MULTICRITERIA ANALYSIS

Analytic Hierarchy Process Method

INTRODUCTION

• The multiobjective optimization field has two major branches:

• The continuous branch, known as Multiobjective Programming or Vector Optimization,

which deals with problems with multiple objectives, in which the alternatives can acquire
an infinite number of values.

• The discrete branch or Discrete Multicriteria Decision (DMD), which analyzes problems in
which the set of decision alternatives is formed by a finite and generally small number of
variables.

2
INTRODUCTION

• One of the main methods developed in the Discrete Multicriteria Decision field is the
AHP-Analytic Hierarchy Process, created by Professor Thomas Saaty in 1980.
• This method allows the use of qualitative as well as quantitative criteria in the evaluation
process.
• The main idea of this method is to divide the decision problem into hierarchical levels,
thus facilitating its understanding and evaluation.
• Is a Pairwise method.
The Goal

Criteria

Subcriteria

Alternatives

3
AHP – ANALYTICAL HIERARCHICAL PROCESS

• The methodology consists of 3 phases:

1. Decomposition of the decision problem in hierarchical levels, facilitating their

understanding and evaluation.
2. Comparative evaluation of preferences made by a semantic ratio scale, with values of 1
(also preferable) to 9 (extremely preferable), being 2, 4, 6, 8 intermediate values of
preferences. Paired comparison matrix of alternatives (A with B, A with C, B with C,
etc.).
3. Summary (aggregation) of priorities.

4
EXAMPLE OF STRUCTURING CRITERIA IN AHP

Buying a good car

Economy Performance Comfort

Engine Power Consumption

• It is important that in the structuring of the criteria there is a homogeneity between the
criteria of the same level, that is, the criteria must have the same level of importance,
allowing them to be compared with each other.

5
ANALYTICAL HIERARCHICAL PROCESS

• The main steps of the AHP method are presented below, as well as its structure,
elements and fundamental concepts.

• To better understand this decision-making process, consider the following example:

Goal: Buy a car

Criteria: Style, Reliability, Consumption
Alternatives: Honda, Toyota, Ford, Mazda

6
PHASE 1 – DEVELOPED AN HIERARCHICAL STRUCTURE (PROBLEM STRUCTURE)

• The first phase of the AHP method consists in the decomposition of the
problem/decision into a hierarchy structure, composed of at least one objective, criteria
and alternatives.

Statement of the general

Buy a good car
objective of decision

Criteria associated with

Style Reliability Consumption
the decision problem

Honda Honda Honda

Toyota Toyota Toyota Available and more
Ford Ford appropriate alternatives
Ford

Mazda Mazda Mazda

7
PHASE 2 - ASSESS THE ALTERNATIVES IN EACH HIERARCHIC LEVEL

• The second step is to establish priorities between elements for each level of the
hierarchy, through a comparison matrix.

• The first point to consider is the determination of a scale of values for comparison, which
should not exceed a total of nine factors in order to maintain the consistent matrix.

• Thus, Saaty defined a Fundamental Scale.

8
PHASE 2 - ASSESS THE ALTERNATIVES IN EACH HIERARCHIC LEVEL
(PAIRWISE COMPARISONS)

Verbal scale trials preferences Value

Extremely preferable to 9
Among very strongly and extremely preferable 8
Very strongly preferable to 7
Among very strongly and strongly preferable 6
Strongly preferable 5
Between Moderately and strongly preferable 4
Moderately better 3
Between equally and moderately better 2
Also preferable 1

9
PHASE 2 - ASSESS THE ALTERNATIVES IN EACH HIERARCHIC LEVEL

• The decision-maker gives a weight 𝑤𝑖 (𝑖 = 1, 2, … , 𝑛) to the criterion 𝑖.

