185l_2013_Lecture2.pdf
16/62
Interpreting Probability
One way to interpret probability is this:
If a random experiment is repeated an infinitenumber oftimes, the relative frequency for any given outcome is the
probability of this outcome.
For example, the probability of heads in flip of a balanced coinis .5, determined using the classical approach. The probability
is interpreted as being the long-term relative frequency of
heads if the coin is flipped an infinite number of times.
8/14/2019 9185l_2013_Lecture2.pdf
17/62
Joint, Marginal, ConditionalProbability
We study methods to determine probabilities of events that
result from combiningother events in various ways.
There are several types of combinations and relationships
between events:
Complement of an event [everything other than that event]
Intersection of two events [event A andevent B] or [A*B] Union of two events [event A orevent B] or [A+B]
8/14/2019 9185l_2013_Lecture2.pdf
18/62
Example 1
Why are some mutual fund managers more successful than others?
One possible factor is where the manager earned his or her MBA.
The following table compares mutual fund performance against the
ranking of the school where the fund manager earned their MBA:Where do we get these probabilities from? [population or sample?]
Mutual fund
outperforms the market
Mutual fund doesnt
outperform the market
Top 20 MBA program .11 .29
Not top 20 MBA program .06 .54
E.g. This is the probability that a mutualfund outperformsANDthe manager was
in a top-20 MBA program; its ajointprobability [intersection].
8/14/2019 9185l_2013_Lecture2.pdf
19/62
Example 1
Alternatively, we could introduce shorthand notation to represent theevents:
A1= Fund manager graduated from a top-20 MBA program
A2= Fund manager did not graduate from a top-20 MBA programB1= Fund outperforms the market
B2= Fund does not outperform the market
B1 B2
A1 .11 .29
A2 .06 .54
E.g. P(A2and B1) = .06
= the probability a fund outperforms the market
andthe manager isnt from a top-20 school.
8/14/2019 9185l_2013_Lecture2.pdf
20/62
Marginal Probabilities
Marginal probabilitiesare computed by adding across rows and down
columns; that is they are calculated in the marginsof the table:
B1 B2 P(Ai)
A1 .11 .29 .40
A2
.06 .54 .60
P(Bj) .17 .83 1.00
P(B1) = .11 + .06
P(A2) = .06 + .54
whats the probability a fundoutperforms the market?
whats the probability a fund
manager isnt from a top school?
BOTH margins must add to 1
(useful error check)
8/14/2019 9185l_2013_Lecture2.pdf
21/62
Conditional Probability
Conditional probabilityis used to determine how two
events are related; that is, we can determine the probability
of one event giventhe occurrence of another related event.Experiment: random select one student in class.
P(randomly selected student is male) =
P(randomly selected student is male/student is on 3rdrow) =
Conditional probabilities are written as P(A | B)and read as
the probability of A givenB and is calculated as:
8/14/2019 9185l_2013_Lecture2.pdf
22/62
Conditional Probability
Again, the probability of an event giventhat another eventhas occurred is called a conditional probability
P( A and B) = P(A)*P(B/A) = P(B)*P(A/B) both are true
Keep this in mind!
8/14/2019 9185l_2013_Lecture2.pdf
23/62
Example 2Conditional Probability
Whats the probability that a fund will outperform the market
giventhat the manager graduated from a top-20 MBA
program?
Recall:
A1= Fund manager graduated from a top-20 MBA program
A2= Fund manager did not graduate from a top-20 MBA program
B1= Fund outperforms the market
B2= Fund does not outperform the market
Thus, we want to know what is P(B1|A1)?
8/14/2019 9185l_2013_Lecture2.pdf
24/62
Conditional Probability
We want to calculate P(B1| A1)
Thus, there is a 27.5% chance that that a fund will outperform the market giventhat the
manager graduated from a top-20 MBA program.
B1 B2 P(Ai)
A1 .11 .29 .40
A2 .06 .54 .60
P(Bj) .17 .83 1.00
8/14/2019 9185l_2013_Lecture2.pdf
25/62
Independence
One of the objectives of calculating conditional probability is to
determine whether two events are related.
In particular, we would like to know whether they are
independent, that is, if the probability of one event is not
affectedby the occurrence of the other event.
