Sensitivity Analysis

Using SensIt
4
4.1 SENSIT OVERVIEW
SensIt is a sensitivity analysis add-in for Windows Excel 2010 & 2013 & 2016 & 2019 & 365 and
Mac Excel 2016 & 2019 & 365.
SensIt performs sensitivity analysis on your worksheet what-if model and creates simple plots,
spider charts, and tornado charts. Sensitivity analysis can help you determine which of your input
assumptions are critical.
Before using the SensIt options, you must have a spreadsheet model with one or more inputs and
an output. With SensIt it is easy to see how sensitive the output is to changes in the inputs.
In Windows Excel and Mac Excel, SensIt adds two buttons to the TreePlan ribbon: SensIt Plot
What-If and SensIt Tornado.
Use SensIt’s Plot What-If feature to see how your model’s output depends on changes in a single
input variable. This feature creates a table and an XY Scatter chart type.
Use SensIt’s Tornado feature to see how your model’s output depends on ranges you specify for
each of the model’s input variables. This feature creates tornado charts (an Excel Bar chart type)
and a spider chart (an XY Scatter chart type). The spider chart is useful for detecting a nonlinear
relationship between the input variables and the output variable of your model. The two-factor
tornado chart can help you determine how your model output is affected by each pair of your input
assumptions.
All of SensIt’s functionality, including its built-in help, is a part of the SensIt XLAM file. There is
no separate setup file or help file. When you use SensIt on a Windows computer, it does not create
any Windows Registry entries (although Excel may use such entries to keep track of its add-ins).

4.2 EXAMPLE PROBLEM

This example is a simple economic model of a prospective new product taken from the
Eschenbach (1992) article. The model assumes that the FirstCost investment occurs at the start of
the first year and all subsequent cash flows are end-of-year. The product life is N years with the
Salvage value at the end of year N. At the start of the project there is no revenue for the first
NNoRev years with subsequent annual Revenue during years NNoRev+1 through N. FracComp is
the fraction of Revenue lost to competition. Annual operations and maintenance cost O&M
applies for all N years of product life. The discount rate i is used to determine PresentWorth.
Eschenbach uses the notation P/A,i,N and P/F,i,N, respectively, for the uniform periodic and the
38 Chapter 4 Sensitivity Analysis Using SensIt

single payment present worth factors. The spreadsheet model includes some intermediate variables
for determining various discount factors.

Figure 4.1 Model Display

A B
1 Input Variables Input Cells
2 FirstCost $120,000
3 Salvage $20,000
4 N 12
5 i 10%
6 O&M $6,000
7 Revenue $55,000
8 NNoRev 1
9 FracComp 0.2
10 Intermediate Variables
11 P/F,i,N $0.32
12 P/A,i,N $6.81
13 N-NNoRev 11
14 P/A,i,N-NNoRev $6.50
15 P/F,i,NNoRev $0.91
16 Revenue Discount $5.90
17 Output Variable
18 Present Worth $105,293

Figure 4.2 Model Formulas

A B
10 Intermediate Variables
11 P/F,i,N =1/(1+B5)^B4
12 P/A,i,N =PV(B5,B4,-1)
13 N-NNoRev =B4-B8
14 P/A,i,N-NNoRev =PV(B5,B13,-1)
15 P/F,i,NNoRev =1/(1+B5)^B8
16 Revenue Discount =B14*B15
17 Output Variable
18 Present Worth =-B2+B3*B11-B6*B12+B7*(1-B9)*B16

4.3 ONE INPUT, ONE OUTPUT

Use SensIt’s Plot What-If feature to see how your model’s output depends on changes in a single
input variable.
In Windows Excel and Mac Excel, choose TreePlan > Plot What-If.
SensIt’s One Input, One Output dialog box appears.
 4.3 One Input, One Output 39

Figure 4.3 SensIt Plot What-If Dialog Box

Cells for Input Variable

In the Label reference edit box, type a cell reference, or point to the cell containing a text label and
click. In the Value reference edit box, type a cell reference, or point to the cell containing a
numeric value that is an input cell of your model.

Cells for Output Variable

In the Label reference edit box, type a cell reference, or point to the cell containing a text label and
click. In the Value reference edit box, type a cell reference, or point to the cell containing a
formula that is the output of your model.

