INDUSTRIAL ENGINEERING DEPARTMENT

Introduction to Operations Research III

Decision Theory

1. The MBA Movie Studio is trying to decide to distribute its new movie “Claws”. The
movie has the potential of being a great financial success (a “smash”), but the
executives are not sure because the subject is controversial. And they have seen
some films heralded as “smashes” become “flops” with disastrous financial
consequences.

The decision facing MBA is whether or not to put out the movie “Claws” on a limited
first run basis. This means that the movie will show only in a few select theaters
during the first six months. After six months it will be released generally. If the movie
turns out to be a success, this is clearly the best approach because the studio makes
considerable profit from these theaters.

The other alternative is to release the film for wide distribution immediately. The
profits for the two alternatives are given in the table below, classified in terms of
whether the film is a “smash”, or “medium” success, or a “flop”.
Profits from Film “Claws”
Level of Success Probability Limited Initial Widespread
Release (in Release
millions) (in millions)
Smash 0.3 22 12
Medium 0.4 8 8
Flop 0.3 -10 -2

There is considerable discussion in MBA about the potential of Claws. Management has
finally agreed on the probabilities shown in the table. But which decision to make is
still not clear. One possibility is to have a few sneak previews of the movie and get the
audience’s opinions. The cost of such a process would be about Php 50,000. And
several executives in the company feel it would be money wasted, since sneak preview
audience tends to rate a movie as good or outstanding even when it later tuns out to
be flop. To support this, the following table was produced, describing the company’s
past experience with sneak preview audience reactions.
Sneak Preview Audience Reaction
Audience Smash Medium Flop Total
Rating
Outstanding 9 12 3 24
Good 1 6 5 12
Poor 0 2 2 4
Total 10 20 10 40
a. Draw the decision tree for this problem.
b. Calculate the posterior probabilities for “smash”, “medium”, and “flop” given the
various audience reactions.
c. Assume that MBA is willing to base its decision on Expected Monetary Value. What
decision should the MBA Movie Studio make about the movie “Claws”?
2. The Tarheel Manufacturing Company must decide whether to build a large plant or a
small one to process a new product with an expected life of 10 years. Demand may be
high during the first 2 years, but if many users find the product unsatisfactory,
demand will be low for the remaining 8 years. High demand during the first 2 years
may indicate high demand for the next 8 years. If demand is high during the first 2
years and company does not expand within the first 2 years, competitive products will
be introduced, thus lowering the benefits.

If the company builds a large processing plant, it must keep it for 10 years. If its
builds the small plant, the plant can be expanded in 2 years if demand is high, or the
company can stay in the small plant while making smaller benefits on the small volume
of sales. Estimates of demand are these:

Probabilit Probabilit
y y
High demand (first 2 0.5 High demand during 0.6
years) followed by first 2 years
high demand (next 8
years)
High demand (first 2 0.1
years followed by low
demand (next 8 years)
Low demand (first 2 0.4 Low demand during first 0.4
years) followed by 2 years
continuing low demand
(next 8 years)
Low demand (first 2 0
years) followed by
high demand (next 8
years)

Financial costs and profits are as follows:

 A large plant with high demand would yield Php 1 million annually in profits.
 A large plant with low demand would yield Php 200,000 annually because of
production inefficiencies.
 A small plant, not expanded, with a low demand would yield annual profits of Php
250,000 for 10 years.
 A small plant during a 2-year period of high demand would yield Php 450,000
annually; if high demand continued and if the plant was not expanded, this would
drop to Php 300,000 annually for the next 8 years as a result of competition.
 A small plant which was expanded after 2 years would yield Php 100,000 annually
for 8 years if low demand occurred during that period.
 A large plant would cost Php 5 million to build and put into operation.
 A small plant would cost Php 1,500,000 to build and put into operation.
 Expanding a small plant after 2 years would cost Php 2,500,000.

Under the conditions stated and with the information furnished, analyze the
alternatives and choose the best decision.
3. A machine in a group of 50 machines is serviced when it breaks down. At the end of T
period, preventive maintenance is performed by servicing all 50 machines. The cost of
repairing a broken machine is Php 1000 while Php 100 is the preventive maintenance
cost per machine. Determine the optimum T that minimizes the total cost per period.

Probability Distribution of Breakdown before T

0.03, t  1

P (t )  Pt 1  0.01, t  2 10
  11,12
0.125, t

4. An automatic machine produces  (thousands of) units of a certain product per day.
As  increases, the proportion of defective p goes up. The probability density function
of p in terms of  is given by

p  1 ,
 0  p 1
f ( p)  

0, otherwise

Each defective item incurs a loss of Php 50. A good item produces a profit of Php 5.
Determine the optimal value of .