Two events A and B are said to be independentif
P(A|B) = P(A)
and
P(B|A) = P(B)
P(you have a flat tire going home/radio quits working)
8/14/2019 9185l_2013_Lecture2.pdf
26/62
Independence
For example, we saw that
P(B1| A1) = .275
The marginal probability for B1is: P(B1) = 0.17
Since P(B1|A1) P(B1), B1and A1are not independent
events.
Stated another way, they are dependent. That is, the
probability of one event (B1) is affectedby the occurrence of
the other event (A1
).
8/14/2019 9185l_2013_Lecture2.pdf
27/62
Union
Determine the probability that a fund outperforms (B1)
orthe manager graduated from a top-20 MBA program (A1).
B1 B2 P(Ai)
A1 .11 .29 .40
A2 .06 .54 .60
P(Bj) .17 .83 1.00
A1or B1occurs whenever:
A1and B1occurs,A1and B2occurs, orA2and B1occurs
P(A1or B1) =.11+.06+.29 =.46
8/14/2019 9185l_2013_Lecture2.pdf
28/62
Probability Rules and Trees
We introduce three rules that enable us to calculate the
probability of more complex events from the probability of
simpler events The Complement Rule:May be easier to calculate the probability
of the complement of an event and then subtract it from 1.0 to getthe probability of the event.
P(at least one head when you flip coin 100 times)
= 1P(0 heads when you flip coin 100 times)
The Multiplication Rule: P(A*B) way I write it
The Addition Rule: P(A+B) way I write it
8/14/2019 9185l_2013_Lecture2.pdf
29/62
Example 3
A graduate Risk Assessment course has seven maleand three
Female students. The professor wants to select two students at
random to help her conduct a research project. What is theprobability that the two students chosen are female?
P(F1 * F2) = ???
Let F1represent the event that the first student is female
P(F1) = 3/10 = .30
What about the second student?
P(F2/F1) = 2/9 = .22
P(F1 * F2) =P(F1) * P(F2/F1) = (.30)*(.22) = 0.066
NOTE: 2 events are NOTindependent.
8/14/2019 9185l_2013_Lecture2.pdf
30/62
Example 4
The professor in Example 3 is unavailable. Her replacement
will teach two classes. His style is to select one student at
random and pick on him or her in the class. What is theprobability that the two students chosen are female?
Both classes have 3 female and 7 male students.
P(F1 * F2) =P(F1) * P(F2/F1) = P(F1) * P(F2)= (3/10) * (3/10) = 9/100 = 0.09
NOTE: 2 eventsAREindependent.
8/14/2019 9185l_2013_Lecture2.pdf
31/62
Addition Rule
Additionruleprovides a way to compute the probability of
eventA orBorboth A and Boccurring; i.e. the union of A
and B.
P(A or B) = P(A + B) = P(A) + P(B)P(A and B)
Why do we subtract the joint probability P(A and B) from thesum of the probabilities of A and B?
P(A or B) = P(A) + P(B)P(A and B)
8/14/2019 9185l_2013_Lecture2.pdf
32/62
Addition Rule
P(A1) = .11 + .29 = .40
P(B1) = .11 + .06 = .17
By adding P(A) plus P(B) we add P(A and B) twice.To correct we subtract P(A and B) from P(A) + P(B)
B1 B2 P(Ai)
A1 .11 .29 .40
A2 .06 .54 .60
P(Bj) .17 .83 1.00
P(A1or B1) =P(A) + P(B)P(A and B) =.40+.17- .11 =.46
B1
A1
8/14/2019 9185l_2013_Lecture2.pdf
33/62
Addition Rule for MutuallyExcusive Events
If and A and B are mutually exclusive the occurrence of one
event makes the other one impossible. This means that
P(A and B) = P(A * B) = 0
The addition rule for mutually exclusive events is
P(A or B) = P(A) + P(B)
Only if A and B are Mutually Exclusive.
8/14/2019 9185l_2013_Lecture2.pdf
34/62
Example 5In a large city, two newspapers are published, the Sun and
the Post. The circulation departments report that 22%of the
citys households have a subscription to the Sunand 35%
subscribe to the Post. A survey reveals that 6%of all
households subscribe to both newspapers. What proportion
of the citys households subscribe to either newspaper?
That is, what is the probability of selecting a household at
random that subscribes to the Sun or the Post or both?