Input Values
Type numbers in the Start, Step, and Stop edit boxes to specify values to be used in the input
variable’s cell. Cell references are not allowed.
Click OK: SensIt uses the Start, Step, and Stop values to prepare a table of values. Each value is
copied to the input variable Value cell, the worksheet is recalculated, and the value of the output
variable Value cell is copied to the table.
(You could do this manually in Excel using the Edit > Fill > Series and Data > What-If Analysis >
Data Table commands.) SensIt uses the paired input and output values to prepare an XY (Scatter)
chart. The text in the label cells you identified are used as the chart’s axis labels. (You could do
this manually using Insert > Chart.)
40 Chapter 4 Sensitivity Analysis Using SensIt

Figure 4.4 SensIt Numerical and Chart Output

SensIt
One Input, One Output
SensIt
Date (current date) $200,000
Time (current time)
Workbook (file name) $180,000
Input Cell OneInput!$B$4
$160,000
Output Cell OneInput!$B$18
$140,000
N Present Worth
6 $16,789 $120,000
Present Worth
7 $35,263
8 $52,057 $100,000
9 $67,325
$80,000
10 $81,204
11 $93,822 $60,000
12 $105,293
13 $115,721 $40,000
14 $125,201
15 $133,819 $20,000
16 $141,654
$0
17 $148,776
18 $155,251 6 8 10 12 14 16 18 20 22 24
19 $161,137 N
20 $166,488
21 $171,353
22 $175,775
23 $179,796
24 $183,451

Insight From the table and chart, we observe a nonlinear relationship between Present Worth and
product life N. That is, the change in Present Worth between years 6 and 7 is much greater than
the change between years 23 and 24.

4.4 MANY INPUTS, ONE OUTPUT TORNADO

Use SensIt’s Tornado feature to see how your model’s output depends on ranges you specify for
each of the model’s input variables. Before using this option, arrange your model input cells in
adjacent cells in a single column, arrange corresponding labels in adjacent cells in a single
column, and arrange Low, Base, and High input values for each input variable in three separate
columns. Alternatively, the three columns containing input values can be worst case, likely case,
and best case. An appropriate arrangement is shown below.
 4.4 Many Inputs, One Output Tornado 41

Figure 4.5 Model Display with Lower and Upper Bounds

A B C D E F G H I
1 Input Variables Input Cells OneExtreme BaseCase OtherExtreme LowerLimit% UpperLimit%
2 FirstCost $120,000 $108,000 $120,000 $180,000 90% 150%
3 Salvage $20,000 $0 $20,000 $30,000 0% 150%
4 N 12 6 12 24 50% 200%
5 i 10% 6% 10% 20% 60% 200%
6 O&M $6,000 $4,800 $6,000 $7,500 80% 125%
7 Revenue $55,000 $33,000 $55,000 $68,750 60% 125%
8 NNoRev 1 0 1 3 0% 300%
9 FracComp 0.2 0 0.2 0.4 0% 200%
10 Intermediate Variables (Input Values for SensIt) (Eschenbach's Inputs
11 P/F,i,N $0.32 Relative To Base Case)
12 P/A,i,N $6.81
13 N-NNoRev 11
14 P/A,i,N-NNoRev $6.50
15 P/F,i,NNoRev $0.91
16 Revenue Discount $5.90
17 Output Variable
18 Present Worth $105,293

Eschenbach expressed the ranges as lower and upper limit percentages of the base case. To use
SensIt, those percentages have been multiplied by the base case so that the OneExtreme and
OtherExtreme values are in the same measurement units as the BaseCase values. For more
information about choosing input values, refer to the “Tips for Many Inputs, One Output” section.
In Windows Excel and Mac Excel, choose TreePlan > Tornado. SensIt’s Many Inputs, One Output
dialog box appears.

Figure 4.6 SensIt Many Inputs, One Output Dialog Box

Ranges for Input Variables

Type a range reference, or point to the range (click and drag) containing text labels and the range
containing numeric values that are inputs to your model. If the range is not contiguous, select the
42 Chapter 4 Sensitivity Analysis Using SensIt

first portion and then hold down the Control key (Windows) or Command key (Mac) while
making the remaining selections. Alternatively, type a comma between each portion.