5. A doctor must diagnose the condition of one of her patients. She is certain that the
condition is not life threatening, so there is not risk to the patient’s life. However, in
diagnosing the condition the doctor would like to minimize the cost to the patient.
There are three tests she could conduct on her patient. The doctor has narrowed the
range of possibilities down to three disease conditions, and indicates her professional
judgmental probability assessments of the three states as 40 percent, 25 percent, and
35 percent that the disease is type 1, type 2, and type 3, respectively. The cost (in
Php) to the patient for diagnosis and treatment is given in the table of possibilities
shown in the table below depending on the test employed. Answer the following by
dealing directly with costs (the problem can also be solved by constructing a table of
negative payoffs):
a. Determine the Bayes decision (the test which minimizes the expected cost).
b. Compute the EVPI.
c. Suppose the doctor first administers a special blood test that could be used to
refine her probability estimates. The cost of the blood test is Php 50 and it would
indicate one of two results – O 1 or O2. The likelihood probabilities are given in the
following table:
Disease Condition
S=1 S=2 S=3
O1 0.80 0.05 0.40 P(O1/S), i = 1, 2, 3
O2 0.20 0.95 0.60 P(O2/S), i = 1, 2, 3
Should the doctor administer the blood test prior to selecting one of the three major
tests?
Cost of Diagnosis and Treatment
 Test 1 2 3
1 500 300 400
2 600 500 300
3 300 550 450

6. A large mill is faced with the problem of extending Php 100,000 credit to a new
customer, a dress manufacturer. The mill classifies typical companies into the
categories: poor risk, average risk, and good risk. Their experience indicates that 20
percent of similar companies are poor risks, 50 percent are average risks, and 30
percent are good risks .If credit is extended, the expected profit for poor risks is- Php
15,000, for average risks Php 10,000, and for good risks Php 20,000. If credit is not
extended, the dress manufacturers will turn to another mill. The mill is able to consult
a credit rating organization for a fee of Php 2,000. Their experience with this credit –
rating company is given by

Credit company Actual credit rating %

evaluation Poor Average Good
Poor 50 40 20
Average 40 50 40
Good 10 10 40

a) What is the Bayes’ action, assuming the credit rating company is not used?
b) How much money can be paid for “ perfect information”?
c) What is the optimal expected loss if the credit rating company data is used? Does
it pay to utilize these data?
d) What is the Bayes’ action if the credit-rating company determines the dress
manufacturer to be a poor risk?

7. The Breezy Breakfast Foods Company is considering marketing a new breakfast cereal.
If the new cereal is successful, it will mean a Php 10 million profit (present value) over
the life of the product. If unsuccessful, a Php 2 million, loss on investment will be
incurred. Management currently feels there is a 50-50 chance that the product will be
successful.

Two market research firms have approached Breezy with proposals to obtain more
information. Attitude Surveys collects data on consumer attitudes with respect to
specific characteristics of a product, such as sweetness, caloric content, nutritive value,
etc., and produces a forecast of “success” or “fail”. Of the studies this company has
performed on similar products recently, their experience has been as follows:

Attitude Surveys Experience

Actual outcome
Forecast Success Failure
Success 20 5
Failure 5 20
 A second company, Market Competition Inc. ; performs analysis in an entirely
different, independent manner. This company performs extensive analysis on
competitive products, and produce a recommendation of “success” or ”fail” based on
the anticipated amount and quality of competitive products. Their experience with 50
studies has been as follows:

Market Competition Experience

Actual outcome
Forecast Success Failure
Success 22 3
Failure 0 25

Attitude Surveys charges Php 100,000 per survey, while Market competition charges
Php 150,000.

a. Consider only Attitude Surveys. Use a decision tree to decide whether or not Breezy
should purchase this survey.
b. Consider only Market Competition, Inc. Use a decision tree to decide whether or not
Breezy should purchase this survey.