P(Sunor Post)= P(Sun) + P(Post)P(Sunand Post)
= .22 + .35 .06= .51
8/14/2019 9185l_2013_Lecture2.pdf
35/62
Probability Trees [DecisionTrees]
Aprobability treeis a simple and effective method of applying the
probability rules by representing events in an experiment by lines. The
resulting figure resembles a tree.
First selection Second selectionThis is P(F), the probability ofselecting a female student first
This is P(F|F), the probability ofselecting a female student
second, given that a female wasalready chosen first
8/14/2019 9185l_2013_Lecture2.pdf
36/62
Probability TreesAt the ends of the branches, we calculatejoint probabilities asThe productof the individual probabilities on the preceding branches.
First selection Second selectionP(FF)=(3/10)(2/9)
P(FM)=(3/10)(7/9)
P(MF)=(7/10)(3/9)
P(MM)=(7/10)(6/9)
Joint probabilities
Sample Space:[F1*F2, F1*M2, M1*F2, M1*M2]
8/14/2019 9185l_2013_Lecture2.pdf
37/62
Probability Trees
Note: there is no requirement that the branches splits bebinary, nor that the tree only goes two levels deep, or thatthere be the same number of splits at each sub node
8/14/2019 9185l_2013_Lecture2.pdf
38/62
Example 6
Law school grads must pass a bar exam. Suppose pass rate
for first-time test takers is 72%.They can re-write if they
fail and 88% pass their second attempt
[P(pass take2/fail take 1)].
What is the probability that a randomly grad passes the bar?
[sample space?]
8/14/2019 9185l_2013_Lecture2.pdf
39/62
Example 6
P(Pass) = .72
P(Fail and Pass)=
(.28)(.88)=.2464
P(Fail and Fail) =
(.28)(.12) = .0336
First exam Second exam
8/14/2019 9185l_2013_Lecture2.pdf
40/62
Bayes Law
Bayes Law is named for Thomas Bayes, an eighteenthcentury mathematician.
In its most basic form, if we knowP(B | A),
we can apply Bayes Law to determine P(A | B)
P(B|A) P(A|B)for example
8/14/2019 9185l_2013_Lecture2.pdf
41/62
Breaking News: New test for early detection ofcancer has been developed.
LetC = event that patient has cancer
Cc = event that patient does not have cancer
+ = event that the test indicates a patient has cancer- = event that the test indicates that patient does not have cancer
Clinical trials indicate that the test is accurate 95% of the time
in detecting cancer for those patients who actually havecancer: P(+/C) = .95
but unfortunately will give a + 8% of the time for those
patients who are known not to have cancer: P(+/ Cc) = .08
8/14/2019 9185l_2013_Lecture2.pdf
42/62
Breaking News: New test for early detection ofcancer has been developed.
It has also been estimated that approximately 10% of the
population have cancer and dont know it yet:P(C) = .10
You take the test and receive a + test results.
Should you be worried? P(C/+) = ?????
8/14/2019 9185l_2013_Lecture2.pdf
43/62
What we know.P(+/C) = .95 P(+/ Cc) = .08 P(C) = .10From these probabilities we can find
P(-/C) = .05 P(-/ Cc) = .92 P(Cc) = .90
Have Cancer: C Do Not Have Cancer: CC
+
-
True State of Nature
TestR
esults
8/14/2019 9185l_2013_Lecture2.pdf
44/62
Bayesian Terminology
The probabilities P(A) and P(AC) are calledprior Probabilities
because they are determinedpriorto the decision about
taking the preparatory course.
The conditional probability P(A | B) is called aposterior
probability(or revised probability), because the priorrobability
is revised afterthe decision about taking the preparatory
course.
8/14/2019 9185l_2013_Lecture2.pdf
45/62
Students WorkBayes Problem
The Rapid Test is used to determine whether someone hasHIV [H]. The false positive and false negative rates are0.05P(+/ Hc)and 0.09P(-/H)respectively.
The doctor just received a positive test results on one oftheir patients [assumed to be in a low risk group for HIV].The low risk group is known to have a 6% P(H)probability of having HIV. What is the probability that thispatient actually has HIV [after they tested positive]. Feel
free to use a table to work this problemP(H) = 0.06 **** P(Hc) = ?
P(+/ Hc) = 0.05 **** P(-/ Hc) = ?
P(-/H) = 0.09 **** P(+/H) = ?