Cells for Output Variable

Type a cell reference, or point to the cell containing a text label and the cell containing a formula
that’s the output of your model.

Ranges for Input Values

Type a range reference, or point to the range (click and drag) containing numeric values for each
of your model’s inputs. You can make non-contiguous selections similar to the ranges for input
variables. Be sure that all five range selections have the appropriate cells in the same order.

Figure 4.7 SensIt Tornado Numerical and Chart Output

A B C D E F G H I
1 SensIt
2 Many Inputs, One Output
3 Single-Factor Sensitivity Analysis
4
5 Date (current date) Workbook (file name)
6 Time (current time) Output Cell ManyInputs!$B$18
7
8 Present Worth
9 Corresponding Input Value Output Value Percent
10 Input Variable Low Output Base Case High Output Low Base High Swing Swing^2
11 Revenue $33,000 $55,000 $68,750 $1,372 $105,293 $170,244 $168,872 25.0%
12 N 6 12 24 $16,789 $105,293 $183,451 $166,661 24.3%
13 i 20% 10% 6% $14,267 $105,293 $167,016 $152,749 20.5%
14 FracComp 0.4 0.2 0 $40,342 $105,293 $170,244 $129,901 14.8%
15 NNoRev 3 1 0 $35,871 $105,293 $145,293 $109,421 10.5%
16 FirstCost $180,000 $120,000 $108,000 $45,293 $105,293 $117,293 $72,000 4.5%
17 O&M $7,500 $6,000 $4,800 $95,072 $105,293 $113,469 $18,397 0.3%
18 Salvage $0 $20,000 $30,000 $98,920 $105,293 $108,479 $9,559 0.1%
19
20
SensIt
21
22
23 Revenue $33,000 $68,750
24
25 N 6 24
26
27 i 20% 6%

28
FracComp 0.4 0
29
30 NNoRev 3 0
31
32 FirstCost $180,000 $108,000
33
34 O&M $7,500 $4,800
35
36 Salvage $0 $30,000
37
38 ‐$40,000 $0 $40,000 $80,000 $120,000 $160,000 $200,000
39
Present Worth
40
41

For each input variable, SensIt sets all other input values at their Base case values, copies the One
Extreme input value to the input variable cell, recalculates the worksheet, and copies the value of
 4.5 Customizing SensIt Charts 43

the output variable cell to the table. The same steps are repeated using each Other Extreme input
value. For each input variable, SensIt computes the range of the output variable values (the swing),
sorts the table from largest swing down to smallest swing, and prepares a bar chart.
For each Low output value in column E, the corresponding input value is shown in column B. For
each High output value in column G, the corresponding input value is shown in column D. The
tornado chart shows single-factor sensitivity analysis, i.e., for each output value, only one input
value is changed from its base case value. The tornado chart summarizes eight separate single-
factor sensitivity analyses.
Insight The uncertainty about annual Revenue (the topmost bar in the tornado chart) is associated
with the widest swing in Present Worth. O&M and Salvage have little effect on Present Worth.

4.5 CUSTOMIZING SENSIT CHARTS

If the labels on the horizontal axis are numbers with many digits, some of the labels may wrap
around so that some of the digits display below the others. There are many ways to remedy this
anomaly.
(1) One way is to widen the chart. Click just inside the outer border of the chart so that eight chart
handles are shown and then drag the middle chart handle on the left or right to widen the chart.
(2) Another way is to select the horizontal axis (click between the labels on the horizontal axis)
and change the font size in the Font group of the Home ribbon.
(3) Yet another way is to select the horizontal axis, right-click and select Format Axis, or double-
click. Then click Axis Options in the Format Axis dialog box, and change the Minimum,
Maximum, and Major unit.
(4) Finally, you can obtain more spacing on the horizontal axis by using a custom format. If you
want to show values in thousands instead of the original units, select the axis, and use a method in
the previous step to display the Format Axis dialog box. Click Number in the Format Axis dialog
box, enter a comma at the end of the current format shown in the Format Code edit box, and click
the Add button. For example, you might change the format from "$#,##0" to "$#,##0," and then
modify the horizontal axis title to indicate that the axis values are in thousands.