8. Inventory Management. Blumberg’s Department Store will hold a one-month suit sale.
The suits can be purchased in lots of 25 each, an the wholesale post per suit is a
function of the number of suites ordered, as shown below:

Number of suits ordered 25 50 75 100

Cost per suit (Php) 800 750 700 650

Each suit left over at the end of the month will be sold at the clearance sale for half the
retail sales price of Php 1400. If a shopping arises during the month, nothing will be
done to replenish inventory, The possible sales levels are 25, 50, 75, and 100, having
probabilities of 0.20, 0.30, 0.40 and 0.10, respectively.

a. Construct a payoff table for this decision problem.

b. Construct a decision tree to represent the problem.
c. Fold back the decision tree to determine the optical inventory ordering the decision.
d. Marketing Analysis. Suppose Blumberg’s could purchase the services of a marketing
consultant who would conduct a survey that could be summarized by one of two
outcomes: O1 (a favorable market exists for the sale) or O2 (an unfavorable market
exists). The likelihoods, estimated by Blumberg’s from past experience with this
consultant, are given in Table 2. What is the maximum amount Blumberg’s should
pay for the consultant’s services? (Hint: conduct a preposterior analysis).

Table 2
Sales (states of nature)
P(0/S) S = 25 S = 50 S = 75 S = 100
O1 0.05 0.50 0.70 0.85
O2 0.95 0.50 0.30 0.15
9. The Profit & and Gambit Company has a major product that has been losing money
recently because of declining sales. In fact, during the current quarter of the year
sales will be 4 million units below the break-even point. Since the marginal revenue for
each unit sold exceeds the marginal cost by Php 1. This amount to a loss of Php 4
million for shutting down. The other alternative is to undertake an intensive
advertising campaign to increase sales and then abandon the product (at the cost of
Php 4 million) only if the campaign is not sufficiently successful. Tentative plans for
this advertising campaign have been developed and analyzed. It would extend over
the next three quarters (subject to early cancellation), and the cost would be Php 6
million in each of the three quarters. It is estimated that the increase in sales would be
approximately 3 million units in the first quarter, another 2 million units in the second
quarter and another 1 million units in the third quarter. However, because of a
number of unpredictable market variables, there is considerable uncertainty as to what
impact the advertising actually would have, and careful analysis indicates that the
estimate for each quarter could turn out to be off by as much as 2 million units in
either direction. (To quantify this uncertainty, assume that the additional increase in
sales in the three quarters are independent of random variables having a uniform
distribution with a range from 1 to 5 million, from 0 to 4 million, and from –1 to 3
million, respectively). If the actual increases are too small, the advertising campaign
can be discontinued and the product abandoned at the end of either of the next 2
quarters.

If the intensive advertising campaign were to be initiated and continued to its

competition, it is estimated that the sales for some time thereafter would continue to
be at about the same level as in the third (last) quarter of the campaign. Therefore, if
the sales in that quarter still are below the break-even point, the product would be
abandoned. Otherwise, it is estimated that the expected discounted profit thereafter
would be Php 8 for each unit sold over the break-even point in the third quarter.

Determine the optimal policy maximizing expected profit.

10. An oil company has some land that is purported to contain oil. The company classifies
such land into four categories by the total number of barrels that are expected to be
obtained from the well, i.e. a 500,000 – barrel well, a 200,000 barrel well, a 50,000-
barrel well, or a dry well. The company is faced with deciding whether to drill for oil, to
unconditionally lease the land, or to conditionally lease the land at a rate depending
upon the oil strike. The cost of drilling a producing well is Php 100,000, and the cost of
drilling a dry well is Php 75,000. For producing wells the profit per barrel of oil is Php
1.50 (after deducting all production costs). Under the unconditional lease agreement,
the company receives Php 45,000 for the land, whereas under the conditional lease
arrangement the company receives 50 cents for each barrel of oil extracted, provided
the land yields a 200,000 – or 500,000- barrel strike; otherwise, it receives nothing.
The possible profits for the oil company are shown below:

a. What should the company’s decision be? Why?

b. If the company has had some experience with well in similar geographical areas
and has concluded that about 10% of the strikes are 500,000 barrel wells, 15% are
200,000 barrel wells, 25% are 50,000 barrel wells, and 50% are dry wells, what
 will be the decision maker’s decision? Why? (Please use different methods to justify
you answer.)
c. The decision-maker has the choice of not utilizing seismic soundings or utilizing
seismic sounding which would mean a research study which would cost Php 12,000.
Seismic soundings may reveal information on four possible seismic classification
denoted as follows:

1. there is definitely a closed geological structure to the site ( a very favorable

outcome if the presence of oil is desired);
2. there is probably a closed structure to the site;
3. there is a non-closed structure( a relatively unfavorable report);
4. there is no structure to the site an unfavorable condition).
Based upon past examination of similar geological areas (100 such examinations), the
data below are obtained:

State of 500,000- 200,000- 50,000-barrel Dry well

seismic barrel well barrel well well
nature
class
f Prob. f Prob. f Prob. f Prob.
1 7 7/12 9 9/16 11 11/24 9 9/48
2 4 4/12 3 3/16 6 6/24 13 13/48
3 1 1/12 2 2/16 3 3/24 15 15/48
4 0 0/12 2 2/16 4 4/24 11 11/48

Use decision tree to come up with your decision analysis of the problem.