8/14/2019 9185l_2013_Lecture2.pdf
46/62
9185 Chemical & Biochemical Engineering 2013
Event Tree & Decision Tree Analysis
8/14/2019 9185l_2013_Lecture2.pdf
47/62
Event Tree
An event tree is a graphical representationof a series of possible events in an accident
sequence. Using this approach assumes that as each event occurs,
there are only two outcomes, failure or success. A successends the accident sequence and the postulated outcome
is either the accident sequence terminated successfully orwas mitigated successfully.
8/14/2019 9185l_2013_Lecture2.pdf
48/62
Example Event Tree
A fire starts in a plant. This is the initiating event.
Then the fire suppression system is challenged. If
the system actuates, the fire is extinguished orsuppressed and the event sequence ends. If the fire
suppression system fails then the fire is not
extinguished or suppressed and the accidentsequence progresses.
8/14/2019 9185l_2013_Lecture2.pdf
49/62
Accident Sequence
8/14/2019 9185l_2013_Lecture2.pdf
50/62
Accident Sequence Event Tree
8/14/2019 9185l_2013_Lecture2.pdf
51/62
ProbabilityPath Line
h
8/14/2019 9185l_2013_Lecture2.pdf
52/62
Event Sequence withProbability
8/14/2019 9185l_2013_Lecture2.pdf
53/62
Results
The probability derived for a fire in which the fire suppressionsystem actuates and the consequence is minimal damage isapproximately 1/1000 or 1 103.
The probability derived for a fire in which the fire suppressionsystem fails to actuate but a fire alarm signal is successfullytransmitted to the local fire department and there is onlymoderate damage is 1/100,000 or 1 105.
Finally, the probability that a fire occurs and both the firesuppression system and the fire alarm system fail and severedamage occurs is 5/10,000,000 or 5 108.
8/14/2019 9185l_2013_Lecture2.pdf
54/62
Decision Trees
Decision analysis and their associated tree analysis techniquesassume that a decision is being made under risk.
The style presented here is one that is commonly used inindustry. Influence diagrams are a close cousin to this style ofdecision tree.
Decision tree format
8/14/2019 9185l_2013_Lecture2.pdf
55/62
Example:
Decision Tree Analysis
In this scenario, you are evaluating four systems.
Table :The information about the systems.
8/14/2019 9185l_2013_Lecture2.pdf
56/62
Decision Tree 2.
8/14/2019 9185l_2013_Lecture2.pdf
57/62
Final outcomes
System A
0.3 100,000,000 = 30,000,000
0.5 25,000,000 = 12,500,000
0.2 ()4,000,000 = ()800,000 Average benefit = $41,700,000
Cost is 1,000,000
Net benefit = Benefit Cost = 41,700,000 1,000,000 = $40,700,000
8/14/2019 9185l_2013_Lecture2.pdf
58/62
Final outcomes
System B
0.3 80,000,000 = 24,000,000
0.5 35,000,000 = 17,500,000
0.2 ()1,000,000 = ()200,000 Average benefit = $41,300,000
Cost is 750,000
Net benefit = Benefit Cost = 41,300,000 750,000 =
$40,550,000
8/14/2019 9185l_2013_Lecture2.pdf
59/62
Final outcomes
System C
0.3 125,000,000 = 37,500,000
0.5 45,000,000 = 22,500,000
0.2 0 = 0 Average benefit = $60,000,000
Cost is 1,500,000
Net benefit = Benefit Cost = 60,000,000 1,500,000 =
$58,500,000
8/14/2019 9185l_2013_Lecture2.pdf
60/62
Final outcomes
System D
0.3 95,000,000 = 28,500,000
0.5 65,000,000 = 32,500,000
0.2 ()80,000,000 = ()1,600,000 Average benefit = $45,000,000
Cost is 450,000
Net benefit = Benefit Cost = 45,000,000 450,000 =$44,550,000
System C: provides the best return on the investment.Another benefit of System C is that there is no potential for a negative outcome.
8/14/2019 9185l_2013_Lecture2.pdf
61/62
Summary
Event and decision trees are highly effectivetools that have broad application in providing
visual and analytical means to betterunderstand both simple as well as complexevents.
One can use the tools to model events in
failure, as well as success space to helpbetter understand them.
8/14/2019 9185l_2013_Lecture2.pdf
62/62
Questions?