Chapter 4.

Multiple attribute decision making

methods and applications

Dr. Nguyen Duc Duy

This slide is adopted course materials from Ph.D. Le Ngoc Quynh Lam
1
1. Introduction

2
1.1 Multiple attribute decision making- MCDM

▪ MCDM deals with decision problems under the presence of a number of

decision criteria
▪ MCDM includes MODM and MADM:
• MODM (Multiple Objective DM): Finding out the set of non-dominated
solutions from infinite solutions, according to multiple conflicting objectives.
• MADM (Multiple Attribute DM): selection of the best solution from a set of
alternatives, each of which is evaluated against multiple conflicting criteria.
▪ Conflicting property: by nature, people consider conflicting objectives
or conflicting criteria, if not, they can be combined.

Duc Duy Nguyen [email protected] 3

1.2 Multiple attribute decision making- MCDM

Comparison of MODM and MADM Approaches (Adapted from: Hwang and Yoon
1981; Starr and Zeleny 1977)

Duc Duy Nguyen [email protected] 4

Major attributes

sub-attributes

choices of actions

Duc Duy Nguyen [email protected] 5

1.3 Alternatives

▪ Represent the different choices of action available to the decision maker.

▪ Usually, the set of alternatives is assumed to be finite
▪ They are supposed to be screened, prioritized and eventually ranked.

Duc Duy Nguyen [email protected] 6

1.3 Multiple attributes

▪ Each MADM problem is associated with multiple attributes.

▪ Attributes are also referred to as "goals" or "decision criteria".
▪ Attributes represent the different dimensions from which the alternatives
can be viewed

Duc Duy Nguyen [email protected] 7

1.3 Multiple attributes

▪ In cases in which the number of attributes is large (e.g., more than a few
dozens), attributes may be arranged in a hierarchical manner.
• Some attributes may be major attributes.
• Each major attribute may be associated with several sub-attributes.
• Similarly, each sub-attribute may be associated with several sub-sub-
attributes and so on.

Duc Duy Nguyen [email protected] 8

1.4 Problems in doing trade-off

• Criteria have different units

• Some are quantitative, some are qualitative
• For some, smaller is better (resource use), for others, bigger is better
(quality issues)

How to picture diverse criteria together so

we can decide between alternatives?

Duc Duy Nguyen [email protected] 9

1.5 Classification of MCDM Methods

▪ Based on data:
• Deterministic
• Stochastic
• Fuzzy
▪ Based on number of decision makers:
• Single
• Group

single decision maker deterministic

MADM methods

Duc Duy Nguyen [email protected] 10

2. Weighted Sum Model (WSM)

11
2.1 Weighted Sum Model (WSM)

▪ Step 1: Define all factors or attributes

▪ Step 2: Assign weight to each factor
▪ Step 3: List all non-dominated alternatives
▪ Step 4: Evaluate each alternative based on each factor using scale of (0, 1); (1,
10); or (1, 100)
▪ Step 5: Calculate weighted sum for each alternative
▪ Step 6: Compare and select the best alternative

Duc Duy Nguyen [email protected] 12

2.2 Example

A just graduated student consider for a job

▪ Step 1: Three important factors:
• Salary
• Promotion
• Company location

Duc Duy Nguyen [email protected] 13

2.2 Example

▪ Step 2: Assign weight

• Salary 0.5
• Promotion 0.3
• Company location0.2

Duc Duy Nguyen [email protected] 14

2.2 Example

▪ Step 3: Alternatives
• Company A, B and C
▪ Step 4: Evaluate

Factor A B C
Salary 0.8 0.4 0.7
Promotion 0.3 0.9 0.4
Location 0.6 0.6 0.2

Duc Duy Nguyen [email protected] 15

2.2 Example

▪ Step 5: Calculate weight sum for each alternative

▪ Step 6: Decision making Select company A

Factor Weight A B C
Salary 0.5 0.8 0.4 0.7
Promotion 0.3 0.3 0.9 0.4
Location 0.2 0.6 0.6 0.2
Weighted sum 0.61 0.59 0.56