CASE I

A manufacturer produces items having a probability p of being defective. These

items are formed into lots of 150. Past experience indicates that p is either 0.05 or 0.25,
and furthermore, in 80 percent of the lots produced p equals 0.05 (and in 20 percent of
the lots p equals 0.25). These items are then used in an assembly, and ultimately, their
quality is determined before the final assembly leaves the plant. The manufacturer can
initially screen each item in a lot at a cost of P15 per item, replacing defective items
found, or use the items directly without screening. If the latter action is chosen, the cost
per lot can be calculated as:
p = 0.05 p = 0.25
Screen 2,250 2,250
Do not screen 750 3,750

Because screening requires scheduling of inspectors and equipment, the decision to

screen or not screen must be made 2 days before the potential screening takes place.
However, one item can be taken from the lot and sent to a laboratory, and its quality
(defective or non-defective) can be reported before the screen, no-screen decision must
be made. The cost of this initial inspection is P125.

a) What is the Bayes’ action without looking at the single item?

b) How much money can be paid for “perfect information”
c) What is the optimal expected cost if the quality of items is determined before the
screen, no-screen decision is made?
d) What is the Bayes’ action if the quality of one of the items is determined and found to
be defective?

CASE II

In a manufacturing process, lot having 8,10, 12, or 14% defectives are produced
according to the respective probabilities 0.4, 0.3, 0.25, and 0.05. Three customers have
contracts to receive lots from the manufacturer. The contracts specify that percentages of
defective in lots shipped to customers A, B, C should not exceed 8, 12, 14, respectively. If
a lot has a higher percentage of defectives than stipulated, a penalty of Php 100 per
percentage point is incurred. On the other hand, supplying better quality than required
costs the manufacturer Php 50 per percentage point. A sample of rise n = 20 is inspected
before each lot is shipped lot to customers. Suppose that 4 defectives are found in the
sample. What would be optimal decision of the firm?

CASE III

A new type of airplane is to be purchased by the Air force, and the number of spare
engines accompanying the order must be determined. The Air Force must order these
spare engines in bathes of 5 and can only choose among 15, 20, or 25 spares. The
supplier of these engines has two plants, and the Air Force must make its decision prior to
knowing which plant will be used. From past experience it is known that the number of
spare engines required when production takes place at plant A is approximated by a
Poisson distribution with parameter  = 21, whereas the number of spare engines when
production takes place at plant B is approximated by Poisson distribution with parameter 
= 24. The cost of a spare engine purchased now is Php 400,000, whereas the cost of
spare engine purchased at a later date is Php900,000. Spares must always be supplied if
they are demanded, and unused engines will be scrapped when the airplanes become
obsolete. Holding costs and interest are to be neglected. From these data, the total costs
(negative payoffs) have been computed as follows:

States of Nature
Actions  = 21 = 24
7
Order 15 1.155 x 10 1.414 x 107
Order 20 1.012 x 107 1.207 x 107
7
Order 25 1.047 x 10 1.135 x 107

The Air Force knows from past experience that 2/3 of all types of airplane engines
are produced in plant A and only 1/3 in plant B. Furthermore, it is known that a similar
type of engine was produced for an earlier version for the current airplane under
consideration. The order size for this earlier type was the same as for the current model.
Furthermore, its non-obsolete life is identical with that planned for the present version.
The engine for the current order will be produced in the same plant as the previous
model, although the Air Force is not aware of which of the two plant this is. The reason for
this lack of knowledge is due to the haste in which the spare engine decision must be
made. The Air Force has access to the data which the spare engine decision must be
made. The Air Force has access to the data on the number of spares actually required for
the older version (which had a Poisson distribution), but it does not have time determine
the production location.

a. What is the Bayes’ action, assuming that the information on the old airplane model is
unavailable?
b. How much money can be paid for “perfect information”?
c. Assuming that cost of data on the old airplane model is free and 30 spares were
required, determine the Bayes’ action.

