This document provides a tutorial for using the SuperDecisions software to build multi-criteria decision models. It explains how to install the software, build a decision hierarchy, make pairwise comparisons between criteria and alternatives, view results, and perform sensitivity analysis. The tutorial uses a sample model to select the best car out of three alternatives based on criteria like price, miles per gallon, prestige, and comfort. It demonstrates how to construct the hierarchy in SuperDecisions, enter pairwise comparison judgments, and view the resulting supermatrix before and after synthesis.
1. SuperDecisions Software Tutorial
• Installing the SuperDecisions Software
• Glossary (3 slides)
• Building the decision hierarchy
• Removing unintended loops
• Making Pairwise Comparisons
• Improving Consistency of pairwise comparisons
• Entering Direct Data instead of Pairwise Comparing
• The Three Types of Supermatrices
• Getting Results
• Sanity Check
• Graphical Sensitivity
• Dynamic Sensitivity
• Restoring backups
• Exporting Supermatrices for use in Excel or Word
• Reports
2. Glossary
Alternative– an alternative is a node representing one of the choices or outcomes for a decision
model. The alternatives are grouped together in a single cluster.
Cluster – a cluster is a collection of nodes that have some logical relationship together in a frame
inside a SuperDecisions Window.
Comparison Group – consists of a parent node linked to a group of children nodes that will be
pairwise compared with respect to the parent node for importance, preference or likelihood. The
children nodes must be in a cluster together; the parent node may be in a different cluster or in
the same cluster as its children nodes, and may have children groups in several clusters.
Criterion – a criterion (always use this word when you mean just one!) is a decision factor, something
that must be considered when making a decision. A criterion is represented by a node in a
SuperDecisions model.
Goal – a goal in a model is a single node in a cluster with links only “from” it. In hierarchical models
there should be only one goal. In hierarchies criteria nodes are linked from the goal and
judgments are made about their importance with respect to the goal. In ANP subnetworks the
goal is not an explicit node but is external to the network, often being represented by the name of
the network, and is the concept kept in mind while making all judgments on nodes in the network.
Hierarchy– In a hierarchy the goal is at the top, the criteria are in a separate cluster connected from
the goal, the subcriteria are in clusters connected from a parent criterion. Clusters are often
arranged hierarchically with the goal cluster at the top of the window, the criteria cluster below
that, the subcriteria clusters below that and the alternative cluster at the bottom.
Inconsistency– if element A is preferred to element B by 2 and element B is preferred to element C by
3, then element A should be preferred to element C by their product, 6. If it is not 6, then there is
inconsistency. All such triples of judgments for a comparison group are checked for consistency
and SuperDecisions gives a measure of the inconsistency as a decimal number that should be
less than about 0.10.
3. Glossary (Cont’d)
Judgments – a dominance judgment is entered for each pair of children nodes in a comparison group
using the Fundamental Scale: 1-Equal, 3-Moderate, 5-Strong, 7- Very Strong, 9-Extremely
Strong, plus numbers in between for intermediate judgments, as well as decimals for finer
distinction. The judgment modes in SuperDecisions are:
Graphical – use ratios of bars or relative areas in a circle
Verbal– use a continuous sliding scale on which the words from the Fundamental Scale are
indicated
Matrix– enter the dominance judgments as numbers in a table or matrix for each pair
Questionnaire–use a questionnaire form in which the appropriate whole number from the
Fundamental Scale is selected to indicate dominance for each pair
Direct–this mode allows the direct entry of priorities for a group of children nodes; numbers which
do not add up to 1.0 can also be entered and they will be normalized to give the priorities
Link or Connection – a link goes from one node to another. A node most often has links to several
other nodes.
Model– a SuperDecisions model may be a simple network contained in a single window, or a complex
model of 2 or 3 or more levels consisting of a main network (window) with attached sub-networks
(windows) linked together.
Network – any collection of clusters, nodes, and their connections in a single window (a window is a
box or frame). A network may be either a hierarchy or a feedback structure.
4. Glossary (Cont’d)
Node – a node is an element or factor in a decision such as the goal, a criterion, a subcriterion, or an
alternative. Nodes are smaller rectangular frames inside a cluster frame.
Normalization– mathematical procedure of summing a group of numbers and dividing each by the
sum so that the resulting numbers will sum to 1; the numbers are then said to be normalized to 1.
Priorities are sets of numbers normalized to 1. To obtain priorities from any group of numbers
apply the procedure above.
