Decision Analysis

Industrial Engineering University of Brawijaya

SENSITIVITY ANALYSIS
2
5
/
1
1
/
2
0
1
3

D
e
c
i
s
i
o
n

A
n
a
l
y
s
i
s

-

G
a
n
j
i
l

2
0
1
3

2
EAGLE AIRLINES
Dick Carothers, president of Eagle Airlines, had been considering
expanding his operation, and now the opportunity was available.
An acquaintance had put him in contact with the president of a small
airline in the Midwest that was selling an airplane.
Many aspects of the situation needed to be thought about, however,
and Carothers was having a hard time sorting them out.
Eagle Airlines owned and operated three twin-engine aircraft.
With this equipment, Eagle provided both charter flights and scheduled
commuter service among several communities in the eastern United
States.
Scheduled flights constituted approximately 50% of Eagle's flights,
averaging only 90 minutes of flying time and a distance of some 300
miles.
The remaining 50% of flights were chartered.
The mixture of charter flights and short scheduled flights had proved
profitable, and Carothers felt that he had found a niche for his company.
He was aching to increase the level of service, especially in the area of
charter flights, but this was impossible without more aircraft.
2
5
/
1
1
/
2
0
1
3

D
e
c
i
s
i
o
n

A
n
a
l
y
s
i
s

-

G
a
n
j
i
l

2
0
1
3

3
EAGLE AIRLINES
A Piper Seneca was for sale at a price of $95,000, and Carothers
figured that he could buy it for between $85,000 and $90,000.
This twin-engine airplane had been maintained according to FAA
regulations.
In particular, the engines were almost new, with only 150 hours of
operation since a major overhaul.
Furthermore, having been used by another small commercial
charter service, the Seneca contained all of the navigation and
communication equipment that Eagle required.
There were seats for five passengers and the pilot, plus room for
baggage.
Typical airspeed was approximately 175 nautical miles per hour
(knots), or 200 statute miles per hour (mph).
Operating cost was approximately $245 per hour, including fuel,
maintenance, and pilot salary.
Annual fixed costs included insurance ($20,000) and finance charges.
2
5
/
1
1
/
2
0
1
3

D
e
c
i
s
i
o
n

A
n
a
l
y
s
i
s

-

G
a
n
j
i
l

2
0
1
3

4
EAGLE AIRLINES
Carothers figured that he would have to borrow some 40% of
the money required, and he knew that the interest rate would
be two percentage points above the prime rate (currently
9.5% but subject to change).
Based on his experience at Eagle, Carothers knew that he
could arrange charters for $300 to $350 per hour or charge a
rate of approximately $100 per person per hour on scheduled
flights.
He could expect on average that the scheduled flights would
be half full.
He hoped to be able to fly the plane for up to 1000 hours per
year, but realized that 800 might be more realistic.
In the past his business had been approximately 50% charter
flights, but he wanted to increase that percentage if possible.
2
5
/
1
1
/
2
0
1
3

D
e
c
i
s
i
o
n

A
n
a
l
y
s
i
s

-

G
a
n
j
i
l

2
0
1
3

5
EAGLE AIRLINES
The owner of the Seneca has told Carothers that he would either sell
the airplane outright or sell Carothers an option to purchase it
within a year at a specified price. (The current owner would
continue to operate the plane during the year.)
Although the two had not agreed on a price for the option, the
discussions had led Carothers to believe that the option would cost
between $2500 and $4000.
Of course, he could always invest his cash in the money market and
expect to earn about 8%.
As Carothers pondered this information, he realized that many of
the numbers he was using were estimates.
Furthermore, some were within his control (for example, the
amount financed and prices charged) while others, such as the cost
of insurance or the operating cost, were not.
How much difference did these numbers make? What about the
option? Was it worth considering? Last, but not least, did he really
want to expand the fleet? Or was there something else that he
should consider?
2
5
/
1
1
/
2
0
1
3

D
e
c
i
s
i
o
n

A
n
a
l
y
s
i
s

-

G
a
n
j
i
l

2
0
1
3

6
Sensitivity Analysis: A Modeling
Approach
Sensitivity analysis answers the question, "What makes a
difference in this decision?
Requisite decision model is a model whose form and content
are just sufficient to solve a particular problem.
The issues that are addressed in a requisite decision model are
the ones that matter, and those issues left out are the ones that
do not matter.
Determining what matters and what does not requires
incorporating sensitivity analysis throughout the modeling
process.
2
5
/
1
1
/
2
0
1
3