4.6 TORNADO SORTED BY DOWNSIDE RISK

The tornado chart is originally sorted by Swing. In the numerical results above the chart, to sort by
downside risk, select the single row of labels and all data rows. On the Home ribbon in the Editing
group, choose Sort & Filter > Custom Sort. Check that "My data has headers" is selected, select
Column Sort by “Low,” select Sort On Cell Values, select Order Smallest to Largest, and click
OK. If necessary, press F9 (Windows) or Command+= (Mac) to refresh the chart.
The results are shown below.
44 Chapter 4 Sensitivity Analysis Using SensIt

Figure 4.8 SensIt Tornado Sorted by Downside Risk

A B C D E F G H I
8 Present Worth
9 Corresponding Input Value Output Value Percent
10 Input Variable Low Output Base Case High Output Low Base High Swing Swing^2
11 Revenue $33,000 $55,000 $68,750 $1,372 $105,293 $170,244 $168,872 25.0%
12 i 20% 10% 6% $14,267 $105,293 $167,016 $152,749 20.5%
13 N 6 12 24 $16,789 $105,293 $183,451 $166,661 24.3%
14 NNoRev 3 1 0 $35,871 $105,293 $145,293 $109,421 10.5%
15 FracComp 0.4 0.2 0 $40,342 $105,293 $170,244 $129,901 14.8%
16 FirstCost $180,000 $120,000 $108,000 $45,293 $105,293 $117,293 $72,000 4.5%
17 O&M $7,500 $6,000 $4,800 $95,072 $105,293 $113,469 $18,397 0.3%
18 Salvage $0 $20,000 $30,000 $98,920 $105,293 $108,479 $9,559 0.1%
19
20
21 SensIt
22
23 Revenue $33,000 $68,750
24
25 i 20% 6%
26
27 N 6 24

28
29 NNoRev 3 0

30
FracComp 0.4 0
31
32
FirstCost $180,000 $108,000
33
34 O&M $7,500 $4,800
35
36 Salvage $0 $30,000
37
38 ‐$40,000 $0 $40,000 $80,000 $120,000 $160,000 $200,000
39
Present Worth
40
41

4.7 TORNADO SORTED BY UPSIDE POTENTIAL

To sort by upside potential, select the single row of labels and all data rows, on the Home ribbon
in the Editing group choose Sort & Filter > Custom Sort, check that "My data has headers" is
selected, select Column Sort by “High,” select Sort On Values, select Order Largest to Smallest,
and click OK. If necessary, press F9 (Windows) or Command+= (Mac) to refresh the chart.
The results are shown below.
 4.8 Tornado Showing Major Uncertainties 45

Figure 4.9 SensIt Tornado Sorted by Upside Potential

A B C D E F G H I
8 Present Worth
9 Corresponding Input Value Output Value Percent
10 Input Variable Low Output Base Case High Output Low Base High Swing Swing^2
11 N 6 12 24 $16,789 $105,293 $183,451 $166,661 24.3%
12 Revenue $33,000 $55,000 $68,750 $1,372 $105,293 $170,244 $168,872 25.0%
13 FracComp 0.4 0.2 0 $40,342 $105,293 $170,244 $129,901 14.8%
14 i 20% 10% 6% $14,267 $105,293 $167,016 $152,749 20.5%
15 NNoRev 3 1 0 $35,871 $105,293 $145,293 $109,421 10.5%
16 FirstCost $180,000 $120,000 $108,000 $45,293 $105,293 $117,293 $72,000 4.5%
17 O&M $7,500 $6,000 $4,800 $95,072 $105,293 $113,469 $18,397 0.3%
18 Salvage $0 $20,000 $30,000 $98,920 $105,293 $108,479 $9,559 0.1%
19
20
21 SensIt
22
23 N 6 24
24
25 Revenue $33,000 $68,750
26
27 FracComp 0.4 0

28
29 i 20% 6%

30
NNoRev 3 0
31
32
FirstCost $180,000 $108,000
33
34 O&M $7,500 $4,800
35
36 Salvage $0 $30,000
37
38 ‐$40,000 $0 $40,000 $80,000 $120,000 $160,000 $200,000
39
Present Worth
40
41

