6 Modeling uncertainty with

probability

 ZHAN, Wenjie (Professor) 詹文杰

 School of Management, Huazhong Unive
rsity of Science & Technology
 Tel: 027-87556472
 Email: [email protected]
 6 Modeling uncertainty with
probability
 Objectives:
 In this chapter we will introduce the main
ideas and rules which are used in
probability calculations.
 Also, we will get to know some methods
for eliciting probabilities.
 6 Modeling uncertainty with
probability
 6.1 Introduction to probability
 6.2 Three approaches to probability
 6.3 Concepts and rules of probability
 6.4 The axioms of probability theory
 6.5 Methods for eliciting probabilities
6.1 Introduction to probability
 In the previous chapter we know that
uncertainty about the state of Nature is
a major factor to decision maker.

 In this chapter we will consider how the

concept of probability can be used to
provide a measure of uncertainty.
 6.1 Introduction to probability
 There are a number of ways in which uncertainty can be
measured and expressed. The simplest method involves the use of
words such as ‘unlikely’, ‘almost impossible’, ‘probable’, ‘doubtful’
and ‘expected’.
 Unfortunately, it has been found that different people attach very
different meanings to these expressions, and even individuals are
not consistent over time in their use of them.
 For example, Moore and Thomas discuss an experiment at a business
school where a large number of executives were asked to rank 10
words or phrases in decreasing order of uncertainty. The ranking of
the word ‘likely’ ranged from second to seventh, while the ranking for
‘unlikely’ ranged from third to tenth.

 Decision makers prefer a much more precise way of measuring

uncertainty. Because of this, we will be using the concept of
probability in our decision models.
 6.1 Introduction to probability
 Probabilities are measured on a scale which
runs from 0 to 1.
 If the probability of an outcome occurring is zero
then this implies that the outcome is impossible.
 At the opposite extreme, if it is considered that an
outcome is certain to occur then this will be
represented by a probability of 1 and the greater
the chances of the event occurring, the closer its
probability will be to 1.
6.1 Introduction to probability
 Throughout the book we will be using
the notation p () to mean ‘the
probability of . . .’.
 For example, we will write ‘the probability
of rain’ as p (rain).
 Definition of probability ：
basic concepts
 Probability as defined by Webster’s dictionary is “the chanc
e that a given event will occur”.

 Example 1: The probability that it will rain the next day.

 If we say that there is a 30% chance of rain, we believe that if iden
tical conditions prevail, then 3 times out of 10, rain will occur the ne
xt day.

 Example 2: The toss of a fair dice （骰子） .

 The chance of one dot is 1/6.
 Definition of probability ：
basic concepts
 Experiment （试验） : the toss of a fair dice

 Outcome/sample point （结果、样本点） : s = two dots

 Sample space （全体、样本空间） : The set of all possible outcome

s. The sample space is denoted as S.
S = {one dot, two dots, · · · , six dots}

 Event （事件） : a subset of the sample space

E = {one dot, two dots, three dots}

 Impossible event （不可能事件） : the set with no outcomes ∅

 Mutually exclusive events （互斥事件 / 不相容事件） : two events h

ave no elements in common.
E1 = {one dot, two dots}; E2 = {four dots, five dots}
 Definition of probability ：
operations on events
 Subsets of events （蕴含、包含） : A ⊂ B

 Equivalent events （相等） : A=B

Venn diagram （文氏图）
 Complement event （对立事件） :

 Mutually exclusive events （互斥事件） : two events have no elements

in common.
E1 = {one dot, two dots}; E2 = {four dots, five dots}

 Union of events （事件的和（或称并）） : A+B, A∪B

 Intersection of events （事件的积（或称交）） : AB, A∩B

 Difference of events （事件的差） :

 Definition of probability:
set theory vs. probability
 The application of set theory （集合） to probability is summarized in the
next table.

 Operational formula for event is the formula for set theory.