CASE IV

Carlo Steerio is a chain of retail outlets specializing in sound equipment. Because of

its high volume, Charles stocks its inventory of stereo pickup cartridges by ordering lots of
100 units from various manufacturers. Carlo Hagano, the owner, decides randomly
whether to accept or reject a particular lot from a supplier, KICO. Each cartridge in a
rejected lot is thoroughly inspected by Carlo, and cartridges that are found to be defective
are replaced by the supplier. An accepted lot is parceled without inspection to the retail
stores for sale to customers, who are relied on to find the defective cartridge, which
Charles replaces without charge from its inventory. KICO will not give Carlo credit for
cartridges that have been used by retail customers, even if they were originally defective.

One of Carlo’s employees suggested that the company sample incoming lots, using
the information thereby obtained as a basis of deciding whether a lot should be accepted
or rejected. Carlo is skeptical about the adventures of the sampling, since the test costs
Php 300 per cartridge tested.

Detailed records of KICO cartridge received by Carlo have been maintained, and the
frequencies of lot proportions defective have been found. These frequencies serve as
estimates of the following prior for values of p of .10, .20 and .30

Possible lot Proportion Defective p Prior Probability

0.10 0.40
0.20 0.30
0.30 0.30
Total 1.00

The cartridges are bought at P150 each by Carlo Steerio from KICO and sold for Php
20,000 per lot customers.

Carlo calls on you to evaluate his employee’s suggestion but tells you that if sampling is
beneficial, he is noting to sample up to 2 only.

What would your recommendation be?

CASE V
Artex Computers is going to purchase 10,000 units of a certain part that is to be
assembled into the Artex products. The order is to be placed with the lowest bidder with
the 10,000 units to be delivered at a rate of 1,000 per month over the next 10 months.

The Frank Machine Shop (FMS) is considering bidding on this order. Two factors are
puzzling Mr. Frank in his attempt to fix a bid price. The first factor deals with FMS’s
chances of winning the bid. Mr. Frank finally decides to consider only two bids - either Php
12 per unit or Php 13 per unit. He estimates that the chances are two thirds of winning
the former price and one third of winning the latter price.

The second factor involved in the decision is the FMS unit manufacturing cost. Two
production processes are available. The first, process A, is known to cost Php 10 per unit.
The second, process B is one that FMS has not used before. The chief foreman says there
is a one-fourth chance that the per unit cost will be P9; a one-half chance that the cost
will be P10; and a one-fourth chance that the cost will be P11, if process B is used.

The chief foreman has suggested that he conduct an experiment with the new
process (B). He could produce 10 or 15 units; and from the experience gained, he
believes he could estimate unit cost “ pretty well”. The cost of this test would be P500.
Then asked to be more specific about how accurate his estimate of cost would be, the
chief foreman provided the following table.

Chance of Various Estimated Cost

Foreman’s Actual per unit cost
estimated Php 9 Php 10 Php 11
cost
Php 9 0.8 0.1 0.1
Php 10 0.1 0.8 0.1
Php 11 0.1 0.1 0.8
1 1 1
Mr. Frank is not sure that this information is at all relevant to his problem. The
controller has argued that the test suggested by the foreman may be valuable but should
be performed after the bid is awarded. Otherwise, he argues, the firm may spend Php 500
and not win the contract. The foreman feels that the test should be done before the bid,
since it may influence FMS’s bid price.

Draw the complete Decision tree diagram. What specific sequence of actions should
be FMS take?

CASE VI

Two batches of a product have had their labels torn off in error. One batch was
produced on Machine A. The proportion of defectives of A is given by the following density
function.
 80,000 p
f ( PA )  , 0  p  0.015
9
 0, otherwise

The other was produced by machine B whose density function given by

f ( PA )  10,125 p 2 , 0  p  1 / 15
 0, otherwise

The output of Machine B is sol a “ as is” to a user who incurs very little cost when a
defective unit is found. However, Machine A output is sold to a user who charges the
manufacturer P5 for each defective unit, since it is installed in a complex piece of
machinery. There are 100 units in each batch. The manufacturer must find out which
batch is which, as he has an urgent order for a batch of Machine A output.

A statistician has suggested has suggested taking a sample of 20 units from one batch. If
there are one or more defectives in that batch, label it “Machine B output” while if there
are zero defectives, label it “Machine A output”; and label the other batch in the opposite
manner. Then ship the batch labeled
“ Machine A output.”

a. Compute the conditional probabilities of misleading the batches.

b. Assuming that the two machines have been used with the same frequency (based on
historical data), compute the unconditional expected cost of the statisticians’s
suggestion.
c. Suppose taking a sample of 20 costs P 2, but a complete batch of 100 units would cost
only P 10. Based on your results from (b), should the manufacturer take 100% sample
of one of the batches?