D
e
c
i
s
i
o
n

A
n
a
l
y
s
i
s

-

G
a
n
j
i
l

2
0
1
3

7
Problem Identification and Structure
A decision-analysis process flowchart
2
5
/
1
1
/
2
0
1
3

D
e
c
i
s
i
o
n

A
n
a
l
y
s
i
s

-

G
a
n
j
i
l

2
0
1
3

8
Problem Identification and
Structure
Solving the wrong problem sometimes is called an "error of
the third kind.
An error of the third kind, or Type III error, implies that the
wrong question was asked; in terms of decision analysis, the
implication is that an inappropriate decision context was used,
and hence the wrong problem was solved.
How can one avoid a Type III error?
The best solution is simply to keep asking whether the problem
on the surface is the real problem.
Is the decision context properly specified?
What exactly is the "unscratched itch" that the decision maker
feels?
2
5
/
1
1
/
2
0
1
3

D
e
c
i
s
i
o
n

A
n
a
l
y
s
i
s

-

G
a
n
j
i
l

2
0
1
3

9
Problem Identification and
Structure
In the case of Eagle Airlines, Carothers appears to be eager to
expand operations by acquiring more aircraft.
Could he "scratch his itch" by expanding in a different direction?
In particular, even though he, like many pilots, may be dedicated
to the idea of flying for a living, it might be wise to consider the
possibility of helping his customers to communicate more
effectively at long distance.
To some extent, efficient communication channels such as those
provided by computer links and facsimile service, coupled with an
air cargo network, can greatly reduce the need for travel.
Pursuing ideas such as these might satisfy Carothers's urge to
expand while providing a more diversified base of operations.
So the real question may be how to satisfy Carothers's desires for
expansion rather than simply how to acquire more airplanes.
2
5
/
1
1
/
2
0
1
3

D
e
c
i
s
i
o
n

A
n
a
l
y
s
i
s

-

G
a
n
j
i
l

2
0
1
3

10
Problem Identification and
Structure
A decision situation might be represented in a variety of
different ways.
Sensitivity analysis can aid the resolution of the problem of
multiple representations by helping to identify the
appropriate perspective on the problem as well as by
identifying the specific issues that matter to the decision
maker.
2
5
/
1
1
/
2
0
1
3

D
e
c
i
s
i
o
n

A
n
a
l
y
s
i
s

-

G
a
n
j
i
l

2
0
1
3

11
Problem Identification and
Structure
Is problem structuring an issue in the Eagle Airlines case? In
this case, the alternatives are to purchase the airplane, the
option, or neither.
Although Carothers might consider a variety of fundamental
objectives, such as company growth or increased influence in
the community, in the context of deciding whether to
purchase the Seneca, it seems reasonable for him to focus on
one objective: maximize profit.
Carothers could assess the probabilities associated with the
various unknown quantities such as operating costs, amount
of business, and so on.
Thus, it appears that a straightforward decision tree or
influence diagram may do the trick.
2
5
/
1
1
/
2
0
1
3

D
e
c
i
s
i
o
n

A
n
a
l
y
s
i
s

-

G
a
n
j
i
l

2
0
1
3

12
Problem Identification and Structure
Influence diagram representing the Eagle Airlines decision
2
5
/
1
1
/
2
0
1
3

D
e
c
i
s
i
o
n

A
n
a
l
y
s
i
s

-

G
a
n
j
i
l

2
0
1
3

13
Problem Identification and Structure
Input variables and ranges of possible values for Eagle Airlines aircraft-purchase decision.
2
5
/
1
1
/
2
0
1
3

D
e
c
i
s
i
o
n

A
n
a
l
y
s
i
s

-

G
a
n
j
i
l

2
0
1
3

14
Problem Identification and
Structure
2
5
/
1
1
/
2
0
1
3

D
e
c
i
s
i
o
n

A
n
a
l
y
s
i
s

-

G
a
n
j
i
l

2
0
1
3

15
The annual profit would be the total annual revenue minus
the total annual cost
Problem Identification and
Structure
Thus, using the base values, Carothers's annual profit is
estimated to be $230,000 - $220,025 = $9,975.
This represents a return of approximately 19% on his
investment of $52,500 (60% of the purchase price).
2
5
/
1
1
/
2
0
1
3

D
e
c
i
s
i
o
n

A
n
a
l
y
s
i
s

-

G
a
n
j
i
l

2
0
1
3

16
Sensitivity to Probabilities
Modelling the uncertainty surrounding the critical variables:
Capacity of Scheduled Flights
Operating Cost
Hours Flown
Charter Price a decision variable set by Carothers.