 Definition of probability:
properties
 Probability: the unpredictability （不可预测性） of a
n event is measured by a number that satisfies the fo
llowing axioms （公理） :
 Axiom 1. 0 ≤ P(E) for every event E
 Axiom 2. P(S) = 1 for any sample space
 Axiom 3. P (E1∪E2) = P(E1) + P(E2) for E1 and E2 mutuall
y exclusive.
 Definition of probability:
properties
 Example 1: P(E={one dot}) = 1/6

 Example 2: P(E={two dots, four dots, six dots}) = 1/6 + 1/6 + 1/6

 Example 3: What is the probability of the toss of a fair dice coming

up either less than or equal to 3 dots or equal to 3 dots?

 Example 4: What is the probability of the toss of a fair dice coming

up more than 6 dots?

 Example 5: What is the probability of the toss of a fair dice coming

up with even numbers?
 Outcomes and events
 Suppose that a company is thinking of
simultaneously launching two new products, A
and B. The company’s marketing manager
decides to list all the possible things that can
happen if the simultaneous launch goes ahead.
His list is shown below:
 Both products fail.
 Product A succeeds but B fails.
 Product A fails but B succeeds.
 Both products succeed.
 Outcomes and events
 Each of the four possible things that can
happen is called an outcome.
 An event consists of one or more possible
outcomes. For example:
 The event ‘just one product succeeds’ consists of
the two outcomes: ‘A succeeds but B fails’ and ‘A
fails but B succeeds’.
 The event ‘at least one product succeeds’ consists
of the last three outcomes in the list.
 However, the event ‘both products fail’ clearly
consists of only one outcome.
6.2 Approaches to probability
 There are three different approaches to deriving
probabilities:
 (1) the classical approach;
 (2) the relative frequency approach;
 (3) the subjective approach.
 The first two methods lead to what are often referred
to as objective probabilities because, if they have
access to the same information, different people
using either of these approaches should arrive at
exactly the same probabilities.
 In contrast, if the subjective approach is adopted it is
likely that people will differ in the probabilities which
they put forward.
 (1) The classical approach
 The classical approach to probability involves
the application of the following formula:
 (1) The classical approach

A example:

 (1) The classical approach
 In order to apply the classical approach to a problem
we have to assume that each outcome is equally likely
to occur, so in this case we would have to assume that
the customer is equally likely to select each
component.

 Of course, this would not be the case if you knew that

the customer tended to select a component from the
top of the box and you deliberately packed the
defective components in the bottom.

