Module 3 - Probabilities

INSTRUCTIONAL MATERIAL
Module 3: Probability
Overview
This module explores some of the more common and simpler methods of quantitative
analysis in probability. It is the aim of this module that you will develop your practical and
transferable skills in this area of mathematical modelling as well as developing cognitive abilities
in effective model selection and justification. To achieve this process requires you to think about
modelling as a specific decision-making activity and set of practical and cognitive skills to
develop.
Module Objectives:
Upon completion of this module you will be able to:
 Understand the distribution functions and queuing theory

 Perform distribution functions and queuing theory and conduct corresponding analysis
Topic Outline
Lesson 2
A. Developing Rational Models with Quantitative Methods and Analysis: Probability
B. The Decision Tree
C. Decision Analysis
D. Expected Monetary Values
E. Bayes Theorem
F. The Value of Sample and Perfect Information
References and Resources:

Watch or read as applicable the following provided links to help you understand the topics in
this module:
 Effective Management Decision Making: An Introduction by Ian Pownall, ©2012 (eBook),

pp. 100 – 128.
 Youtube video lecture - Math Antics - Basic Probability https://www.youtube.com/watch?
v=KzfWUEJjG18
 Youtube video lecture – Decision Tree Tutorial in 7 minutes with Decision Tree Analysis
& Decision Tree Example (Basic) https://www.youtube.com/watch?v=a5yWr1hr6QY
 Youtube video series – Decision Analysis https://www.youtube.com/watch?v=NQ-
mYn9fPag&list=PLD3fYc0bAjC-FbzlruUEYIO-El4PjDVCv
 Youtube video lecture – Bayes’ Theorem: Hate it or Love it, can't ignore it!
https://www.youtube.com/watch?v=bUI8ovd07uI
MANA 4033 – Decision Making
Abelardo R. dela Cruz 1
 Youtube video - Introduction to Bayesian data analysis - part 1: What is Bayes?

https://www.youtube.com/watch?v=3OJEae7Qb_o

Lesson 2
Developing Rational Models with Quantitative Methods and Analysis:

Probability
The availability of measurable data and its reliability is perhaps not as common as
managers would wish it was. As a result, we need to begin to consider decision-making
methods which are able to include greater levels of uncertainty in the process. In particular,
there is a focus upon the individual’s view of the uncertainty.
It is the aim of this module that you will develop your practical and transferable skills in this
area of mathematical modelling as well as developing cognitive abilities in effective model
selection and justification. To achieve this process requires you to think about modelling as a
specific decision-making activity and set of practical and cognitive skills to develop.
Modelling is described as the process through which reality can be understood. Models of
reality are nothing more than simplifications of reality.
 They allow the modeler to understand situations and environments cost effectively and
with less risk, than taking an actual decision or intervention in that environment.
 The modeler chooses to focus upon particular views of data to understand something
that is occurring or will occur.
 The modeler will have to be clear as to why they have chosen one particular view over
another.
There are four generic forms of modelling:
1) Iconic – where the modeler creates a physical representation of the process or entity
to be understood. These might be scalar clay representation of cars, or mock ups of
a product.
2) Analog models – are also physical form, but do not physically resemble the object
being modelled. Perhaps the most well-known example is the use of water to
represent the five-sector circular flow of money in an economy.
3) Virtual models – follow from analog models, but rather that create physical models,
a computer-generated model is constructed.
4) Mathematical models – represent real world problems through a system of
mathematical formulas and expressions based on key assumptions, estimates, or
statistical analyses. This modelling is the focus of this module.
There are three distinct stages in the development of mathematical and logical models and
probabilities or models of higher levels of uncertainty regarding the confidence of data that is
available.
1. Model Development is concerned with determining what are the key data upon which
we can apply a method or develop a policy or understanding that allows us to make