Priority – Priorities result from making a set of pairwise comparison judgments on a group of children
nodes. The priorities sum to 1.
Sensitivity– To perform sensitivity with respect to a criterion in a hierarchy means to vary the priority
of that node, maintaining the same relative proportion of the other nodes with respect to the goal,
and see how the outcome changes. To perform sensitivity in a network is a more complicated
process because every node may be involved in many different comparison groups, but it still
means to vary the priority of a given element in the structure and see how the outcome changes.
Supermatrix – the judgment data for a model is stored in supermatrices (think of an Excel
spreadsheet).
Synthesis – after judgments are made the model is synthesized to give the best alternative; that is,
the one with the highest synthesized priority.
Window– a box or frame, sometimes with a menu of commands across the top, or sometimes not, in
which case it is called a dialogue box. The opening screen of SuperDecisions is a blank window.
5. SuperDecisions’ Opening Screen
Main Menu Commands
• File – Open file, Close file, Recent Files, Backups, Export supermatrices, Print preview for text version of model
Each model is contained in a separate file. Old files have .mod extensions, new files have .sdmod extensions
• Design – Build a network by creating clusters and nodes and making node connections
• Assess/Compare – Perform pairwise comparisons, access the Ratings spreadsheet if there is one
• Computations – Synthesize results, look at supermatrices, perform sensitivity, do sanity check for errors and
incomplete comparisons
• Network – quickly transit around the sub-networks in a complex model and go going directly into a selected subnet
• Test – Programmer menu for development work
• Help – Sample models, including some in other languages, Help (not yet implemented in 2.1.16 version of
SuperDecisions), for now use this old Help file: http://www.superdecisions.com/SuperDecisions_Help.pdf
6. Main Menu Commands
• File – New – brings up templates, Open file, Close file, Recent Files, Backups, Import
model in .txt format, Export supermatrices to .txt files, Print model report, Old files
have .mod extensions, new files have .sdmod extensions
• Design – Build a network by creating clusters and nodes and making node
connections
• Assess/Compare – Perform pairwise comparisons, access the Ratings spreadsheet if
there is one
• Computations – Synthesize results, look at supermatrices, perform sensitivity, do
sanity check for errors and incomplete comparisons
• Network – quickly transit around the sub-networks in a complex model and go going
directly into a selected subnet
• Test – Programmer menu for development work
• Help – Sample models, including some in other languages, Help, for now use this old
Help file: http://www.superdecisions.com/SuperDecisions_Help.pdf
7. Relative Decision Model
• We will demonstrate how to use the
SuperDecisions software to create a
simple hierarchical relative model for
selecting the best of three cars.
• In a relative model the alternatives are
pairwise compared against the criteria.
• In a ratings model the alternatives are
rated against standards. See Tutorial 2 for
how to build a ratings model.
8. A Three-Level Hierarchy to Choose
the Best Car
Goal
Buy Best Car
Price MPG
Prestige (Miles per Comfort
gallon)
Acura TL Toyota Camry Honda Civic
9. The Cars
• Acura TL
– Cost $30,000-$35,000
– Miles per Gallon 20/29 (City/Hwy)
– Prestige is very good
– Comfort is excellent
• Toyoto Camry
– Cost $22,000 - $28,000
– Miles per gallon 22/30 (City/Hwy)
– Prestige is good
– Comfort is good
• Honda Civic
– Cost $16,000 - $20,000
– Miles per gallon 29/38 (City/Hwy)
– Prestige is medium to low
– Comfort is medium to low
10. The Decision Hierarchy as it appears in the
SuperDecisions Software
Cluster
Node
All links are among
nodes: the cluster link
is automatically created
because some node(s)
in the Criteria cluster
are connected to some
node(s) in the
Alternatives cluster
See the SuperDecisions software model: Tutorial_1_Acura_Relative_Model.sdmod
11. Building the Decision Hierarchy
• File>New and choose Simple Network
(simply leaves you with a blank window) or just
start building on the blank opening screen
To create the Goal Cluster select the
Design command:
• Design>Cluster>New brings up
the cluster editing window. Type a name
and description for the cluster and set the
parameters: font, text size, color,
icon (if you want) and save.
12. Decision Hierarchy with a new
cluster to hold the Goal Node
Double-click on a cluster to minimize it to an icon or expand from an icon.
13. Add Goal Node to Goal Cluster
• Design>Node>New to get to the node editing window.
Select cluster to add node to, enter goal node name (and
description if you wish) selecting parameters as before, and save.