4.8 TORNADO SHOWING MAJOR UNCERTAINTIES

In some situations you may have twenty or more input variables and you wish to show the
variation of only the top five or ten. To illustrate this modification, note that in this example the
variation in Present Worth due to O&M and Salvage is minimal, so consider showing only the top
six input variables, here sorted by swing. Click one of the bars on the left side of the vertical base
case line to select Series 1 (shown at the right end of the SERIES function in the formula bar), and
then click and drag the fill handle from A18 up to A16 and the fill handle from F18 up to F16.
Click one of the bars on the right side of the vertical base case line to select Series 2, and then
click and drag the fill handle from H18 up to H16. To resize the chart, click just inside its outer
border and drag the bottom center fill handle upward. The resulting chart is shown below.
46 Chapter 4 Sensitivity Analysis Using SensIt

Figure 4.10 SensIt Tornado Showing Only Major Uncertainties

SensIt

Revenue $33,000 $68,750

N 6 24

i 20% 6%

FracComp 0.4 0

NNoRev 3 0

FirstCost $180,000 $108,000

‐$40,000 $0 $40,000 $80,000 $120,000 $160,000 $200,000

Present Worth

4.9 TORNADO WITH VERTICAL LINE

A tornado chart shows ranges for a single alternative. To compare with a second alternative, it is
useful to show a vertical line representing the value of the performance measure for the second
alternative. For example, the second alternative might be "Pursue a fixed cost contract," with
present worth of $30,000.
Select the chart. In Windows Excel, on the Insert ribbon in the Illustrations group, click Shapes,
and in the Lines area click the Line icon. In Mac Excel, from Mac Excel’s topmost menu, choose
Insert > Shape > Line. On the tornado chart, click at the $30,000 value on the horizontal axis and
drag upward to above the top horizontal bar. Before releasing the mouse button, position the top
end of the line so that it is vertical. The result is shown below.
 4.10 Many Inputs, One Output Spider 47

Figure 4.11 Tornado Chart with Vertical Line

SensIt

Revenue $33,000 $68,750

N 6 24

i 20% 6%

FracComp 0.4 0

NNoRev 3 0

FirstCost $180,000 $108,000

O&M $7,500 $4,800

Salvage $0 $30,000

‐$40,000 $0 $40,000 $80,000 $120,000 $160,000 $200,000

Present Worth

To format the vertical line, select it. In Windows Excel and Mac Excel, choose Shape Format >
Shape Outline > Weight 1 pt.
Insight In this example, we note that Revenue, N, and i are critical input assumptions. These are
the three input variables that could result in Present Worth less than the " Pursue a fixed cost
contract" alternative. However, three other input assumptions are nearly as critical. We should
focus on these six inputs, either by trying to control the value, by gathering more information to
reduce the uncertainty (the width of the bars), or by treating the uncertainty explicitly using
probability assessments and subsequent Monte Carlo simulation. We should not be as concerned
about the O&M and Salvage values.

4.10 MANY INPUTS, ONE OUTPUT SPIDER

In addition to the tornado chart, SensIt’s Tornado-Spider feature also creates a spider chart to
show how your model’s output depends on the percentage changes for each of the model’s input
variables.
SensIt Spider uses the same base case and extreme input values as the single-factor sensitivity
analysis for the tornado chart, but you can specify a Step Percent for evaluating the model at
intermediate values. The results are shown with each input value expressed as a percentage of the
base case input value. If the base case value for an input variable is zero, that variable is not
included in the Spider analysis (because it would not be possible to express an extreme or
intermediate input value as a percentage of the zero base case value). SensIt prepares an XY
(Scatter) chart where the horizontal axis is each input value as a percentage of the base case input
value and the vertical axis is the associated model output value.
48 Chapter 4 Sensitivity Analysis Using SensIt