Duc Duy Nguyen [email protected] 16

Exercises
▪ You're deciding on a college for further studies and have four criteria: Reputation,
Campus Facilities, Faculty Quality, and Tuition fees. Assign weights to each criterion:
• Reputation: Weight = 9
• Campus Facilities: Weight = 8
• Faculty Quality: Weight = 7
• Tuition fees: Weight = 6 (Negative factor)
▪ You have four college options:
• Option A: Reputation = 9/10, Campus Facilities = 8/10, Faculty Quality = 7/10, Tuition
fees = $20,000/year
• Option B: Reputation = 8/10, Campus Facilities = 9/10, Faculty Quality = 8/10, Tuition
fees = $25,000/year
• Option C: Reputation = 7/10, Campus Facilities = 7/10, Faculty Quality = 9/10, Tuition
fees = $18,000/year
• Option D: Reputation = 9/10, Campus Facilities = 8/10, Faculty Quality = 6/10, Tuition
fees = $22,000/year
Duc Duy Nguyen [email protected] 17
3. Simple Additive Weighting (SAW)

18
 3.1 SAW (Simple Additive Weighting)

Multiplies the normalized value of the criteria for the alternatives with the importance of the criteria.
The alternative with the highest score is selected as the preferred one.

Duc Duy Nguyen [email protected] 1

3.2 Example

Assume the bank evaluation problem can be described as follows. Suppose the
criteria of evaluating banks can be represented by investment income (x1), number
of customers (x2), brand image (x3), and branch numbers (x4). Let the five banks
and the corresponding evaluation ratings be described:

Duc Duy Nguyen [email protected] 20

3.2 Example

X1: Max= 3,100 million

X1A= 2.500/3100=0.806

X4: Max= 18
X4E=………………….

Duc Duy Nguyen [email protected] 21

3.2 Example

▪ Next, the utility of alternative A can be obtained as:

▪ With the same procedure as above, the utilities of other alternatives can also be
obtained as:

Duc Duy Nguyen [email protected] 22

3. The Analytic Hierarchy Process (AHP)

23
3.1 The Analytic Hierarchy Process (AHP)

▪ Founded by Saaty in 1980.

▪ It is a popular and widely used method for MCDM.
▪ Allows the use of qualitative, as well as quantitative criteria in evaluation.
▪ Wide range of applications exists:
• Supplier ranking/selection
• Site selection
• Technology ranking
• ……

Duc Duy Nguyen [email protected] 24

3.1 AHP-General Idea

▪ Develop an hierarchy of decision criteria and define the alternative courses

of actions.

▪ AHP algorithm is basically composed of two steps:

1. Determine the relative weights of the decision criteria
2. Determine the relative rankings (priorities) of alternatives

Duc Duy Nguyen [email protected] 25

3.2 Example: Car Selection

▪ Objective
• Selecting a car
▪ Criteria
• Style, Reliability, Fuel-economy
▪ Alternatives
• Civic Coupe, Saturn Coupe, Ford Escort, Mazda Miata

Duc Duy Nguyen [email protected] 26

3.2 Example: Hierarchy tree

Select a new car

Style Reliability Fuel Economy

Civic Saturn Escort Miata

Alternative courses of action
Duc Duy Nguyen [email protected] 27
3.3 Example: Ranking of Criteria and Alternatives

▪ Pairwise comparisons are made with the grades ranging from 1-9.
▪ A basic, but very reasonable assumption for comparing alternatives:
If attribute A is absolutely more important than attribute B and is rated at 9,
then B must be absolutely less important than A and is graded as 1/9.
▪ These pairwise comparisons are carried out for all factors to be considered,
usually not more than 7, and the matrix is completed.

Duc Duy Nguyen [email protected] 28

3.3 Example: 1 - 9 Scale
Intensity of Importance Definition
1 Equal Importance
3 Moderate Importance
5 Strong Importance
7 Very Strong Importance
9 Extreme Importance
2, 4, 6, 8 For compromises between the above

Reciprocals of above In comparing elements i and j

- if i is 3 compared to j
- then j is 1/3 compared to i
Rationals Force consistency
Measured values available