better informed decisions in the future. Cost and benefit considerations will also shape
the appropriate selection of data use in a given mathematical model.
2. Data preparation is concerned with generating data in an appropriate form as
depending upon the model developed, it might be that data is not in the required form or
units, to be immediately utilized. You may therefore need to progress through a series of
transformations, to be able to apply a given method to a dataset. This is explored with
data forecasting.
3. Model solution is important for the manager/modeler to also be aware of the
mechanistic limitations of their model, so that the solution generated through its
application to a dataset, will also have constraints and assumptions that limit the
solution’s effectiveness and/or validity.
The choice of a decision model requires a cost and benefit consideration to be made in
selecting an appropriate set of mathematical relationships where frequently a less complicated
model is more appropriate than a more complex and accurate one due to cost and ease of
solution considerations. Decision variables and decision body can also increase the complexity
of the model chosen and the data selected. The model’s solution therefore is always seeking to
maximize or minimize a give event or outcome, subject to the constraints of the model chosen.
 The values of the decision variables that provide the mathematically-best outputs are
referred to as the optimal solution of the model.
 In the area of linear programming which is concerned with resource efficiency use, it is
important to attempt to identify the optimizing solution from a range of viable solutions.
Mathematical models are deterministic when concerned with discrete events and outcomes.
 Where events and outcomes are subject to variations in inputs that shape the decision-
making process, they are called stochastic models.
 Such models are more problematic for managers and difficult to generalize from.
 Perhaps the most famous of such model in studies of Business and Management is that
of Gibrat’s Law.
o According to the Financial Dictionary, “law of proportionate effect (or Gibrat
process) an explanation of firm growth that suggests that the proportionate
growth of each firm in a market is random and independent of the size of the firm,
so that the ‘chance’ that a firm will grow by a given percentage is the same
whatever its size. The randomness of firms’ growth rates is a consequence of the
multiple economic and chance factors that influence firms, such as effectiveness
of advertising campaigns, successful launch of new products, strikes, effect of
exchange rates, etc. The law of proportionate effect predicts that the size
inequality of firms will tend to increase over time as a result of chance factors and
so will lead to increasing concentration. This will tend to occur even if all firms
have the same level of unit costs. Although chance may play some part in a
firm's growth or decline, however, modern theories of firm growth focus more on
the part played by firms’ ability to create and sustain competitive.” Source:
https://financial-dictionary.thefreedictionary.com/law+of+proportionate+effect
These concerns, through consideration of probability and decision analysis of finite and
“state of nature events” will be developed.
 The term “event” refers to an action of interest to a decision maker.
 “State of nature events” specifically refers to mutually exclusive events or outcomes of
interest to a decision maker, where if a given event occurs, other events that were
possible, can no longer happen. For example:
o If you choose to buy car A, you are then unable to buy car B. C or D.
o These are discrete event outcomes, which are from a very small set of possible
events or outcomes that a decision maker might choose from.
 Where there might be many thousands or more potential event outcomes, we clearly are
not dealing with a small, manageable discrete set of possible events – but in fact are
considering a distribution of events. This type of event outcome is considered as
distribution functions.
The Decision Tree

Discrete event outcomes of unknown or uncertain decisions can be described
mathematically by using logical analysis called decision analysis.
 Solution set introduced in Module 1 is a range of solutions which describe the range of
possible outcomes for a given decision.
 This range of solutions can be represented by a logical decision tree which is a visual
representation of outcomes that emerge from a given decision and which, when drawn,
can look like a tree.
o In this manner, this is a logical model reflecting the RAT models outlined in
Module 1. All outcomes can be articulated for a given series of sequential
decisions so as to determine the optimal outcome given different likelihoods for
those outcomes occurring.
o Hence a logical “tree” of outcome is generated which is a chronological
representation of the decision problem.
There are conventions to adopt to construct such decision trees using a particular form and
structure. Each decision tree has two types of outcome or nodes.
1. Round nodes correspond to the states of nature (the outcomes)
2. Square nodes are used to represent (further or subsequent) decision alternatives.
In such a way, decision alternatives can lead to further state of nature outcome nodes and
hence further decisions, and so forth.
 In this way, the full decision solution set can be explored and visually represented.
 When all solution set outcomes have been mapped, then it is also common practice to
identify payoffs/costs/benefits that accrue should that branch of decision tree manifest in
reality.

 In constructing these trees, it is common practice to begin at the first decision (which is
constructed on the left of the subsequent diagram) and logically proceed right, using the
appropriate nomenclature (square and round nodes) to represent outcomes from
decisions.
 Only when all solution set has been captured and visualized, can data be added to the
tree.
 Such a simple articulation of a solution set of course, reflects available information at a
given point and time in the model’s development.
Undertaking actions to improve understanding about the likelihood of a given outcome
occurring can subsequently require a revision of the original tree or mapping of the solution set.
 It is common practice to describe the original probabilities (likelihood) of a given decision
occurring from a solution set as being prior probabilities.
 When the tree is then revised in light of new knowledge of the problem (and hence an
improved model has then been developed), the probabilities of a given decision then
occurring from a solution set are described as being posterior probabilities.
 Consider a simple example and review some simple refreshers on basic probability.
o The probability of a given event (E) can be written as: P(E)
o Mutually exclusive events (States of Nature) describe that if one outcome event
occurs, the others from the solution set cannot
o Independent events are described when the occurrence of one event does not
affect the likelihood of an alternative event occurring
o The logical rule of independent events occurring (without regards to their
combination) is the OR rule [i.e. P(throwing a 3 or 6 on a single throw of a die)] =
1/6 + 1/6 = 1/3.
o The logical rule of sequential non-independent events occurring is the AND rule
[i. e. P(throwing a 3 and then a 6 on a single throw of a die)] = 1/6 × 1/6 = 1/36.
o The likelihood of several given event outcomes occurring following a decision
being made must sum 1 (as one of the event outcomes will occur).
o Conditional probabilities described the context where one event DOES affect the
likelihood of another then occurring, i.e. probability of even X occurring after Y
event has already occurred = P(X × Y)
The Production Line
On a production line, 90% of the time it is set up correctly. When this happens, 95% of the
output it produces are good parts. If the line is incorrectly setup, then the chance of it producing
a good part is only 30%. On any given day, when the line is set up and the first goods made are
found to be good, what is the probability that the machine is set up correctly?
If we draw a decision tree to cover the solution set,
1. We begin with the first decision, which is the decision to “Set-up Production Line”
represented by the square node.