Example:
Assume that, in an initial attempt to model the uncertainty,
Carothers chooses two values for each variable, one representing
an optimistic and one a pessimistic scenario.
2
5
/
1
1
/
2
0
1
3

D
e
c
i
s
i
o
n

A
n
a
l
y
s
i
s

-

G
a
n
j
i
l

2
0
1
3

17
Sensitivity to Probabilities
Influence diagram of Eagle Airlines decision. Note that only three variables are considered to be
uncertain, and that Hours Flown and Capacity are considered to be probabilistically dependent.
2
5
/
1
1
/
2
0
1
3

D
e
c
i
s
i
o
n

A
n
a
l
y
s
i
s

-

G
a
n
j
i
l

2
0
1
3

18
Sensitivity to Probabilities
Decision tree for Eagle Airlines with uncertainty for three variables. Profit is calculated with
all other variables held at their base values.
2
5
/
1
1
/
2
0
1
3

D
e
c
i
s
i
o
n

A
n
a
l
y
s
i
s

-

G
a
n
j
i
l

2
0
1
3

19
Sensitivity to Probabilities
A relevance arc leads from "Capacity of Scheduled Flights" to
"Hours Flown" in the influence diagram, and in the decision
tree the value for r = P(Low Hours | Low Capacity) may not be
the same as the value for s = P(Low Hours | High Capacity).
In fact, our argument suggests that r will be greater than s.
On the other hand, Operating Cost is judged to be
independent of the other variables.
2
5
/
1
1
/
2
0
1
3

D
e
c
i
s
i
o
n

A
n
a
l
y
s
i
s

-

G
a
n
j
i
l

2
0
1
3

20
Sensitivity to Probabilities
Assessing some values for probabilities p, q, r, and s.
Suppose that Carothers is comfortable with an assessment
that p = 0.5, or that Operating Cost is just as likely to be high
($253) as low ($237).
Suppose that Carothers feels that a reasonable way to
represent the dependence between Hours and Capacity is to
let s be 80% of r.
If Capacity is high (55%), then the probability that Hours = 650 is
only 80% of the probability that Hours = 650 when Capacity is
low.
With these two specifications, we now have only two
unspecified probabilities left to consider, q and r.
2
5
/
1
1
/
2
0
1
3

D
e
c
i
s
i
o
n

A
n
a
l
y
s
i
s

-

G
a
n
j
i
l

2
0
1
3

21
Sensitivity to Probabilities
Eagle Airlines' decision tree with probabilities substituted for p and s. This decision tree is
now ready for a two-way sensitivity analysis on q and r.
2
5
/
1
1
/
2
0
1
3

D
e
c
i
s
i
o
n

A
n
a
l
y
s
i
s

-

G
a
n
j
i
l

2
0
1
3

22
Sensitivity to Probabilities
We now can create a two-way sensitivity graph for q and r. As
with the two-way sensitivity analysis above, the graph will
show regions for which the expected value of purchasing the
Seneca is greater than investing in the money market.
To create the graph, we first write out the expected value of
purchasing the airplane in terms of q and r, including the
specifications that p = 0.5 and s = 0.8r.
This equation comes from solving the decision tree:
EMV(Purchase) = 0.5{q[-9725r - 4225(1 - r)] + (1 - q)
[6525(0.8r) + 18,275(1 - 0.8r)]} + 0.5{q[675r + 10,175(1 - r)] +
(1 - q)[16,925(0.8r) + 32,675(1 - 0.8r)]}
EMV(Purchase) = q(3500r - 22,500) - 11,000r + 25,475
2
5
/
1
1
/
2
0
1
3

D
e
c
i
s
i
o
n

A
n
a
l
y
s
i
s

-

G
a
n
j
i
l

2
0
1
3

23
Sensitivity to Probabilities
2
5
/
1
1
/
2
0
1
3

D
e
c
i
s
i
o
n

A
n
a
l
y
s
i
s

-

G
a
n
j
i
l

2
0
1
3

24
We would want to purchase the airplane if EMV(Purchase) >
4200.

q(3500r - 22,500) - 11,000r + 25,475 > 4200
25,475 - 4200 - 11,000r > q(22,500 - 3500r)
Sensitivity to Probabilities
Two-way sensitivity graph for Eagle Airlines.
2
5
/
1
1
/
2
0
1
3