Duc Duy Nguyen [email protected] 29

3.3 Example: Ranking of criteria

Style Reliability Fuel Economy

Style 1 1/2 3

Reliability 2 1 4

Fuel Economy 1/3 1/4 1

Duc Duy Nguyen [email protected] 30

3.3 Example: Ranking of priorities

▪ Consider [Ax = λmaxX] where

• A is the comparison matrix of size n×n, for n criteria, also called the priority matrix.
• x is the Eigenvector of size n×1, also called the priority vector.
• λ max is the Eigenvalue, λmax  > n.
▪ To find the ranking of priorities, namely the Eigen Vector X:
1) Normalize the column entries by dividing each entry by the sum of the column.
2) Take the overall row averages.
Row
Normalized
1 0.5 3 Column Sums
0.30 0.29 0.38 averages 0.3202
A= 2 1 4 0.60 0.57 0.50 0.5571
X=
0.33 0.25 1.0 0.10 0.14 0.13 0.1226
Priority vector
Column sums 3.33 1.75 8.00 1.00 1.00 1.00

Duc Duy Nguyen [email protected] 31

3.3 Example: AHP results

Criteria weights
▪ Style 0.3202
▪ Reliability 0.5571
▪ Fuel Economy 0.1226
Selecting a New
Car
1.00

Style Reliability Fuel Economy

0.3202 0.5571 0.1226

Duc Duy Nguyen [email protected] 32

3.3 Example: Checking for Consistency

▪ Consistency Ratio (CR): measure how consistent the judgments have

been relative to large samples of purely random judgments.
• AHP evaluations are based on the assumption that the decision maker is
rational, i.e., if A is preferred to B and B is preferred to C, then A is preferred
to C.
• If the CR is greater than 0.1 the judgments are untrustworthy because they
are too close for comfort to randomness and the exercise is valueless or
must be repeated.

Duc Duy Nguyen [email protected] 33

3.3 Example: Consistency Ratio

▪ The final step is to calculate the Consistency Ratio, CR by using the table
below, derived from Saaty’s book.
▪ The upper row is the order of the random matrix, and the lower row is the
corresponding index of consistency for random judgments (Random index - RI).

Duc Duy Nguyen [email protected] 34

3.3 Example: Calculation of Consistency Ratio
▪ The next stage is to calculate λmax so as to lead to the Consistency
Index and the Consistency Ratio.
▪ Consider [Ax = λmax x] where x is the Eigenvector.
A x Ax x

1 0.5 3 0.3202 0.9667 0.3202

0.5571 1.6881 0.5571
2 1 4 = = max
0.1226 0.3687 0.1226
0.333 0.25 1.0

λmax=average{0.9667/0.3202, 1.6881/0.5571, 0.3687/0.1226}=3.0183

CI=
◼Consistency index , CI is found by 0.0092
CR=
CI=(λmax-n)/(n-1)=(3.0183-3)/(3-1)= 0.0092 0.0158

Duc Duy Nguyen [email protected] 35

3.3 Example: Consistency Ratio

▪ An inconsistency of 10% or less implies that the adjustment is small as

compared to the actual values of the eigenvector entries.
▪ A CR as high as, say, 90% would mean that the pairwise judgments are just
about random and are completely untrustworthy! In this case, comparisons
should be repeated.
▪ In the above example:
• CR=CI/0.58=0.0092/0.58=0.0158
• 0.0158 < 0.1, so the evaluations are consistent!

Duc Duy Nguyen [email protected] 36

 Matlab code

▪ Work in team
▪ Develop codes for the
remaining steps.

Duc Duy Nguyen [email protected] 37

Results

Duc Duy Nguyen [email protected] 38

3.3 Example: Ranking alternatives
Style Priority vector
Civic Saturn Escort Miata
Civic 1 1/4 4 1/6 0.13
Saturn 4 1 4 1/4 0.24
Escort 1/4 1/4 1 1/5 0.07
Miata 6 4 5 1 0.56

Reliability Civic Saturn Escort Miata

Civic 1 2 5 1 0.38
Saturn 1/2 1 3 2 0.29
Escort 1/5 1/3 1 1/4 0.07
0.26
Miata 1 1/2 4 1

Duc Duy Nguyen [email protected] 39

3.3 Ranking alternatives
Miles/gallon Normalized

Fuel Economy Civic 34 .30

Saturn 27 .24
Escort 24 .21
Miata 28 .25
113 1.0

! Since fuel economy is a quantitative measure, fuel consumption ratios

can be used to determine the relative ranking of alternatives; however
this is not obligatory. Pairwise comparisons may still be used in some
cases.