2. This can have two state of nature outcomes as defined in the question represented by
the round nodes
a. The line can be set up correctly or
b. It can be set up incorrectly
3. The inspection of parts from both potential branches of setting up the production line (the
next decision) can result (as outcomes) in parts then being assessed as either good or
bad. Hence the tree would look like the diagram below:
Set up Parts
Parts Good
PartsGood
Not Good
Production Line Parts Not Good
Not
Ok?
Ok?
4. Forward passing through the logic of the question (from left to right), the simple tree now
contains all possible solutions: Good parts can arise from both a good and bad initial set
up.
5. Backward passing through the tree (from right to left), we then add measured data in this
case
a. the likelihood of producing good parts from set up,
b. the likelihood of being set up correctly, and
c. the likelihood of producing a good part from a bad set up
Therefore, the solution set on the diagram in the succeeding page that contains good parts as
an outcome (event) of this model can be described as:
 Probability of parts good when good initial set up, Probability of parts good when bad
initial set up
 Conditional probabilities then occurring, i.e. probability of even X occurring after Y event
has already occurred. Our decision to resolve to learn if the production line is set up
correctly or not initially results to the following:
o The likelihood of an initial good set up is 0.90 × 0.95 = 0.855 or 86%
o The likelihood of an initial bad set up is 0.10 × 0.30 = 0.03 or 3%
o Therefore, the likelihood of good parts from a good or bad initial set up is 0.855 +
0.03 = 0.885 or 88.5%.
 Now, we can finally answer the question initially asked: On any given day, when the line
is set up and the first goods made are found to be good, what is the probability that the
machine is set up correctly? expressed as P(good production line set up) is 0.855 ÷
0.885 = 0.9661 or 96.61%.
Set up 0.10 0.95

0.30 Parts
PartsGood
Parts Good
Not Good
0.70
Production Line 0.90 0.05 Parts Not Good
Not
Ok?
Ok?
The example given did not include pay-offs (financial results following a particular series of
decisions and outcomes) which would be associated with the outcome event solution set.
 When pay-offs are added to a decision tree, they sit on the very right of the diagram and
denote profit, cost, time distance or any appropriate measurement argued to
preferentially differentiate between preferred outcome events.
 Pay-offs are therefore the result of a specific combination of a decision alternative and a
state of nature.
A diagram example of a solution set decision tree with pay-offs is shown below:
To sum up the recently concluded discussion and activity:

The likelihood of an event is known as probability. The probability of an event is a number
that always satisfies 0 ≤ p ≤ 1, where 0 indicates an impossible event and 1 indicates a certain
event. A probability model is a mathematical description of an experiment listing all possible
outcomes and their associated probabilities.
Decision tree analysis involves making a tree-shaped diagram to chart out a course of
action or a statistical probability analysis. It is used to break down complex problems or
branches. Each branch of the decision tree could be a possible outcome. The tree structure in
the decision model helps in drawing a conclusion for any problem which is more complex in
nature.
Additional Sources:
 https://courses.lumenlearning.com/ivytech-collegealgebra/chapter/constructing-
probability-models/
 https://economictimes.indiatimes.com/definition/decision-tree-model
 https://toggl.com/blog/decision-tree-analysis
Decision Analysis
Let us consider how decisions are made where we may lack specific probability data to
support the development of a complete solution set. This is often called decision analysis
under uncertainty. Three approaches to modelling decisions are described when we are
considering alternate states of nature outcomes without comparative probabilities which are as
follows:
1. Optimistic approach – where the decision maker acts rationally (following RAT models)
and chooses the largest payoff (or alternately the payoff which uses a minimum of
resources).
o It is also called the equal likelihood criteria as each state of nature is equally
likely to occur or sometimes the Laplace decision criterion named after its
originator Pierre-Simon Laplace.
o This approach does not consider the risk of incurring large losses as it is focused
upon the highest average return.
 This method of decision making is unattractive for smaller organization
who are more concerned with short term cash flow.
Example: Ms. Firmness decides to build production plant considering the following options:
 A large production plant, and it turns out that the market is favorable, she determines
that the project will have a net benefit to his company of P9.00M. If she builds that
large plant, but the market is unfavorable, then the net monetary effect will be a loss
of P8.10M.
 A small production plant, that during favorable market conditions, the project will
have a NPV of P4.50M. But when the market is unfavorable, the NPV will be a loss
of P0.90M.
Please be reminded of the following that are redacted from the problem context and
procedure:
 In determining the amount of gain or loss (cash flow), the modeler/decision maker
must project have projected the corresponding revenues and expenses (cost-benefit
analysis) of each project for each “State of Nature.”