Figure 4.12 SensIt Spider Numerical and Chart Output

Present Worth
Corresponding Input Value Input Value as % of Base Output Value
Input Variable Low Output Base Case High Output Low % Base % High % Low Base High Swing
Revenue $33,000 $55,000 $68,750 60.0% 100.0% 125.0% $1,372 $105,293 $170,244 $168,872
N 6 12 24 50.0% 100.0% 200.0% $16,789 $105,293 $183,451 $166,661
i 20% 10% 6% 200.0% 100.0% 60.0% $14,267 $105,293 $167,016 $152,749
FracComp 0.4 0.2 0 200.0% 100.0% 0.0% $40,342 $105,293 $170,244 $129,901
NNoRev 3 1 0 300.0% 100.0% 0.0% $35,871 $105,293 $145,293 $109,421
FirstCost $180,000 $120,000 $108,000 150.0% 100.0% 90.0% $45,293 $105,293 $117,293 $72,000
O&M $7,500 $6,000 $4,800 125.0% 100.0% 80.0% $95,072 $105,293 $113,469 $18,397
Salvage $0 $20,000 $30,000 0.0% 100.0% 150.0% $98,920 $105,293 $108,479 $9,559

SensIt
$220,000

$200,000

$180,000

$160,000

$140,000
Revenue
$120,000
Present Worth

$100,000 i
FracComp
$80,000 NNoRev
FirstCost
$60,000
O&M
$40,000 Salvage

$20,000

$0

‐$20,000
‐50.0% 0.0% 50.0% 100.0% 150.0% 200.0% 250.0% 300.0% 350.0%

Input Value as % of Base Case

Insight On a spider chart, lines that are nearly horizontal generally indicate an input variable
where small percentage changes do not have much effect on the output value. Lines that are more
vertical indicate an input variable where small percentage changes have a greater effect on the
output value. Unlike the tornado chart, the spider chart makes it easy to identify nonlinear
relationships.

4.11 TWO-FACTOR TORNADO CHART

SensIt can also produce a two-factor tornado chart, where SensIt considers each pair of input
variables. If N is the number of input variables, there are N*(N-1)/2 pairs to evaluate. For each
pair, SensIt considers all nine combinations of the One Extreme, Base Case, and Other Extreme
input values for the two variables. For the calculation of output for each combination, the other
input variables are kept at their Base Case values. For each pair, SensIt summarizes the
calculations by showing the combinations of input values that produce the lowest and highest
output value. Finally, the pairs are sorted by swing, and a tornado chart is created.
The example has N=8 input variables, so there are N*(N-1)/2 = 8*(8-1)/2 = 28 pairs to evaluate.
SensIt’s output shows data rows and chart bars for all 28 pairs. To save space, the figure shows
numerical results for only the top ten (by hiding rows for the other 18 pairs and resizing the chart).
 4.12 Two-Factor Downside Risk 49

Figure 4.13 SensIt Two-Factor Tornado Partial Output

Present Worth
Corresponding Input Values Output Value
Input Variables Low Output Base Case High Output Low Base High Swing
N&i 6 & 20% 12 & 10% 24 & 6% -$23,599 $105,293 $320,344 $343,943
N & Revenue 6 & $33,000 12 & $55,000 24 & $68,750 -$43,863 $105,293 $272,283 $316,146
i & Revenue 20% & $33,000 10% & $55,000 6% & $68,750 -$49,197 $105,293 $248,861 $298,058
N & FracComp 6 & 0.4 12 & 0.2 24 & 0 -$21,118 $105,293 $272,283 $293,401
Revenue & FracComp $33,000 & 0.4 $55,000 & 0.2 $68,750 & 0 -$37,598 $105,293 $251,432 $289,030
N & NNoRev 6&3 12 & 1 24 & 0 -$52,632 $105,293 $223,451 $276,083
i & FracComp 20% & 0.4 10% & 0.2 6% & 0 -$25,398 $105,293 $248,861 $274,259
Revenue & NNoRev $33,000 & 3 $55,000 & 1 $68,750 & 0 -$40,281 $105,293 $220,244 $260,524
i & NNoRev 20% & 3 10% & 1 6% & 0 -$41,752 $105,293 $208,525 $250,277
FirstCost & Revenue $180,000 & $33,000 $120,000 & $55,000 $108,000 & $68,750 -$58,628 $105,293 $182,244 $240,872