• Pairwise Matrix
𝑤1 𝑤1 𝑤1 If 𝑤1 = 1/2 and 𝑤𝑖 = 1/4 then the criterion 1
... ... is twice important as the criterion 𝑖
𝑤1 𝑤𝑖 𝑤𝑛
⋮ ⋱ ⋮ ⋱ ⋮ 𝑤𝑖
𝑤𝑖 𝑤𝑖 𝑤𝑖 Then:
𝐴 = 𝑎𝑖𝑗 = ... ... 𝑎𝑖𝑗 = =2
𝑤1 𝑤𝑖 𝑤𝑛 𝑤𝑗
⋮ ⋱ ⋮ ⋱ ⋮
𝑤𝑛 𝑤𝑛 𝑤𝑛
... ... 𝑎𝑖𝑗 ≥ 0 Positive
𝑤1 𝑤𝑖 𝑤𝑛
𝑎𝑖𝑗 = 1 ; 𝑎𝑗𝑗 = 1
𝑎𝑖𝑗 = 1/𝑎𝑗𝑖 Reciprocity
𝑎𝑖𝑗 = 𝑎𝑖𝑘 . 𝑎𝑘𝑗 Consistency

10
PHASE 2 - ASSESS THE ALTERNATIVES IN EACH HIERARCHIC LEVEL
(PAIRWISE COMPARISONS)
• Considering the 3 criteria of the hierarchical structure (example) the following square comparison
matrix was developed.

Second Level Criteria Comparison Matrix

Style Reliability Consumption

Style 1 1/2 3
Reliability 2 1 4
Consumption 1/3 1/4 1
The Reliability criterion is slightly important to the Style criterion
• This analysis should be made for each level of the hierarchy, i.e., the existing sub-criteria for each of
the criteria considered must also go through the same form of comparison, with the same scale of
values

11
 PHASE 2 - ASSESS THE ALTERNATIVES IN EACH HIERARCHIC LEVEL
Relative priority of each criterion

To obtain the relative priority of each criterion it is necessary to:

a) Normalize the values of the matrix of comparations (matrix A) - aims to match all
criteria to the same unit, for this each value of the matrix is divided by the total of its
respective column.

Second Level Criteria Comparison Matrix Second Level Criteria Normalized Comparison Matrix

Style Reliability Consumption Style Reliability Consumption

Style 1 1/2 3 Style 3/10 2/7 3/8

Reliability 2 1 4 Reliability 3/5 4/7 1/2

Consumption 1/3 1/4 1 Consumption 1/10 1/7 1/8

Sum 10/3 7/4 8

12
PHASE 2 - ASSESS THE ALTERNATIVES IN EACH HIERARCHIC LEVEL
Relative priority of each criterion

b) Obtain the priority vector - aims to identify the order of importance of each criterion,
for this is calculated the arithmetic mean of the values of each line of the normalized
matrix obtained in the previous item.

Second Level Criteria Standardized Normalized Matrix

Relative
Style Reliability Consumption
Priority
Style 3/10 2/7 3/8 0.3202

Reliability 3/5 4/7 1/2 0.5571

Consumption 1/10 1/7 1/8 0.1226

From the results obtained, the Criterion Reliability appears first, followed by Style and
Consumption.

13
 PHASE 2 - ASSESS THE ALTERNATIVES IN EACH HIERARCHIC LEVEL
Relative priority of each criterion
Comparison Matrix
Second Level Criteria Comparison Matrix

Style Reliability Consumption

1 1/2 3
Style 1 1/2 3 𝐀= 2 1 4
Reliability 2 1 4 1/3 1/4 1
Consumption 1/3 1/4 1

Second Level Criteria Standardized Normalized Matrix

Priority Vector
Relative
Style Reliability Consumption
Priority 0,3202
Style 3/10 2/7 3/8 0.3202 𝐰 = 0,5571
Reliability 3/5 4/7 1/2 0.5571
0,1226
Consumption 1/10 1/7 1/8 0.1226