 In most practical situations (e.g. the simultaneous

product launch above) the outcomes will not be
equally likely and therefore the usefulness of this
approach is limited.
 (2) the relative frequency
approach
 In the relative frequency approach the
probability of an event occurring is regarded
as the proportion of times that the event
occurs in the long run if stable conditions
apply.
 This probability can be estimated by
repeating an experiment a large number of
times or by gathering relevant data and
determining the frequency with which the
event of interest has occurred in the past.
 (2) the relative frequency
approach
 Example 1:
 A quality control inspector at a factory might test
250 light bulbs and find that only eight are
defective. This would suggest that the probability
of a bulb being defective is 8/250 (or 0.032).
 The reliability of the inspector’s probability
estimate would improve as he gathered more
data: an estimate based on a sample of 10 bulbs
would be less reliable than one based on the
sample of 250.
 Of course, the estimate is only valid if
manufacturing conditions remain unchanged.
 (2) the relative frequency
approach
 Example 2:
 If the publisher of a weekly magazine found that
circulation had exceeded the break-even level in
35 out of the past 60 weeks then he might
estimate that the probability of sales exceeding
the break-even level next week is 35/60 (or
0.583).
 Clearly, for this probability estimate to be reliable
the same market conditions would have to apply
to every week under consideration; if there is a
trend or seasonal pattern in sales it would not be
reliable.
 (3) the subjective approach
 Most of the decision problems which we will consider
in decision analysis will require us to estimate the
probability of unique events occurring (i.e. events
which only occur once).
 Example:
 If a company needs to estimate the probability that a new
product will be successful or that a new state-of-the-art
machine will perform reliably, then, because of the
uniqueness of the situation, the past data required by the
relative frequency approach will not be available.
 The company may have access to data relating to the
success or otherwise of earlier products or machines, but it
is unlikely that the conditions that applied in these past
situations will be directly relevant to the current problem.
 In these circumstances the probability can be estimated by
using the subjective approach.
 (3) the subjective approach
 A subjective probability is an expression of an
individual’s degree of belief that a particular
event will occur.
 Example:
 Thus a sales manager may say: ‘I estimate that
there is a 0.75 probability that the sales of our
new product will exceed $2 million next year.’
 Of course, such a statement may be influenced by
past data or any other information which the
manager considers to be relevant, but it is
ultimately a personal judgment, and as such it is
likely that individuals will differ in the estimates
they put forward even if they have access to the
same information.
 (3) the subjective approach
 Many people are skeptical about subjective
probabilities and yet we make similar sorts of
judgments all the time.
 Some people may be concerned that subjective
probability estimates are likely to be of poor quality.
 We will introduce a number of elicitation methods to
help decision makers to make judgments about
probabilities .
 However, it is worth pointing out that such judgments
rarely need to be exact. As we shall see, sensitivity
analysis often reveals that quite major changes in the
probabilities are required before it becomes apparent
that the decision maker should switch from one
course of action to another.
 6.3 Concepts and rules of
probability
 Having looked at the three approaches
to probability, we now need to consider
the concepts and rules which are used in
probability calculations.

 These calculations apply equally well to

classical, relative frequency or subjective
probabilities.
 (1) Mutually exclusive and
exhaustive events
 Mutually exclusive event
 Two events are mutually exclusive (or
disjoint) if the occurrence of one of the
events precludes the simultaneous
occurrence of the other.
 For example, if the sales of a product in
the USA next year exceed 10 000 units,
they cannot also be less than 10 000 units.
 (1) Mutually exclusive and
exhaustive events
 Exhaustive event
 The events of ‘dollar rises against the yen
tomorrow’ and ‘the Dow-Jones index falls
tomorrow’ are not mutually exclusive: there is
clearly a possibility that both events can occur
together.
 If you make a list of the events which can occur
when you adopt a particular course of action then
this list is said to be exhaustive if your list includes
every possible event.
 (2) The addition rule
 In some problems we need to calculate the
probability that either one event or another event will
occur (if A and B are the two events, you may see ‘A
or B’ referred to as the ‘union’ of A and B).
 For example, we may need to calculate the
probability that a new product development will take
either 3 or 4 years, or the probability that a
construction project will be delayed by either bad
weather or a strike.
 In these cases the addition rule can be used to
calculate the required probability but, before applying
the rule, it is essential to establish whether or not the
two events are mutually exclusive.
 (2) The addition rule
(cont.)
In these cases the addition rule can be used to
calculate the required probability but, before applying
the rule, it is essential to establish whether or not the
two events are mutually exclusive.

If the events are mutually exclusive then the addition

rule is:

P(A U B) = P(A) + P(B) ;

(where A and B are the events)
(2) The addition rule
(cont.)
 (2) The addition rule
(cont.)
 If the events are not mutually exclusive
we should apply the addition rule as foll
ows:

p(A U B) = p(A) + p(B) − p(A and B)

 The last term has the effect of negating

the double-counting.
 (2) The addition rule
(cont.)
Table4.1 gives details of rainfall during April for the past 20 years and
also whether or not the river caused flooding. Suppose that in order to
make a particular decision we need to calculate the probability that next
year there will either be heavy rain or the river will flood.
 (2) The addition rule
(cont.)
We decide to use the relative frequency approach based on the
records for the past 20 years and we then proceed as follows:

The mistake we have made is to ignore the fact that heavy rain
and flooding are not mutually exclusive: they can and have
occurred together.
This has meant that we have double-counted the nine years
when both events did occur, counting them both as heavy rain
years and as flood years.
 (2) The addition rule
(cont.)
Thus the correct answer to our problem is:
 (3) Complementary events
 If A is an event then the event ‘A does not
occur’ is said to be the complement of A.
 The complement of event A can be written as
A (pronounced ‘A bar’).
 For example:
 the complement of the event ‘project completed
on time’ is the event ‘project not completed on
time’;
 the complement of the event ‘inflation exceeds
5% next year’ is the event ‘inflation is less than or
equal to 5% next year’.
 (3) Complementary events
(cont.)
Since it is certain that either the event or its complement
must occur their probabilities always sum to one. This leads to
the useful expression:

For example, if the probability of a project being completed

on time is 0.6, what is the probability that it will not be
completed on time?

p (not completed on time) = 1 − p (completed on time)

= 1 – 0.6
= 0.4
 (4) Marginal and conditional
probabilities
Consider Table 4.2, which shows the results of a survey of 1000
workers who were employed in a branch of the chemicals industry.
The workers have been classified on the basis of whether or not they
were exposed in the past to a hazardous chemical and whether or
not they have subsequently contracted cancer.
 (4) Marginal and conditional
probabilities (cont.)
 marginal probability
 Suppose that we want to determine the probability that a
worker in this industry will contract cancer irrespective of
whether or not he or she was exposed to the chemical.
 Assuming that the survey is representative and using the
relative frequency approach, we have:

p (worker contracts cancer) = 268/1000 = 0.268

 This probability is called an unconditional or marginal

probability because it is not conditional on whether or not
the worker was exposed to the chemical (note that it is
calculated by taking the number of workers in the margin of
the table).
 (4) Marginal and conditional
probabilities (cont.)
 conditional probability
 Suppose that now we wish to calculate the

probability of a worker suffering from cancer given

that he or she was exposed to the chemical.
 The required probability is known as a conditional

probability because the probability we are

calculating is conditional on the fact that the worker
has been exposed to the chemical.
 The probability of event A occurring given that event

B has occurred is normally written as p (A|B).

 (4) Marginal and conditional
probabilities (cont.)
 So in our case we wish to find:
p (worker contracts cancer|exposed to chemical).

 We only have 355 records of workers who were exposed to the

chemical and of these 220 have contracted cancer, so:

p (worker contracts cancer | exposed to chemical) = 220/355

= 0.620

 Note that this conditional probability is greater than the margina

l probability of a worker contracting cancer (0.268), which impli
es that exposure to the chemical increases a worker’s chances o
f developing cancer.
 (5) Independent and
dependent events
 Two events, A and B, are said to be independent
if the probability of event A occurring is
unaffected by the occurrence or non-occurrence
of event B.
 For example:
 The probability of a randomly selected husband
belonging to blood group O will presumably be
unaffected by the fact that his wife is blood group O.
 Similarly, the probability of very high temperatures
occurring in England next August will not be affected
by whether or not planning permission is granted
next week for the construction of a new swimming
pool at a seaside resort.
 (5) Independent and
dependent events (cont.)
 If two events, A and B, are independent then clearly:
p (A|B) = p (A)
 Because the fact that B has occurred does not
change the probability of A occurring.
 In other words, the conditional probability is the
same as the marginal probability.

 In the previous section we saw that the probability of

a worker contracting cancer was affected by whether
or not he or she has been exposed to a chemical.
These two events are therefore said to be
dependent.
 (6) The multiplication rule
 We saw earlier that the probability of either
event A or B occurring can be calculated by
using the addition rule. In many
circumstances, however, we need to calculate
the probability that both A and B will occur.

 For example:
 what is the probability that both the New York and
the London Stock Market indices will fall today?
 what is the probability that we will suffer strikes
this month at both of our two production plants?
 (6) The multiplication rule
(cont.)
 The probability of A and B occurring is known as
a joint probability, and joint probabilities can be
calculated by using the multiplication rule.