 The projected net cash flows are converted at their Net Present Values (NPV). It can
be simple using Excel application.
 The decision tree is not presented (which you can do on your own guided by the
previous illustration).
Therefore, we can compute for the average NPV by adding the NPVs for the “State of
Nature” then divide by 2. Our solution using Excel is as follows:
State o f Nature
Average
Alternative Favo rable Unfavo rable
NPV
Market Market
Co nstruct a large p lant 9,000,000 (8,100,000) 450,000
Co nstruct a small p lant 4,500,000 (900,000) 1,800,000
Do no thing - - -
Since the decision alternative “Construct a small plant” has the highest average, Ms.
Firmness would select that choice as the decision to implement.
2. Conservative approach – where the decision maker acts rationally and chooses the
best of the worst possible payoffs i.e. a payoff may be the maximum of the minimum
payoffs (Maximin) achievable.
o This is also called the Wald decision criteria named after its originator,
Abraham Wald.
o The strategic objective of this approach is to avoid the worst outcomes. In these
situations
 the decision maker may gain more in hindsight by following different
decisions
 but cannot gain less choosing the maximin payoff
Using the same example, the objective is to avoid the worst outcomes; therefore, our
solution is as follows:
State o f Nature
Alternative Favo rable Unfavo rable Maxim in
Market Market
Co nstruct a large p lant 9,000,000 (8,100,000) (8,100,000)
Co nstruct a small p lant 4,500,000 (900,000) (900,000)
Do no thing - - -
Since the decision alternative “Do Nothing” has the highest payoff, Ms. Firmness would
select that choice as the decision to implement. Note that this is clearly a risk-reducing
decision.
o Applying the Maximin rule will ensure that the business will not suffer a loss.

o However, it will not also have the possibility of gaining a profit either.
o The payoff of “0” is the largest payoff, as it is higher than either of the negative
numbers.
3. Minimax regret approach – where the decision maker identifies the decision that
results in the minimum of maximum regrets (losses) between different outcome events.
o The objective is to ensure that the larger opportunity losses are avoided.
o This is also called Savage decision criteria after its founder Leonard Jimmie
Savage.
Working on the same example, our solution for this approach will be:
State o f Nature
Alternative Favo rable Unfavo rable Maxim in
Market Market
Co nstruct a large p lant 9,000,000 (8,100,000)
Co nstruct a small p lant 4,500,000 (900,000) (900,000)
Do no thing - -
Since the decision alternative “Construct a small plant” has the lowest payoff, Ms.
Firmness would select that choice as the decision to implement. Note that “Doing nothing”
alternative is not considered because the decision-maker’s objective is to choose the “lesser
worse” outcome or avoiding the larger loss.
Source: Methods for Decision Science, https://www.shsu.edu/law001/site/Home.html
Clearly an important concern for a decision maker, making decisions under uncertainty is to
identify and consider which decision-making criteria to use.
 There is no guarantee or expectation that the three outlined decision-making criterion
will generate the same recommended decisions.
 Which criterion to use, is partially dependent upon who will be judging the success or
otherwise of those decisions.
o If it is not the decision maker for example, then the minimax regret approach
should be adopted.
o It has been noted that a small company may not be able to bear significant
losses and hence will make decisions conservatively
o If, for a larger organization there is little to differentiate between decision
alternatives, then the optimistic decision criteria can be used.
A fourth criteria or approach can be considered on this regard, which is called the “Realism
(Hurwicz) Decision Rule.” It is an attempt to make a tradeoff between complete risk
indifference and total risk aversion. With this procedure, the decision maker will decide how
much emphasis to put on each extreme. Explained in another way, the Criterion of Realism

calculates a weighted average of the best and worst outcomes for each alternative using alpha
(α) and (1 – α) as the weights, respectively.
 The decision maker must choose a Coefficient of Realism, called alpha (α), which is a
decimal number between 0 and 1.
 This number provides the emphasis on the optimistic view.
o The number (1 – α), then, is the amount of emphasis that is placed on the most
pessimistic outcome. For example,
 If a manager chose an alpha of 0.6, he would be placing 60% emphasis
on a risky, high return outcome, and
 40% emphasis (1 – 0.6 = 0.4 or 40%) on a low-risk, pessimistic outcome.
 To determine the decision under this rule, a column is added on the right side of our
Excel solution payoff table. In this column, the decision maker must calculate the factor
by
o multiplying the best outcome in the row by alpha (α),
o multiplying the worst outcome in the row by (1 – α), and
o adding the two results together.
o The highest one that will be in the row of the decision alternative is to be selected
under this rule.
In our example, assuming an alpha (α) of 80%, the alternative to be selected is “Construct
a large plant” which has the maximum pay-off as provided by out solution below.
State o f Nature Wig hted
Alternative Favo rable Unfavo rable Averag e ( α =
Market Market 0 .8 0 )
Co nstruct a large p lant 9,000,000 (8,100,000) 5,580,000