Duc Duy Nguyen [email protected] 40

3.3 Example: Ranking alternatives
Selecting a New Car
1.00

Style Reliability Fuel Economy

0.30 0.60 0.10

Civic 0.13 Civic 0.38 Civic 0.30

Saturn 0.24 Saturn 0.29 Saturn 0.24
Escort 0.07 Escort 0.07 Escort 0.21
Miata 0.56 Miata 0.26 Miata 0.25

Duc Duy Nguyen [email protected] 41

3.3 Example: Ranking of alternatives
In summary AHP provides a

Reliability

Economy
logical framework to determine

Style

Fuel
benefits of each alternatives

Civic .13 .38 .30 .30 .30

Saturn .24 .29 .24 .27
x
.60 =
Escort .07 .07 .21 .08
Miata .56 .26 .25 .10 .35

Winner

Priority matrix Criteria Weights

The Miata is the

highest ranked car
Duc Duy Nguyen [email protected] 42
3.3 Example: Cost consideration

Duc Duy Nguyen [email protected] 43

3.3 Example: Cost consideration

Duc Duy Nguyen [email protected] 44

3.3 Example: Cost consideration

No. Brand Price (cost) Normalized Benefit B/C

$ cost
1 Miata 18,000 0.3333 0.35 1.0501
2 Civic 12,000 0.2222 0.30 1.3501
3 Saturn 15,000 0.2778 0.27 0.9719
4 Escort 9,000 0.1667 0.08 0.4799
Total 54,000 1

The winner is Civic

Duc Duy Nguyen [email protected] 45

3.4 Examle 2: Evaluation of Job Offers

Ex: Peter is offered 4 jobs from Acme Manufacturing (A), Bankers Bank (B), Creative
Consulting (C), and Dynamic Decision Making (D).
He bases his evaluation on the criteria such as location, salary, job content, and long-
term prospects.

Decide upon the relative importance of the selection criteria:

Location Salary Content Long-term

Location 1 1/5 1/3 1/2

Salary 5 1 2 4
Content 3 1/2 1 3
Long-term 2 1/2 1/3 1

Duc Duy Nguyen [email protected] 46

3.4 Examle 2 Priority Vectors:
1) Normalize the column entries by dividing each entry by the sum of
the column.
2) Take the overall row averages
Location Salary Content Long-term Average

Location 0.091 0.102 0.091 0.059 0.086

Salary 0.455 0.513 0.545 0.471 0.496
Content 0.273 0.256 0.273 0.353 0.289
Long-term 0.182 0.128 0.091 0.118 0.130
+ +
1 1 1 1 1

Consistency Ratio (CR)=………………

Duc Duy Nguyen [email protected] 47
3.4 Examle 2 : Evaluation of Job Offers

Step 2: Evaluate alternatives with each criteria

Location Scores Relative Location Scores

A B C D A B C D Avg.
A 1 1/2 1/3 5 A 0.161 0.137 0.171 0.227 0.174
B 2 1 1/2 7 B 0.322 0.275 0.257 0.312 0.293
C 3 2 1 9 C 0.484 0.549 0.514 0.409 0.489
D 0.032 0.040 0.057 0.045 0.044
D 1/5 1/7 1/9 1

Duc Duy Nguyen [email protected] 48

3.4 Examle 2: Calculation of Relative Scores

Relative
weights Relative scores
Relative Scores for Each Criteria for each for each
Location Salary Content Long-Term criteria alternative
A 0.174 0.050 0.210 0.510 0.086 0.164
B 0.293 0.444 0.038 0.012 x 0.496 = 0.256
C 0.489 0.312 0.354 0.290 0.289 0.335
D 0.044 0.194 0.398 0.188 0.130 0.238

Duc Duy Nguyen [email protected] 49

3.4 Concept of Sub-Criteria, Local Weights and global
weights.
▪ Many levels of criteria and sub-criteria exists for complex problems.

Duc Duy Nguyen [email protected] 50