 Before applying this rule we need to establish

whether or not the two events are independent.
If they are, then the multiplication rule is:
p (A and B) = p (A) × p (B)
 (6) The multiplication rule
(cont.)
 Example:
 suppose that a large civil engineering company is involved in two major
projects: the construction of a bridge in South America and of a dam in
Europe.
 It is estimated that the probability that the bridge construction will be
completed on time is 0.8, while the probability that the dam will be
completed on time is 0.6.
 The teams involved with the two projects operate totally independently, and
the company wants to determine the probability that both projects will be
completed on time.

 Since it seems reasonable to assume that the two completion times are
independent, we have:
p (bridge and dam completed on time) = p (bridge completed on time)
× p (dam completed on time)
= 0.8 × 0.6 = 0.48
 (6) The multiplication rule
(cont.)
 The use of the multiplication rule is not limi
ted to two independent events.
 For example, if we have four independent
events, A, B, C and D, then:

P (A and B and C and D) = p (A)× p (B) × p (C) × p (D)

 (6) The multiplication rule
(cont.)
 If the events are not independent the
multiplication rule is:

p (A and B) = p (A) × p (B|A)

 because A’s occurrence would affect B’s

probability of occurrence. Thus we have the
probability of A occurring multiplied by the
probability of B occurring, given that A has
occurred.
 (6) The multiplication rule
(cont.)
 Example:
A new product is to be test marketed in Florida and it is estimated
that there is a probability of 0.7 that the test marketing will be a
success. If the test marketing is successful, it is estimated that there
is a 0.85 probability that the product will be a success nationally.
What is the probability that the product will be both a success in

the test marketing and a success nationally?

Clearly, it is to be expected that the probability of the product being

a success nationally will depend upon whether it is successful in

Florida.

 Applying the multiplication rule we have:

p (success in Florida and success nationally)
= p (success in Florida) × p (success nationally | success in Florida)
= 0.7 × 0.85 = 0.59
 (7) Probability trees
 As you have probably gathered by now,
probability calculations require clear thinking.

 One device which can prove to be particularly

useful when awkward problems need to be
solved is the probability tree, and the following
problem is designed to illustrate its use.
 (7) Probability trees (cont.)
 Example:
 A large multinational company is concerned that
some of its assets in an Asian country may be
nationalized after that country’s next election.
 It is estimated that there is a 0.6 probability that the
Socialist Party will win the next election and a 0.4
probability that the Conservative Party will win.
 If the Socialist Party wins then it is estimated that
there is a 0.8 probability that the assets will be
nationalized, while the probability of the
Conservatives nationalizing the assets is thought to
be only 0.3.
 The company wants to estimate the probability that
their assets will be nationalized after the election.
 (7) Probability trees (cont.)
 We first determine the probability that the
Socialists will win and the assets will be
nationalized using the multiplication rule for
dependent events:

P (Socialists win and assets nationalized)

= p (Socialists win) × p (assets nationalized | Socialists win)
= 0.6 × 0.8 = 0.48
 (7) Probability trees (cont.)
 We then determine the probability that the Co
nservatives will win and that the assets will be
nationalized:

P (Conservatives win and assets nationalized)

= p (Conservatives win) × p (assets nationalized | Conservatives win)
= 0.4 × 0.3 = 0.12
 (7) Probability trees (cont.)
 Now we can obtain the overall probability of the
assets being nationalized as follows:

p (assets nationalized)
= p (either Socialists win and nationalize
or Conservatives win and nationalize)

 These two events are mutually exclusive, since we

assume that the election of one party precludes the
election of the other, so we can simply add the two
probabilities we have calculated to get:

p (assets nationalized) = 0.48 + 0.12 = 0.60

(7) Probability trees (cont.)
 (8) Probability distributions
 So far in this chapter we have looked at how to
calculate the probability that a particular event will
occur. However, when we are faced with a decision it
is more likely that we will be concerned to identify all
the possible events which could occur, if a particular
course of action was chosen, together with their
probabilities of occurrence.