Co nstruct a small p lant 4,500,000 (900,000) 3,420,000
Do no thing - - -
Source: Methods for Decision Science, https://www.shsu.edu/law001/site/u3m3p6.html
Expected Monetary Values

Expected Monetary Values (EMVs) are additional calculations derived from the multiplication
of a given outcome likelihood and the payoff for that likelihood. As there are several outcomes
following a decision, one of these outcomes will happen (should that decision be taken) and
hence the EMV of that decision is the summed weighted payoffs for the outcomes of that
decision. This generates a weighted return for a given decision outcome which can then be
compared by the manager, with other weighted returns, to determine, rationally, the optimal
decision strategy to adopt. The formula is expressed as:
EMV (d¿¿ 1)=Σ P (s 1) ×V i ¿ for i = 1 to n event outcomes (state of nature).

Vi = payoff for ith event outcome (state of nature)

Let’s consider a simple example: In an organization, 2 projects are being considered for
funding. You have been asked to calculate the EMV of each project using the data given in
order to determine which project is likely to yield the largest return.
1. The probability of project X being a very successful for P360K profits are 20%, the
probability of project X being moderately successful for P210K profits are 50% with all
other outcomes generating profits of P150K.
2. The probability of project Y being a very successful for P390K profits are 10%, the
probability of project X being moderately successful for P240K profits are 60% with all
other outcomes generating profits of P60K.
Computed solution by using Excel is as follows:
Profits Probability
Project Very Moderately Other Very Moderately Other EMV
Successful Successful Outcomes Successful Successful Outcomes
X 360,000 210,000 150,000 20% 50% 30% 222,000
Y 390,000 240,000 60,000 10% 60% 30% 201,000
Based from the above, Project X would be supported because it will generate a higher EMV.
Notes:
 The Probability in percent supplied in the “Other Outcomes” column (30% for both
projects) is determined by deducting the sum of the two outcomes (Very Successful and
Moderately Successful which is 20% and 50% for Project X, and 10% and 60% for
Project Y, respectively) from 100% (the total Probability should always be 100%).
 EMV for Project X is computed to wit: (360,000 × 20%) + (210,000 × 50%) + (150,000 ×
30%) = 222,000. Follow this sequence in computing the EMV for Project Y.
As far as computing the EMV for the example given in the eBook on pages 112 to 114, herewith
is the solution:
Profits Probability Profits Probability
Competitor Pricing Composite
Company Price Competitor's Price Probability With No With No
EMV
Competition Competition Competition Competition
Low Medium High Low Medium High
Low 30 42 45 80% 15% 5% 32.55 50 60% 40% 39.53
Medium 34 45 49 20% 70% 10% 43.20 70 60% 40% 53.92
High 10 30 53 5% 35% 60% 42.80 90 60% 40% 61.68
After inputting the given variables, there are two computations involve in determining EMV
(cells or columns shaded in gold) as follows:
1. EMV on Profits considering Competitor Pricing

a. Low Company Price Row or Decision 1 – (Profits, Low × Probability, Low) +

(Profits, Medium × Probability, Medium) + (Profits, High × Probability, High) =
32.55
b. Medium Company Price Row or Decision 2 – (Profits, Low × Probability, Low) +
43.20
c. High Company Price Row or Decision 3 – (Profits, Low × Probability, Low) +
42.80
2. Composite EMV (or final EMV)
a. Low Company Price Row or Decision 1 – (Profits, With Competition × Probability,
With Competition) + (Profits, No Competition × Probability, No Competition) =
39.53
b. Medium Company Price Row or Decision 2 – (Profits, With Competition ×
Probability, With Competition) + (Profits, No Competition × Probability, No
Competition) = 53.92
c. High Company Price Row or Decision 3 – (Profits, With Competition ×
Probability, With Competition) + (Profits, No Competition × Probability, No
Competition) = 61.68
Note: Probability, With Competition is given in the problem at 60%. Considering the rule that
Probability is always equal to 100%, we deduct the 60% from it and the 40% Probability, No
Competition is represented.
Therefore, the optimal strategy is Decision 3, Enter the market place and set a high price
(regardless of what the competitor’s pricing policy is) because of higher payoff or profits. See
the decision tree for this problem on page 114 for you to analyze and dissect how it is done.
Benefits of EMV
 It gives you an average outcome of all identified uncertain events.
 It helps you to calculate the contingency reserve.
 In a decision tree analysis, it helps in selecting the best choice.
 It does not require any costly resources, only experts’ opinions.
 It helps you with a make or buy decision during the plan procurement process.
Drawbacks of EMV Analysis
 This technique is usually not used in small and small-medium-sized projects.