For convenience, just use the word Goal again.
To display
descriptions make
sure the
Icon on the
top menu bar is
depressed then
hold cursor over
node or cluster.
14. Add the Rest of the Clusters
and Nodes
Double-click a
cluster to minimize
or expand
To re-size a cluster
click on bottom right
hand button on a
cluster and drag.
IMPORTANT!
The cluster holding
the alternatives must
be named with some
version of the word
Alternatives
Note: Prefacing Cluster and node names with numbers (prefacing with a
allows you to control their order in the supermatrices as number is OK)
they are appear in alphabetical order there.
15. Build the Hierarchy in SuperDecisions
Step 1. Create the clusters and nodes shown below
Or use the Use the Design>Node
shortcut to Connexions from menu
connect nodes command to connect the
Goal Node to the Criteria:
To turn on the
connection
shortcut mode
left-click the
“connections”
icon.
16. Complete the Rest of the
Connections
Turn on the fan-shaped
“Show Connections” icon by
clicking it
Connect each of the
four nodes in the
Criteria cluster to all
three of the
alternative nodes.
Hold your cursor over a
node – such as “Comfort”
-, to display nodes it is
connected to outlined in
red.
When the judgments for
these nodes with respect
to Comfort have been
marked as completed,
their window will also
appear with a red outline.
17. Connecting Nodes: Menu or Shortcut
1
Click the
“Make
2
connections Left-click the “from” node
” icon to Selected nodes are depressed
depress it and outlined in black
and enter
the shortcut 3
“make
Right-click each “to”
connections
node (which outlines
” mode.
them in red)
Line then
4
automatically
appears from
Goal cluster to
Criteria cluster
To input a picture for a node or cluster icon use the Design>Node>Edit command , click, the Change Icon command and
select a picture. To add your own pictures to the available collection, copy your picture files into the C:/Program
Files/Super Decisions/Icons folder (*.gif files work pretty well).
18. Connect Criteria Nodes to
Alternative Nodes
SHORTCUT
to connect
many nodes
at once
Shift left-click on any
“from” node in a cluster
- selects all nodes
Shift right-click
on any “to” node
connects all “from”
nodes to all “to”
nodes
Line automatically
appears between
clusters
19. Remove Unintended Loops
A loop will appear on a
cluster when a node or
node(s) are connected to
other node(s) in the
same cluster.
REMOVING LOOPS
Right click on the
background of the cluster
with the loop and select
the Remove self loop
command.
20. Unweighted Supermatrix before
making any pairwise comparisons
The nodes connected from a node are shown in the column below that
node; for example, the Goal node is connected to the criteria nodes and
they are equally weighted at .25 before making pairwise comparisons.
Supermatrices are square; every node appears as a column and a row. The
priorities, that add up to 1.0, are read from the columns.
Cluster names
Node names
Priorities of
Criteria nodes
21. Making Pairwise Comparisons
• Click on the Goal Node to select it.
• Click the pairwise comparison/assessment icon on
the menu bar to enter the assessment mode or use the
Assess/Compare command.
• The Questionnaire Pairwise Comparison Mode opens by
default. Click on the Matrix tab to switch to the Matrix
Mode.
22. More About the
Compare/Assess Mode
There are 5 possible modes for entering assessments;
judgments entered in one mode will appear as the
equivalent judgment in any other mode.
Choose tab indicating whether
comparing nodes or clusters
(cluster comparisons not possible
in hierarchies)
Selector for parent node of
comparison
Selector for cluster containing
nodes linked from parent node, to
be compared with respect to it
Restore button will bring back
original judgments for
selected comparison group
when revisiting comparisons
23. Example of a Pairwise comparison Matrix in
SuperDecisions and in traditional AHP
• Only 3 judgments • In the AHP Theory
are necessary view you see 9 judgments
Acura Civic Camry
Acura 1 8 4
Civic 1/8 1 1/4
Camry 1/4 4 1
Diagonal elements are always 1, so they do
not need to be displayed. The elements below
the diagonal are always inverses of the
AHP reciprocals are shown in
judgment in the reciprocal cell above, so they
red in SuperDecisions. For
do not need to be displayed (e.g. 1/8 8).
example, 4.0 means for1/4.
aij = 1/ aji
24. Start Making Pairwise Comparisons
1. Left-click on Goal node to select it
2. Left-click on “make assessments” icon Comparison/Assessment Questionnaire mode
appears.
3. Switch to the Matrix mode by clicking on matrix tab. Change the comparative phrase
1Prestige is ????times more Preference than 2Price by left-clicking on it and choosing
importance.