 This complete statement of all the possible events

and their probabilities is known as a probability
distribution.
 discrete probability distribution
For example, a consortium of business people who are considering
setting up a new airline might estimate the following probability
distribution for the number of planes they will be able to have in
service by the end of the year.
discrete probability distribution
 Note that the probabilities sum to 1, since all the
possible events have been listed. The ‘number of
planes in service’ is known as an uncertain quantity.
 If we plotted, on a continuous scale, the values
which this quantity could assume then there would
be gaps between the points: it would clearly be
impossible for the airline to have 2.32 or 3.2451
planes in service since the number of planes must be
a whole number.
 This is an example of what is known as a discrete
probability distribution.
continuous probability distribution
 In contrast, in a continuous probability
distribution, the uncertain quantity can take
on any value within a specified interval.
 Because continuous uncertain quantities can,
in theory, assume an infinite number of
values, we do not think in terms of the
probability of a particular value occurring.
 Instead, the probability that the variable will
take on a value within a given range is
determined.
continuous probability distribution
Figure 4.3 shows a probability distri
bution for the time to complete a con
struction project.
Note that the vertical axis of the gr
aph has been labeled probability den
sity rather than probability because
we are not using the graph to display
the probability that exact values will
occur.
The curve shown is known as a pro
bability density function (pdf).
The probability that the completion
time will be between two values is fo
und by considering the area under th
e pdf between these two points.
continuous probability distribution
 Since the company is certain that the completion
time will be between 10 and 22 weeks, the whole
area under the curve is equal to 1.
 Because half of the area under the curve falls
between times of 14 and 18 weeks this implies that
there is a 0.5 probability that the completion time will
be between these two values.
 Similarly, 0.2 (or 20%) of the total area under the
curve falls between 10 and 14 weeks, implying that
the probability that the completion time will fall
within this interval is 0.2.
 continuous probability distribution

A summary of the probability distribution is shown below.


 cumulative distribution
function
 When eliciting a probability distribution it is so
metimes easier to think in terms of the proba
bility of a variable having a value less than a
particular figure.
 This can be facilitated by deriving the cumulat
ive distribution function (cdf), which gives the
probability that a variable will have a value le
ss than a particular value.
cumulative distribution
function

Example: The cdf for th

e above project is shown i
n Figure 4.4.
It can be seen that ther
e is a 0.2 probability that
the completion time will b
e less than 14 weeks, a 0.
7 probability that it will be
less than 18 weeks and it
is certain that the time wil
l be less than 22 weeks.
 6.4 The axioms of probability
theory
 If you use subjective probabilities to express
your degree of belief that events will occur
then your thinking must conform to the
axioms of probability theory.

 These axioms have been implied by the

preceding discussion, but we will formally
state them below.
 Axiom 1: Positiveness
 The probability of an event occurring m
ust be non-negative.
 Axiom 2: Certainty
 The probability of an event which is cert
ain to occur is 1.
 Thus axioms 1 and 2 imply that the pro
bability of an event occurring must be a
t least zero and no greater than 1.
 Axiom 3: Unions
 If events A and B are mutually exclusive
then:
p (A or B) = p (A) + p (B)

Note that they are generally referred to as Kol

mogoroff’s axioms and, as stated above, they rel
ate to situations where the number of possible o
utcomes is finite.
 6.5 Methods for eliciting
probabilities
 We have seen in earlier chapters that
subjective probabilities provide a concise and
unambiguous measure of uncertainty and
they are therefore an important element of
many decision models.