 This technique involves expert opinions to finalize the probability and impact of the risk.
Therefore, personal bias may affect the result.
 This technique works well when you have many risks.
 If you miss positive risks, it will affect the outcome.
Additional source: A Short Guide to Expected Monetary Value (EMV),
https://pmstudycircle.com/2015/01/a-short-guide-to-expected-monetary-value-emv/

Bayes Theorem
Bayes' theorem, named after 18th-century British mathematician Thomas Bayes, is a
mathematical formula for determining conditional probability. Conditional probability is the
likelihood of an outcome occurring, based on a previous outcome occurring. Bayes' theorem
provides a way to revise existing predictions or theories (update probabilities) given new or
additional evidence. Bayes' theorem is also called Bayes' Rule or Bayes' Law and is the
foundation of the field of Bayesian statistics.
Applications of the theorem are widespread and not limited to the financial realm. It can be
used to determine the accuracy of medical test results by taking into consideration how likely
any given person is to have a disease and the general accuracy of the test.
 Bayes' theorem relies on incorporating prior probability distributions in order to generate
posterior probabilities.
o Prior probability is the probability of an event before new data is collected. This is
the best rational assessment of the probability of an outcome based on the
current knowledge before an experiment is performed.
o Posterior probability is the revised probability of an event occurring after taking
into consideration new information.
 Posterior probability is calculated by updating the prior probability by
using Bayes' theorem.
 In statistical terms, the posterior probability is the probability of event A
occurring given that event B has occurred.
 Bayes' theorem thus gives the probability of an event based on new information that is,
or may be related, to that event.
 The formula can also be used to see how the probability of an event occurring is affected
by hypothetical new information, supposing the new information will turn out to be true.
For instance, say a single card is drawn from a complete deck of 52 cards.
o The probability that the card is a king is four divided by 52, which equals 1/13 or
approximately 7.69%. Remember that there are four kings in the deck.
o Now, suppose it is revealed that the selected card is a face card. The probability
the selected card is a king, given it is a face card, is four divided by 12, or
approximately 33.3%, as there are 12 face cards in a deck.
 The formula for Bayes’ Theorem is
P ( A ⋂ B) P ( A )× P( A ∣ B)
P ( A ∣ B )= =
P( B) P ( B)
Where:
P(A) = The probability of A occurring
P(B) = The probability of B occurring
P(A ∣ B) = The probability of A given B

P(B ∣ A) = The probability of B given A

P(A⋂B) = The probability of both A and B occurring
Source: Adam Hayes (2020). Bayes’ Theorem Definition.
https://www.investopedia.com/terms/b/bayes-theorem.asp#:~:text=Bayes'%20theorem%2C
%20named%20after%2018th,formula%20for%20determining%20conditional
%20probability.&text=Bayes'%20theorem%20provides%20a%20way,given%20new%20or
%20additional%20evidence.
Example:
It is likely that the current difficult economic climate in the US and UD, will potentially will
affect the numbers of students entering in higher education. As a result, expected sales of
student textbooks, whether electronic or traditional paper copies, will fall and/or fluctuate. Let’s
assume you undertake some internal market research. You ask your colleagues for their view of
this impact upon your company’s performance. It is suggested that the company can expect
 a profit of £1.5M if the student numbers studying for new entry in the coming year fall by
a small amount,
 a profit of £0.5M if numbers fall by a moderate amount, and
 a loss of £2.0M if numbers fall by a heavy amount.
You have estimated that the likelihood of these events is:
 P(small) = 0.40,
 P(moderate) = 0.30, and
 P(heavy) = 0.30.
With the solution set tree indicates a likely profit, the concern with the best “guess” market
probabilities is a concern for the company. To potentially offset this, your company is
considering reallocating some of their production capacity by leasing it to another organization.
If they do this, the potential loss of profit will not be as significant as projected due to the
additional leased income, but this will limit the productive capacity of your company, if expected
decline in numbers of student does not occur. In this scenario, you have projected the following:
 a profit of £1.0M if student numbers fall by a small amount,
 a profit of £0.75M if student numbers fall by a moderate amount, and
 a loss of £0.5M if student numbers fall by a heavy amount.
Your first step is then to draw a simple decision tree for these assumptions of the market
and the likely market value for the company in the coming year. (Please refer to the decision
trees as illustrated in the eBook). After preparing the solution set for the decision tree, we can
use Excel to compute for the Expected Monetary Values (EMV) of the respective State of
Nature (Event) which is shown below.