.
Note the copy commands. You can then paste into Excel or Word.
25. Enter Judgments (Matrix Mode)
1. Enter judgments in cells by typing numbers from Fundamental Scale. The direction of the
arrow indicates which criterion is more important. Double-click on arrow to change dominance
direction. The first element in the comparative phrase is the dominant one. Up arrows are red,
down arrows are blue.
2. The current parent node is the Goal node and the Criteria nodes are being compared with
respect to it for importance. The inconsistency should be less than 0.10.
Inconsistency=0.07685; Derived Priorities
Mark completed
and move to next
comparison
26. Improve Consistency
(available only from Matrix Mode)
• Click on the Inconsistency button (at top left corner of matrix)
• Choose Basic Inconsistency Report; the first cell, Prestige versus MPG, currently has
a red 3 in it meaning that MPG are more important than Prestige (see previous view
of matrix), but the Best Value of 1.05 (in blue) means Prestige is a little more
important than MPG and the inconsistency would be improve down to 0.01 if that
were the judgment
• Left-click on either the Current or Best Value cell to return to the matrix and input a
new value . You can use the suggested value, or a value between it and the original
value, or leave it as it is and go to the number 2 most inconsistent judgment (MPG
versus Comfort) and change that, and so on. See next slide.
27. Matrix View after Improving
Judgment of MPG vs Prestige
Although the original inconsistency of 0.07 passes the .10 test, it can be improved.
New judgment Better Inconsistency of 0.01.
29. Other Comparison Modes
Graphical
Click and
drag on circle
(NOT the
bars) to
change
judgment
Verbal Tracking –
shows which
judgment you
are on in
equivalent
matrix view
Click button to invert
dominance
30. The Questionnaire Mode
The Assessment/Compare command opens the pairwise comparison mode by default. Choose the
judgment on the left or right side of the zero on the questionnaire line that is nearest to the more
important, more preferred, or more likely, node. Here price is more important than prestige. If
necessary, and it is here, change the verbal phrase so it reads correctly, as explained below.
In the view above the questionnaire opened with the wrong dominance word,
“moderately to strongly more Preference”. Left click on the dominance phrase itself
to get the menu of possible phrases and select the most appropriate word.
31. Direct Data Assessment Mode
Enter direct data in the Direct mode. It may be already normalized, as
shown here, or it may be numbers representing costs or distances.
Important! Click <Enter> or move
away from last judgment entered
to make sure it has registered
Click the Invert box when the priorities are inversely related to the data such
as distance, farther is lower priority, or cost, more expensive is lower priority.
32. Weaknesses of using Direct Data
• Data may not be as good as judgments in determining your personal
priorities. It would usually be better to use your judgment about what
the price of a car means to you rather than use the the data directly.
Suppose you are a poor college student. See the two results below.
Which do you think more accurately reflects the reality?
Comparing cars for price using judgments Priorities from judgments Priorities from data
33. Compare Cars for Prestige and Price
Prestige Comparisons
Price Comparisons
34. Compare Cars for MPG and Comfort
MPG Comparisons and Priorities
Comfort Comparisons and Priorities
35. The Supermatrices
1. Computations>Unweighted Supermatrix: matrix
containing the priorities from the pairwise
comparisons.
2. Computations>Weighted Supermatrix: The
unweighted supermatrix components have been
multiplied by cluster weights. In a hierarchy there
are no cluster weights and the weighted
supermatrix is the same as the unweighted.
3. Computations>Limit Supermatrix: The limit matrix is
obtained by raising the weighted supermatrix to
powers until it converges to give the answer.
37. Synthesize to get Overall Results
• Select Computations>Synthesize or click to get
the final results: the priorities of the alternatives. You
MUST name the Alternatives cluster with some
variation of the word alternatives to get an answer.
Acura TL 0.344
Toyota Camry 0.200
Honda Civic 0.456
The Raw values come from the
Limit Supermatrix. The
Normalized values are obtained
from them by summing and
dividing each by the sum. The
Ideals are obtained by dividing
the Raw values by the largest
Raw value
38. Sanity Check
• The Computations>Sanity Check will
reveal incomplete comparisons and
duplicate goals, among other things. Each
time you finish a set of comparisons you
must mark it complete before proceeding
to the next set. Unintentionally skipped
comparisons will also be caught by the
Sanity Check.