 A number of techniques have been developed

to assist the decision maker with the task of
making probability judgments and in this
chapter we will examine some of the more
widely used methods.
 (1) Assessment methods for
individual probabilities
 Direct assessments:
 The simplest way to elicit a probability from a
decision maker is to pose a direct question such as
‘What is the probability that the product will achieve
a break-even sales level next month?’
 Unfortunately, many people would feel uncomfortable
with this sort of approach, and they might be
tempted to give a response without sufficient
thought.
 Asking the individual to mark a point on a scale which
runs from 0 to 1 might be preferred because at least
the scale enables the probability to be envisaged.
 The probability wheel
A probability wheel is a
device like that shown in
right figure, and it
consists of a disk with
two different colored
sectors, whose size can
be adjusted, and a fixed
pointer.
 The probability wheel
 Example: Let us suppose that a manager needs
to assess the probability that a rival will
launch a competing product within the next
week. We could adjust the wheel so that the white
sector takes up 80% of its area and ask her to
choose between the following two hypothetical
gambles:
 Bet One: If the rival launches the product within the next
week you will win $100 000. If the rival does not launch
the product you will win nothing.
 Bet Two: If, after spinning the wheel once, the pointer is in
the white sector you will win $100 000. If it is pointing
toward the black sector you will win nothing.
 Example (cont.)
 Bet One: If the rival launches the product within the
next week you will win $100 000. If the rival does not
launch the product you will win nothing.
 Bet Two: If, after spinning the wheel once, the pointer
is in the white sector you will win $100 000. If it is
pointing toward the black sector you will win nothing.
 If the manager says that she would choose Bet Two then this
implies that she thinks that the probability of the rival
launching the product in the next week is less than 80%.
 The size of the white sector could then be reduced and
the question posed again. Eventually, the manager
should reach a point where she is indifferent between
the two bets.
 For example, if this is achieved when the white sector takes up
30% of the wheel’s area, this clearly implies that she estimates
that the required probability is 0.3.
 Comments on The probability
wheel
 Advantage:
 The use of the probability wheel allowed an assessment to
be made without directly asking the manager to state the
probability. It is therefore an example of an indirect
assessment method. The wheel has the advantage that it
enables the decision maker to visualize the chance of an
event occurring.
 Disadvantage:
 However, because it is difficult to differentiate between the
sizes of small sectors, the probability wheel is not
recommended for the assessment of events which have
either a very low or very high probability of occurrence .
 The analyst should also ensure that the rewards of the two
bets are regarded as being equivalent by the decision maker.
 (2) Assessment methods for
probability distributions
 The probability method
 Graph drawing
 The probability method
 Step 1: Establish the range of values within which the decision maker
thinks that the uncertain quantity will lie.
 Step 2: Ask the decision maker to imagine scenarios that could lead to
the true value lying outside the range.
 Step 3: Revise the range in the light of the responses in Step 2.
 Step 4: Divide the range into six or seven roughly equal intervals.
 Step 5: Ask the decision maker for the cumulative probability at each
interval. This can either be a cumulative ‘less than’ distribution (e.g.
what is the probability that the uncertain quantity will fall below each of
these values?) or a cumulative ‘greater than’ (e.g. what is the
probability that the uncertain quantity will exceed each of these
values?), depending on which approach is easiest for the decision
maker.
 Step 6: Fit a curve, by hand, through the assessed points.
 The probability method
 Step 7: Carry out checks as follows.
 (i) Split the possible range into three equally likely intervals and
find out if the decision maker would be equally happy to place a
bet on the uncertain quantity falling in each interval. If he is not,
then make appropriate revisions to the distribution.
 (ii) Check the modality of the elicited distribution (a mode is a
value where the probability distribution has a peak). For example,
if the elicited probability distribution has a single mode (this can
usually be recognized by examining the cumulative curve and
seeing if it has a single inflection), ask the decision maker if he
does have a single best guess as to the value the uncertain
quantity will assume. Again revise the distribution, if necessary.
 Graph drawing
 Graphs can be used in a number of ways to elicit pro
bability distributions.
 In one approach the analyst produces a set of graphs
, each representing a different probability density fun
ction (pdf), and then asks the decision maker to sele
ct the graph which most closely represents his or her
judgment.
 In other approaches the decision maker might be ask
ed to draw a graph to represent either a probability d
ensity function or a cumulative distribution function
(cdf).
An example: history data of
score in decision analysis
Year
2011 2012 2013
Score