SensIt

N & i 6 & 20% 24 & 6%

N & Revenue 6 & $33,000 24 & $68,750

i & Revenue 20% & $33,000 6% & $68,750

N & FracComp 6 & 0.4 24 & 0

Revenue & FracComp $33,000 & 0.4 $68,750 & 0

N & NNoRev 6 & 3 24 & 0

i & FracComp 20% & 0.4 6% & 0

Revenue & NNoRev $33,000 & 3 $68,750 & 0

i & NNoRev 20% & 3 6% & 0

FirstCost & Revenue $180,000 & $33,000 $108,000 & $68,750

‐$200,000 ‐$100,000 $0 $100,000 $200,000 $300,000 $400,000

Present Worth

Insight The single-factor analysis produces Present Worth values between $1,372 and $170,244.
The two-factor analysis produces Present Worth values between -$58,628 and $320,344. If we
used Monte Carlo simulation, we would observe even more variation in the output, equivalent to
an eight-factor sensitivity analysis.
The single-factor downside-risk tornado chart shows that the lowest Present Worth is for the
Revenue input assumption, followed by i, N, NNoRev, FracComp, and FirstCost. The two-factor
analysis shows that the FirstCost and Revenue pair could produce the lowest Present Worth. So it
is particularly important to find a way to avoid the high FirstCost and low Revenue combination,
even though the single-factor analysis indicates that FirstCost by itself is not so important.

4.12 TWO-FACTOR DOWNSIDE RISK

In the following chart the two-factor results are sorted by the "Low" output value to show
downside risk. Instead of showing all 28 results, the figure shows only the top twelve (by hiding
rows for the other 16 pairs of inputs and resizing the chart).
The results are shown below.
50 Chapter 4 Sensitivity Analysis Using SensIt

Figure 4.14 SensIt Two-Factor Tornado Sorted by Downside Risk

SensIt

FirstCost & Revenue $180,000 & $33,000 $108,000 & $68,750

N & NNoRev 6 & 3 24 & 0

i & Revenue 20% & $33,000 6% & $68,750

FirstCost & i $180,000 & 20% $108,000 & 6%

N & Revenue 6 & $33,000 24 & $68,750

FirstCost & N $180,000 & 6 $108,000 & 24

i & NNoRev 20% & 3 6% & 0

Revenue & NNoRev $33,000 & 3 $68,750 & 0

Revenue & FracComp $33,000 & 0.4 $68,750 & 0

i & FracComp 20% & 0.4 6% & 0

FirstCost & NNoRev $180,000 & 3 $108,000 & 0

N & i 6 & 20% 24 & 6%

‐$200,000 ‐$100,000 $0 $100,000 $200,000 $300,000 $400,000

Present Worth

Insight From the one-factor analyses, we note that the inputs associated with maximum swing are
Revenue, N, and i, in that order. When sorted by downside risk, the ordered inputs are Revenue, i,
and N. Although those three inputs are included in the top ten pairs of the two-factor analysis
shown above, the top two pairs indicate that FirstCost is an important input when combined with
Revenue, and NNoRev is an important input when combined with N.

4.13 COMPARISON WITH OVERALL VARIATION

Using swing as the criterion for measuring variation of the output value, the single most important
input variable is Revenue with output variation from $1,372 to $170,244, a swing of $168,872.
With two-factor sensitivity analysis, the two most important input variables are N and i with
output variation from -$23,599 to $320,344, a swing of $343,943.
When all input assumptions are varied (computations are not shown here), the output varies
from -$180,805 to $702,005, a swing of $882,810. The figure shows the output variation for these
three situations. For example, if you use triangular density functions for each input variable and
perform Monte Carlo simulation with a very large number of trials, the results will show similar
variation.
 4.14 Tips for Many Inputs, One Output 51