14
PHASE 2 - ASSESS THE ALTERNATIVES IN EACH HIERARCHIC LEVEL
Relative priority of each criterion

Buy a good car

1.0

Style Reliability Consumption

0,3202 0,5571 0,1226

Honda Honda Honda

Toyota Toyota Toyota

Ford Ford Ford

Mazda Mazda Mazda

15
PHASE 2 - ASSESS THE ALTERNATIVES IN EACH HIERARCHIC LEVEL
Assess the consistency of relative priorities

• The next step is to calculate the Consistency Ratio (𝐶𝑅) to measure how consistent the
judgments were in relation to large samples of completely random judgments.
• Evaluations of the AHP method assume that the decision-taker is rational, that is, if A is
preferred to B and B is preferable to C, then A is preferred to C.
• If the 𝐶𝑅 is greater than 0.1 the judgments are not reliable because they are too close for
the comfort of randomness, in this case the results obtained do not present consistent
values.
• To calculate the Consistency Ratio (𝐶𝑅) it is necessary to first obtain the value of
𝜆max which represents the highest eigenvalue of matrix 𝐀, obtained from the following
equation:

𝐀. 𝐰 = 𝜆max . 𝐰

16
PHASE 2 - ASSESS THE ALTERNATIVES IN EACH HIERARCHIC LEVEL
Assess the consistency of relative priorities

• Considering the initial example, we have to:

𝐀. 𝐰 = 𝜆max . 𝐰

1 1/2 3 0,3202 0,3202

2 1 4 0,5571 = 𝜆max 0,5571
1/3 1/4 1 0,1226 0,1226

0,9666 0,3202
Weightvector 1,6879 = 𝜆max 0,5571
0,3686 0,1226
Consistency Vector
0,9666 1,6879 0,3686
𝜆max = 𝐀𝐯𝐞𝐫𝐚𝐠𝐞 , , = 3,0183
0,3202 0,5571 0,1226

17
PHASE 2 - ASSESS THE ALTERNATIVES IN EACH HIERARCHIC LEVEL
Assess the consistency of relative priorities

• Once calculated 𝜆max , the Consistency Index (𝐶𝐼) should be calculated to calculate the
Consistency Ratio (𝐶𝑅).
• The consistency index is determined according to the formula below, where n is the
number of criteria:

𝜆max − 𝑛
𝐶𝐼 =
𝑛−1

for the previous example we have to:

3,0183 − 3
𝐶𝐼 = = 0,0091
3−1

18
PHASE 2 - ASSESS THE ALTERNATIVES IN EACH HIERARCHIC LEVEL
Assess the consistency of relative priorities

• The Consistency Ratio (𝐶𝑅) is obtained by the formula:

𝐶𝐼
𝐶𝑅 =
𝑅𝐼

where 𝑅𝐼 is the estimation of the average 𝐶𝐼 obtained from a large enough set of
randomly generated matrices of size 𝑛.
The following table defines the 𝑅𝐼 values according to the number of criteria:

Table - Values of 𝑅𝐼𝑛

𝑛 2 3 4 5 6 7 8 9 10
𝑅𝐼𝑛 0 0.58 0.9 1.12 1.24 1.32 1.41 1.45 1.49

19
PHASE 2 - ASSESS THE ALTERNATIVES IN EACH HIERARCHIC LEVEL
Assess the consistency of relative priorities
Table - Values of 𝑅𝐼𝑛
𝑛 2 3 4 5 6 7 8 9 10
𝑅𝐼𝑛 0 0.58 0.9 1.12 1.24 1.32 1.41 1.45 1.49

• for the previous example, considering 𝑛 = 3 we will have :

𝐶𝐼 0,0091
𝐶𝑅 = = = 0,0158
𝑅𝐼 0,58

• As the 𝐶𝑅 ≤ 0.1 we can conclude that the values of the relative priorities of the example
used are consistent.