Expected Monetary Values

Profits - Profits -
State of Nature EMV EMV
No Leasing Probability Leasing
S1 (Small Market) 1,500 40% 600 1,000 400
S2 (Medium Market) 500 30% 150 750 225
S3 (Large Market) (2,000) 30% (600) (500) (150)
Total 150 475
Your optimal decision therefore, by comparing EMVs is to invest in the leasing of some
productive capacity as your company is expected to earn better at £0.475M compared with
£0.15M if leasing is not considered. However, what is the additional EMV benefit of deciding to
lease some of the company’s productive capacity? By combining the two tree diagrams and
determine EMV with Perfect Information (EMVwPI). This is an expectation of the optimum value
of given event outcomes and represents the maximum return that could be expected given that
the “optimal” event outcomes manifest. It provides therefore some indicative measurement of
the price that a manager would be willing to pay to learn of perfect information about a decision.
Computation of EMVwPI is presented below.
Profits - Profits -
State of Nature Best EMV
No Leasing Leasing Probability
S1 (Small Market) 1,500 1,000 40% 600
S2 (Medium Market) 500 750 30% 225
S3 (Large Market) (2,000) (500) 30% (150)
675
EMVwPI 200
Please observe the following:
 Compare the Profits (No Leasing and Leasing alternatives), the best figure is considered
to compute the EMV.
o S1 – between £1.5M and £1.0M, the higher figure is taken in the computation
o Similar procedures followed for S2 and S3.
 The resulting EMVs for S1, S2, and S3 are summed up and deducted from the EMV
from Leasing decision (D2) of £0.475M which resulted to a difference of £0.20M referred
to as the EMVwPI.
Given that the decision to lease some productive capacity can be made flexibly, which
indeed is a preferred decision, the board of the company decided it would be prudent to wait
until September and then use data on the numbers of students who are applying during the
Clearing Period, as the best indicative guide to actual expected student enrollment. The
Clearing period is traditionally few weeks in August where students who are late in applying to
universities or missed the terms of their offer for a place of study, can contact all the
Universities in the UK and discuss with them, the possibility of gaining an offer of a place of
study.

It is therefore argued that an active clearing period in September is likely to require more
productive capacity from the company and leasing option becomes less attractive. Estimates of
the probabilities for a very busy clearing period (CP) denoting the states of nature outcomes for
a small, moderate, or heavy fall in student numbers respectively are:
 P(CP ∣ s1) = 0.30

 P(CP ∣ s2) = 0.20
 P(CP ∣ s3) = 0.50
To see how this decision affects the solution set for the problem, reconsider the decision
tree and analyze how this new market data changes earlier assumptions in our analyses: D1
and D2 refer to “no leasing” and “leasing” decisions respectively.
We need to revise the prior probabilities in light of this new conditional information which can
be achieved using a Bayesian revision by considering first the Busy Clearing Period. Similarly,
we can generate the posterior probabilities for a Non Busy Clearing Period. Below is the revised
computation of EMVs for both.
VERY BUSY CLEARING PERIOD
State of Nature (after Profits Probability EMV
Clearing) No Leasing Leasing Prior Conditional Joint Posterior No Leasing Leasing
S1 (Small Market) 1,500 1,000 40% 30% 12% 36% 545.45 363.64
S2 (Medium Market) 500 750 30% 20% 6% 18% 90.91 136.36
S3 (Large Market) (2,000) (500) 30% 50% 15% 45% (909.09) (227.27)
Totals 33% 100% (272.73) 272.73
NON BUSY CLEARING PERIOD

State of Nature (after Profits Probability EMV
Clearing) No Leasing Leasing Prior Conditional Joint Posterior No Leasing Leasing
S1 (Small Market) 1,500 225 40% 70% 28% 42% 626.87 94.03
S2 (Medium Market) 500 (150) 30% 80% 24% 36% 179.10 (53.73)
S3 (Large Market) (2,000) 675 30% 50% 15% 22% (447.76) 151.12
Totals 67% 100% 358.21 191.42
Notes:
1. Very Busy Clearing Period
o Conditional probabilities for each State of Nature are given
o Joint probabilities are computed by multiplying the Prior and Conditional
probabilities
o Posterior probabilities are computed by dividing each of the State of Nature by
the total of Joint probabilities
o EMVs are computed in the usual manner.
2. Non Busy Clearing Period
o Conditional probabilities for each State of Nature is computed by subtracting the
conditional probabilities in Very Busy Clearing Period (per State of Nature) from
100% to wit:

 S1 – (100% – 30% = 70%)

 S2 – (100% – 20% = 80%)
 S3 – (100% – 50% = 50%)
Rationale: Probability rule of 100%
 Since the conditional probability of the occurrence of Very Busy
Clearing period is given, and there are only two probability criteria, the
other being the Non Busy Clearing Period, therefore, applying the rule
will mean the difference of 100% from the given conditional probability
which is assigned to Non Busy Clearing Period.
Based from the foregoing computations, we are now provided with an array of probable
information to guide us in our decision.
The Value of Sample and Perfect Information

Market research can be undertaken by companies to improve their likely chances of
competitive success in a given market. Learning more about the market necessitates taking
assumptions of knowing all about the market which is called perfect information.
 Decisions made by managers at potentially impacting performance of considerable
significance to the organization require gathering sufficient information and simulating
this information of their relative and probable contribution is an exercise of responsible
conduct of rational decision-making. Under this circumstance, it is called optimal
decision strategy with perfect information.
 If we now knew what the probability of the events attributable to the specific criteria
were, we would be able to determine the expected value of the optimal strategy in the
normal manner. We then call this the expected value with perfect information.
 If we don’t know or have no access to perfect information, the expected values of a
given decision tree and solution set ca be referred to as the expected value without
perfect information.
 The difference between the two expected values represents the benefit that comes from
certainly knowing how the market or situation will develop is the value of perfect
information.
For a manager, it represents the maximum “price” that hey would be prepared to pay to
know more about how the market would actually develop for them. It is highly unlikely that any
market research would ever generate enough insight to guarantee for knowledge of future
events, so the value of perfect information represents an intangible ceiling upon the fee for any
additional market knowledge.
From the preceding discussions, we were introduced to a refinement of the idea of perfect
knowledge, the sample information.
 Sample information means simply that more tested information is generated about the
states of nature of how more or less likely they actually are to occur in reality.