Figure 4.15 Comparison of Variation of Output Values

Comparison of Variation of Output Values

Eight-Factor

Tw o-Factor

One-Factor

-$200,000 $0 $200,000 $400,000 $600,000 $800,000

Present Worth

4.14 TIPS FOR MANY INPUTS, ONE OUTPUT

When defining the high and low cases for each variable, it is important to be consistent so that the
"high" cases are all equally high and the "low" cases are all equally low. This will ensure that the
output results can be meaningfully compared.
For example, if you are thinking about uncertainty using probability and very extreme values are
possible but with low probability of occurrence, you might take all of the base case values to be
estimates of the mean of the input variable, take low cases to be values such there is a 1-in-10
chance of the variable being below this amount, and take the high cases to be values such that
there is a 1-in-10 chance of the variable being above this amount. Or, you might use the 5th and
95th percentiles for each of the input variables.
Alternatively, in some situations the values for each input variable may have lower and upper
bounds, so you may specify low and high values that are the absolute lowest and highest possible
values.
When you click OK in the Many Inputs, One Output dialog box, SensIt sets all of the input
variables to their base-case values and records the output value. Then SensIt goes through each of
the input variables one at a time, plugs the low-case value into the input cell, and records the value
in the output cell. It then repeats the process for the high case. For each substitution, all input
values are kept at their base-case values except for the single input value that is set at its low or
high value. SensIt then produces a spreadsheet that lists the numerical results as shown in columns
E, F, and G of the worksheet with the tornado chart.
In the worksheet, the variables are sorted by their "swing" -- the absolute value of the difference
between the output values in the low and high cases. "Swing" serves as a rough measure of the
impact of each input variable. The rows of numerical output are sorted from highest swing at the
top down to lowest swing at the bottom. Then SensIt creates a bar chart of the sorted data.
In general, you should focus your modeling efforts on those variables with the greatest impact on
the value measure.
If your model has input variables that are discrete or categorical, you should create multiple
tornado charts using different base case values of that input variable. For example, if your model
has an input variable "Government Regulation" that has possible values 0 (zero) or 1, the low and
high values will be 0 and 1, but you should run one tornado chart with base case = 0 and another
tornado chart with base case = 1.
52 Chapter 4 Sensitivity Analysis Using SensIt

4.15 INTERPRETATION OF PERCENT SWING^2

A tornado chart displays the results of single-factor sensitivity analysis for an alternative. For a
risk averse decision maker, the certain equivalent of an alternative is less than the expected value
(i.e., probability-weighted average) of the payoff distribution. Each Percent Swing^2 value
indicates how the difference between the expected value and the certain equivalent is affected by
the uncertainty of a specific input variable. These approximations are based on the following
assumptions:
1. Exponential Utility. The decision maker's attitude toward risk is described by an
exponential utility function, with a Risk Tolerance parameter indicating the degree of risk
aversion. In this case the certain equivalent of a payoff distribution can be approximated
by
Certain Equivalent = Expected Value – Variance/(2*Risk Tolerance)
The variance of each input variable contributes to the variance of the payoff distribution,
which determines how much the certain equivalent is less than the expected value.
2. Independence. The input variables should be probabilistically independent, in which case
the variance of the payoff distribution is the sum of the variances of the input variables
(for small variations and perhaps a weighted sum, depending on the payoff model). This
assumption of independence is also necessary for single-factor sensitivity analysis to
provide meaningful information.
3. Symmetric Probability Distributions. The approximation of certain equivalent based on
the variance is most accurate for uncertain input assumptions specified by symmetric,
bell-shaped probability density functions. As an example, for an input distribution that
has a normal probability distribution, the standard deviation is proportional to the width
(high extreme minus low extreme) specified as inputs to the single-factor sensitivity
analysis, and the variance is proportional to the width squared. For an unbounded
symmetric probability distribution, the extremes are often specified using the 0.05 and
0.95 fractiles or the 0.10 and 0.90 fractiles.
For a more extensive discussion, see McNamee and Celona (2001).

4.16 ACKNOWLEDGEMENT
The original version of SensIt was written using the Excel 4 XLM macro programming language
by Mike Middleton of the University of San Francisco and Jim Smith of Duke University. During
2005-2020, Mike Middleton used the Excel VBA programming language to rewrite the original
version, and he added spider charts and two-factor tornado charts.
 4.17 References 53

4.17 REFERENCES
Eschenbach, Ted G. "Spiderplots versus Tornado Diagrams for Sensitivity Analysis," Interfaces
(An International Journal of The Institute of Management Sciences and the Operations Research
Society of America), Special Issue: Decision and Risk Analysis, Vol. 22, No. 6 (Nov.-Dec. 1992),
pp. 40-46.
McNamee, Peter, John Celona. Decision Analysis for the Professional. 3rd ed. Menlo Park, CA:
SmartOrg, 2001.