39. WARNING! No Alternatives Found
This error message will appear if there is no cluster named some
version of the word alternatives, IN ENGLISH, but for example,
3alternatives is an acceptable name. The SuperDecisions
software uses this word to find for which nodes it should deliver
synthesized priorities extracted from the raw values in the limit
supermatrix. To obtain synthesized priorities for any other nodes
go to the limit supermatrix, get the raw values, and normalize
them yourself. The Normals are the raw values divided by their
sum. The Ideals are the raw values divided by the largest raw
value.
41. Graphical Sensitivity
1. To do graphical sensitivity select
the Computations>Sensitivity
command
2. Select Edit>Independent Variable
to get to the Sensitivity input
selector box and change the
Independent Variable to the Goal.
The first graph that appears has
the first node, alphabetically,
selected as the “with respect to”
node. It is generally not the one
you want. Here you need to
select the Goal, not the Acura
TL, as the independent variable
42. Graphical Sensitivity (cont’d)
Step one. Select the
Edit>Independent Variable
command
Step two. In the Selected
Node box highlight the
current node (Acura) then
lick Edit.
Step three. In the Input Parameter
Box select Parameter Type:
Supermatrix, the Goal as Wrt
Node (“with respect to”) and
select one of the criteria as
the 1st other node, for
example, choose Prestige.
Click on the button to the
right to get the drop down box
with the other choices.
43. Getting Sensitivity Graph for Prestige
Show Selected Node Box Set Parameters Box
1 2
Sensitivity Graph for Selected node(s)
Updated Parameters – click Update for results
Priority of Prestige
is given on x-axis;
vertical line starts
at Prestige priority
3 4 of 50%; car
priorities for
Prestige=50%
priority are shown
by intersections
with vertical line.
44. Interpreting Sensitivity of Prestige
At Prestige = 50%, Acura is best At Prestige = 9.2% (actual priority in model), Civic is best
Click and drag vertical line to change priority of Prestige on horizontal axis from 0.5 to 0.1.
The analysis: If your priority is less than about 25% for Prestige, the Honda is the car
to buy. For any priority greater than that, the Acura is your best car.
45. Priorities of all Nodes in Model
• Select Computations>Priorities
command to see the priorities of
all nodes in model
• “Limiting priority” column shows
priority of Prestige compared to all
the other nodes in the entire
model.
• “Normalized by Cluster” column
shows the priority of Prestige
(.096) compared to the other
criteria in its cluster.
• Drag the vertical line from .5 to .
096 on the x-axis in Sensitivity to
show the priorities of the cars at
the model priority of .096 for
Prestige.
46. Dynamic Sensitivity
Select Computations>New Sensitivity to get into Dynamic Sensitivity Mode
Click on “Node
for Sensitivity”
selector button
at right and
change from
Goal Node,
shown here, to
1Prestige.
47. Dynamic Sensitivity - Prestige
The Prestige
parameter can be
dragged from 0.00 to
1.00. At 0.50, the
priorities of the cars
are the same as the
overall synthesized
result priorities of
0.344, 0.200 and
0.456.
48. Dynamic Sensitivity
At a parameter
value of 1.00 for
Prestige the most
prestigious car,
the Acura is
clearly the best
choice.
49. Putting Subcriteria into a Hierarchical
Model
• Create separate
clusters for the
subcriteria of the
criteria that will have
subcriteria.
• Price has subcriteria
of initial cost and
maintenance, which
then connect to the
cars.
• Comfort has
subcriteria of Ride and
Driving Performance
which then connect to
the cars.
• Prestige and MPG
have no subcriteria
and connect directly to
the cars.
50. Getting Backups
The most recent backup is at the top of the list. Double-click on the name of the version you want
to open it in a new window with a complete history of backups to its time included. Backups are
made at frequent intervals. However, it is still a good idea to do versioning occasionally yourself,
saving the file with a different version name: Vacation Rev 1.sdmod, Vacation Rev 2.sdmod, and
so on.
51. Exporting Supermatrices
The following files may be exported as .txt files using the File>Export
command:
Unweighted Supermatrix
Weighted Supermatrix
Limit Supermatrix
Cluster Matrix
Import into Excel using the File>Open command . Be sure to select All Files for
type of file with the Open command so the .txt file you exported will appear.
Click yes through the wizard to open in Excel.
52. Reports
• The Computations>Full Report command
and the File>Print command both
generate the same HTML file of reports
about the model. You may use the Print
Preview version, or save as a .html file. It
gives the names and descriptions of the
nodes and clusters and important
priorities.