 As managers improving the subjectivity of initial or prior assumptions about the market
or product or buyer behavior etc. through some tested information on that market or
product or public.
The important mathematical and practical difference in how managers develop the model of
the problem to be understood is how the sample information changes the prior probabilities.
When managers revise their initial or prior probabilities in this way, they become posterior
probabilities.
 It is common assessments that the sample information is viewed as being for or against
a particular decision.
 To articulate this argument fully means managers must also consider how the states of
nature relate to the findings of the market research.

Example:
Assume that a product is being sold by a company which may or may not be faulty in use.
You have a test which can be used to indicate if the product will become faulty in use. You wish
to know the strength of the relationship between the test finding for a product and if that product
then goes on to become faulty (how reliable is the test and should it then be something you use
in all your products). So, you have experience which leads you to believe that 2% of your
products are faulty in service. The test you have available can detect a future fault in 85% (has
been proven) of the time. Sometimes the test generates incorrect findings (test indicates a
future fault but in service, the product remains faultless). Your experience of this suggests this
occurs in 8% of sampled goods and you prepared this table of data:
Product is faulty? Product is OK?

Test is positive for a fault 85% 8%
Test is negative for a fault 15% 92%
We interpret this representation of data as we know 2% of our products are faulty. If we find
a faulty product, then there is an 85% chance our test will verify this. If we choose another
product that we know to be fault free, then there is an 8% chance our test will say that is faulty.
So, suppose you now choose a good at random and test it and the test indicates it is faulty,
what are the actual odds that the product is faulty? With the test finding a fault, we are
concerned with the top row of the table and evaluate it by stating:
 Chance of the product being faulty and the test found the fault = 2% × 85% = 0.017 or
1.7%
 Chance of the product not being faulty (product is OK) and the test found a fault = 98%
× 8% = 0.0784 or 7.8%
Going back to the original question of what is the likelihood that we have chosen a faulty
product given the test indicates the product is faulty?
P (Product is faulty ∣ Test reveals a fault) = 1.7% ÷ (1.7% + 7.8%) = 17.9%
So, the likelihood of a good being faulty and revealed by the test is 18% (or there is an 82%
chance that the product is not faulty).
______________
Activities
.


Module 3 - Probabilities

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Module 3 - Probabilities

Uploaded by

Copyright:

Available Formats

INSTRUCTIONAL MATERIAL

 Understand the distribution functions and queuing theory

References and Resources:

 Effective Management Decision Making: An Introduction by Ian Pownall, ©2012 (eBook),

 Youtube video - Introduction to Bayesian data analysis - part 1: What is Bayes?

MANA 4033 – Decision Making

Developing Rational Models with Quantitative Methods and Analysis:

MANA 4033 – Decision Making

The Decision Tree

MANA 4033 – Decision Making

MANA 4033 – Decision Making

Set up 0.10 0.95

To sum up the recently concluded discussion and activity:

MANA 4033 – Decision Making

MANA 4033 – Decision Making

MANA 4033 – Decision Making

Co nstruct a large p lant 9,000,000 (8,100,000) 5,580,000

Source: Methods for Decision Science, https://www.shsu.edu/law001/site/u3m3p6.html

Expected Monetary Values

EMV (d¿¿ 1)=Σ P (s 1) ×V i ¿ for i = 1 to n event outcomes (state of nature).

MANA 4033 – Decision Making

Vi = payoff for ith event outcome (state of nature)

MANA 4033 – Decision Making

a. Low Company Price Row or Decision 1 – (Profits, Low × Probability, Low) +

 This technique is usually not used in small and small-medium-sized projects.

MANA 4033 – Decision Making

 The formula for Bayes’ Theorem is

MANA 4033 – Decision Making

P(B ∣ A) = The probability of B given A

MANA 4033 – Decision Making

Expected Monetary Values

Please observe the following:

MANA 4033 – Decision Making

 P(CP ∣ s1) = 0.30

NON BUSY CLEARING PERIOD

MANA 4033 – Decision Making

 S1 – (100% – 30% = 70%)

The Value of Sample and Perfect Information

MANA 4033 – Decision Making

MANA 4033 – Decision Making

Product is faulty? Product is OK?

MANA 4033 – Decision Making

You might also like