A: 90~100 3 12 2

B: 80~89 4 13 16

C: 70~79 1 6 10

D: 60~69 8 4 10

Failed: <60 0 3 0

Sum 16 38 38
 An example: history data of
score in decision analysis
Year
2011 2012 2013 Sum pdf cdf
Score

A: 90~100 3 12 2 17 0.1848 0.1848

B: 80~89 4 13 16 33 0.3587 0.5435

C: 70~79 1 6 10 17 0.1848 0.7283

D: 60~69 8 4 10 22 0.2391 0.9674

Failed: <60 0 3 0 3 0.0326 1.000

Sum 16 38 38 92 1.000 /
 Method of relative heights
 The method of relative heights is one well-
known graphical technique that is designed to
elicit a probability density function.
 First, the decision maker is asked to identify the
most likely value of the variable under
consideration and a vertical line is drawn on a
graph to represent this likelihood.
 Shorter lines are then drawn for other possible
values to show how their likelihoods compare with
that of the most likely value.
 Relative heights
 Example: let us suppose that a fire department has been asked
to specify a probability distribution for the number of
emergency calls it will receive on a public holiday.
 The chief administrator of the department considers that two is the
most likely number of calls. To show this, the analyst draws on a
graph a line which is 10 units long.
 Further questioning reveals that the administrator thinks that three
requests are about 80% as likely as two, so this is represented by
a line eight units long.
 The other lines are derived in a similar way, so that the likelihood
of seven requests, for example, is considered to be only 10% as
likely as two and it is thought to be extremely unlikely that more
than seven requests will be received.
 To convert the line lengths to probabilities they need to be
normalized so that they sum to one. This can be achieved by
dividing the length of each line by the sum of the line lengths,
which is 36, as shown below (note that the probabilities do not
sum to exactly one because of rounding).
Relative heights
(3) Consistency and coherence
checks
 Consistency checks are, of course, a crucial element
of probability assessment. The use of different
assessment methods will often reveal inconsistencies
that can then be fed back to the decision maker.
 Indeed, the axioms of probability theory give no
guidance as to which is the best method for the
elicitation of subjective probability.
 One way to answer this question is to use a single
method of assessing subjective probability that is
most consistent with itself. In other words, there
should be high agreement between the subjective
probabilities, assessed at different times by a single
assessor for the same event, given that the
assessor’s knowledge of the event is unchanged.
(3) Consistency and coherence
checks
 One useful coherence check is to elicit
from the decision maker not only the
probability that an event will occur but
also the probability that it will not occur.
The two probabilities should, of course,
sum to one.
(3) Consistency and coherence
checks
 Suppose that I assess the probabilities of a
set of mutually exclusive and exhaustive
events to be
0.001, 0.250, 0.200, 0.100, 0.279
 It is then pointed out to me that these
probabilities sum to 0.830 and hence that the
assessment is incoherent.
 We have to adjust the probabilities by adding
0.034 to each (= (1/5)(1 − 0.830)) to give
0.035, 0.284, 0.234, 0.134, 0.313
 Homework 5:
 Given below are the results of a survey of 100 cars:
Condition of brakes
Condition of tires Faulty Not faulty Total
Faulty 25 5 30
Not faulty 15 55 70
Total 40 60 100

 (a) Assuming that the survey is representative of all the cars on the
road, what is the probability that a car selected at random will have:
 (i) Faulty brakes;
 (ii) Faulty tires;
 (iii) Either faulty brakes or faulty tires;
 (iv) Faulty brakes given that it has faulty tires;
 (v) Faulty tires given that it has faulty brakes?
 (b) What conclusion would you draw about the relationship between
the events ‘faulty tires’ and ‘faulty brakes’?
The end