20
PHASE 2 - ASSESS THE ALTERNATIVES IN EACH HIERARCHIC LEVEL

• All procedures for the construction of the comparison matrix and for determining the
relative priority of each criterion must be done again, observing now the relative
importance of each of the alternatives that make up the hierarchical structure of the
problem in question.

Buy a good car

Style Reliability Consumption

Honda Honda Honda

Toyota Toyota Toyota

Ford Ford Ford

Mazda Mazda Mazda

21
PHASE 2 - ASSESS THE ALTERNATIVES IN EACH HIERARCHIC LEVEL

• Criterion - Style

Comparison Matrix - Style criterion

Relative Priority
Honda Toyota Ford Mazda

Honda 1/1 1/4 4/1 1/6 0,1160

Toyota 4/1 1/1 4/1 1/4 0,2470
Ford 1/4 1/4 1/1 1/5 0,0600
Mazda 6/1 4/1 5/1 1/1 0,5770

22
PHASE 2 - ASSESS THE ALTERNATIVES IN EACH HIERARCHIC LEVEL

• Criterion - Reliability

Comparison Matrix - Reliability criterion

Relative Priority
Honda Toyota Ford Mazda

Honda 1/1 2/1 5/1 1/1 0,3790

Toyota 1/2 1/1 3/1 2/1 0,2900
Ford 1/5 1/3 1/1 1/4 0,0740
Mazda 1/1 1/2 4/1 1/1 0,2570

23
PHASE 2 - ASSESS THE ALTERNATIVES IN EACH HIERARCHIC LEVEL

• Criterion – Economy

Miles/Gallon Normalized

Honda 34 0,3010
Toyota 27 0,2390
Ford 24 0,2120
Mazda 28 0,2480
113
• Since fuel economy is a quantitative measure, fuel consumption rates can be used to
determine the classification or relative priority of alternatives, but this is not mandatory.
Paired comparations can still be used in some cases.

24
 Buy a good car
1.0

Style Reliability Consumption

0,3202 0,5571 0,1226

Honda 0,1160 Honda 0,3790 Honda 0,3010

Toyota 0,2470 Toyota 0,2900 Toyota 0,2390

Ford 0,0600 Ford 0,0740 Ford 0,2120

Mazda 0,5770 Mazda 0,2570 Mazda 0,2480

25
PHASE 3: SUMMARY (AGGREGATION) OF PRIORITIES

• In this last step, we obtain the priorities composed of the alternatives, multiplying the
previous values and the relative priorities obtained at the beginning of the method, i.e.:

Composite
Style Reliability Consumption Priorities
Honda 0,1160 0,3790 0,3010 0,2852
0,3202
0,2470 0,2900 0,2390
 = 0,2699
Toyota
0,5571
Ford 0,0600 0,0740 0,2120 0,0864
0,1226
Mazda 0,5770 0,2570 0,2480 0,3583

Priority of criteria

26
PHASE 3: SUMMARY (AGGREGATION) OF PRIORITIES

• In this last step, we obtain the priorities composed of the alternatives, multiplying the
previous values and the relative priorities obtained at the beginning of the method, i.e.:

Composite
Style Reliability Consumption Priorities
Honda 0,1160 0,3790 0,3010 0,2852
0,3202
0,2470 0,2900 0,2390
 = 0,2699
Toyota
0,5571
Ford 0,0600 0,0740 0,2120 0,0864
0,1226
Mazda 0,5770 0,2570 0,2480 0,3583

Priority of criteria

27
ALTERNATIVE CHOICE

• The alternative “Mazda" appears as the most suitable to buy a good car, depending on
the criteria defined and their respective importance.

Composite
Style Reliability Consumption Priorities
Honda 0,1160 0,3790 0,3010 0,2852
0,3202
0,2470 0,2900 0,2390
 = 0,2699
Toyota
0,5571
Ford 0,0600 0,0740 0,2120 0,0864
0,1226
Mazda 0,5770 0,2570 0,2480 0,3583

Priority of criteria

28