D
e
c
i
s
i
o
n

A
n
a
l
y
s
i
s

-

G
a
n
j
i
l

2
0
1
3

25
Sensitivity to Probabilities
The area below the line contains points where q < (21,275 -
ll,000r)/(22,500 - 3500r); for these (q, r) points, EMV(Purchase)>$4200.
The graph makes sense because q and r are probabilities of the
pessimistic scenarios low Capacity and low number of Hours Flown.
If Carothers thinks that the pessimistic scenarios are likely (q and r close
to 1), then he would not want to buy the airplane.
Carothers may not have especially firm ideas of what the probabilities q
and r should be.
Suppose, for example, that in the process of coming up with the
probabilities he feels that q could be between 0.4 and 0.5 and that r
could be between 0.5 and 0.65.
These probabilities are represented by the points inside Rectangle A in Figure
5.8. All of these points fall within the "Purchase Seneca" region, and so the
conclusion is that the Seneca should be purchased.
The decision is not sensitive to the assessment of the probabilities.
If, on the other hand, Carothers thinks that reasonable values of q and r
fall in Rectangle B, then the optimal choice is not clear.
In this situation, he could reflect on the chances associated with Capacity and
Hours Flown and try to refine his model of the uncertainty.
2
5
/
1
1
/
2
0
1
3

D
e
c
i
s
i
o
n

A
n
a
l
y
s
i
s

-

G
a
n
j
i
l

2
0
1
3

26
Sensitivity to Probabilities
The value of the two-way sensitivity graph is to provide
guidance in determining how much effort is needed to model
uncertainty in a decision problem.
Looking at it another way, the graph can reveal whether the
decision is sensitive to the uncertainty in the problem and to
the modeling of that uncertainty.
2
5
/
1
1
/
2
0
1
3

D
e
c
i
s
i
o
n

A
n
a
l
y
s
i
s

-

G
a
n
j
i
l

2
0
1
3

27
Sensitivity Analysis in Action
2
5
/
1
1
/
2
0
1
3

D
e
c
i
s
i
o
n

A
n
a
l
y
s
i
s

-

G
a
n
j
i
l

2
0
1
3

28
Sensitivity Analysis in Action
Two-way sensitivity analysis for the heart disease treatment decision.
Source: Macartney et al. (1984).
2
5
/
1
1
/
2
0
1
3

D
e
c
i
s
i
o
n

A
n
a
l
y
s
i
s

-

G
a
n
j
i
l

2
0
1
3

29
Warming Up
An orange grower in Florida faces a dilemma. The weather forecast
is for cold weather, and there is a 50% chance that the temperature
tonight will be cold enough to freeze and destroy his entire crop,
which is worth some $50,000. He can take two possible actions to
try to alleviate his loss if the temperature drops. First, he could set
burners in the orchard; this would cost $5000, but he could still
expect to incur damage of approximately $15,000 to $20,000.
Second, he could set up sprinklers to spray the trees. If the
temperature drops, the water would freeze on the fruit and provide
some insulation. This method is cheaper ($2000), but less effective.
With the sprinklers he could expect to incur as much as $25,000 to
$30,000 of the loss with no protective action.
Compare the grower's expected values for the three alternatives he
has, considering the various possible loss scenarios for the burners
and the sprinklers. Which alternative would you suggest the grower
take? Why?
2
5
/
1
1
/
2
0
1
3

D
e
c
i
s
i
o
n

A
n
a
l
y
s
i
s

-

G
a
n
j
i
l

2
0
1
3

30
NEXT . . .
2
5
/
1
1
/
2
0
1
3

D
e
c
i
s
i
o
n

A
n
a
l
y
s
i
s

-

G
a
n
j
i
l

2
0
1
3

31
Creativity and Decision Making
Theories of Creativity
Chains of Thought
Phases of the Creative Process
2
5
/
1
1
/
2
0
1
3

D
e
c
i
s
i
o
n

A
n
a
l
y
s
i
s

-

G
a
n
j
i
l

2
0
1
3

32