Systems Analysis & Improvement

Multi Criteria Decision modelling:

Analytical Hierarchy Process (AHP)

UCD School of Mechanical and Materials Engineering

Multi criteria
decision modelling

2
Multi criteria decision modelling

• Step 1: Identify the decision situation and

understand objectives
– Sometime the surface problem hides the real
problem.
– Define the problem as exactly as possible
– Identify financial and non-financial objectives
– Understanding objectives is critical to subsequent
steps
• Step 2: Identify alternatives
– Analysis of objectives may reveal alternative
solutions not initially obvious
– A step that involves both discovery and creativity

3
Multi criteria decision modelling

• Step 3: Decompose and model the problem

– Decomposition entails structuring the problem into
smaller and more manageable pieces.
– Consider elements of uncertainty in the different
pieces
– Modeling involves reassembling the pieces of the
problem into a simplified representation.
• Keep the relevant pieces and discard the irrelevant
ones
– Models can be mathematical or graphical.
• Mathematical models allow for formal analysis that
can provide insights not otherwise obvious.

4
Multi criteria decision modelling

• Step 4: Choose the best alternative

– Which model is “preferred”?
– Remember that decision making requires human
judgment and is not a process of “solving” a
problem for the right answer.
• Step 5: Sensitivity analysis
– “What if” you change one of more aspects of the
preferred model?

5
Multi criteria decision modelling

• Step 6: Is further analysis needed?

– Sensitivity analysis often leads to new insights that
may require repeating the previous steps.
– Decision-making cycle: think of the overall
decision-making process as iterative
– Usually necessary to cycle through the different
steps several times
– Continue iterations until you arrive at the requisite
decision model, i.e., the model in which no new
insights or intuitions are gained by another cycle of
analysis, or which contains every essential for
solving the problem
• Step 7: Implement the chosen alternative

6
 MCDA Engineering Applications
• Six Sigma project selection
– to prioritise projects which provide maximum financial benefits to the
organization

• Locating a new production facility

– Objectives relate to: distances from major suppliers, distances from
customers, government incentives, quality of prospective employee
base.

• Water resources management

– Objectives relate to: to rank hydropower schemes with respect to scale
considering social, economic and environmental issues.

• Production material supplier selection

– Objectives involve cost associated with plan implementation and quality
of outgoing lots.

• Choosing a schedule for personnel in a manufacturing factory

– Objectives relate to employee satisfaction, cost for employee wages,
customer waiting times.

• Generation planning for electric power utilities

– Objectives relate to cost and system reliability.
MCDA Applications (individuals)

• Choosing a college to attend

– Objectives relate to: cost for tuition and living expenses, quality of
the education, number of miles from home, and quality of the social
life.

• Choosing from among five different job offers

– Important considerations here could involve annual salary,
opportunity for advancement, fringe benefits, number of miles from
home, and amount of travel required by the job.

• Choosing a new car to purchase

– Objectives relate to: purchase price of the car, expected annual
maintenance costs, fuel economy, amount of prestige, safety, and
resale value at some point in the future.
 Analytical Hierarchy Process (AHP)
• AHP helps decision makers learn about their problems in
a decision problem situation in which discretely defined
alternatives are to be evaluated.

• More than one criterion (or attribute) are involved in the

decision process to reach the preferred alternatives in a
ranked fashion.

• Comparisons can be made using verbal scales

• Provides automatic checks on the consistency of

judgments

9
Overview of the AHP

10
 Overview of the AHP
1. Set up decision hierarchy
2. Make pairwise comparisons of attributes &
alternatives
3. Transform comparisons into weights & check
consistency
4. Computing the matrix of option scores.
5. Use weights to obtain scores for options
6. Ranking the options.

11
 AHP – an example
• A manager in a food processing company must choose a
new packaging machine to replace the existing one
which is wearing out and caused a quality degradation
of the packed products. The manager has a limited
budget for the purchase and has narrowed down the
possible options to three: (i) the Aztec, (ii) the Barton
and (iii) the Congress.

• Difficult decision because of the variety of attributes

associated with the machines
– such as the purchase price, reputation for reliability
and the quality of after-sales support provided by 12
the different manufacturers.
 A hierarchy for packaging machine
problem

• The process of breaking down attributes continues until all the essential
criteria for making the decision have been specified.
13
 Make pairwise comparisons

• This is used to determine the relative importance of

attributes and to compare how well the options perform
on the different attributes.

Scales of pairwise comparisons:

equally important (1)
weakly more important (3)
strongly more important (5)
very strongly more important (7)
extremely more important (9)

• For example, how much more important is the initial

purchase price than the cost of upgrading the machine
at a later date? 14
 Comparing the importance of costs and
quality

Costs Quality
Costs 1 5
Quality 1

• Pairwise comparisons using verbal responses.

• Converted response to the number shown in brackets.
– For example, if ‘Costs’ are ‘strongly more important’ than ‘Quality’
then they are assumed to be five times more important.
• Intermediate responses are allowed if the decision maker
prefers these.
– between weakly and strongly more important, which would be
converted to a ‘4’.
15
 Comparing the importance of the
quality attributes
Reliability After- Speed of Customization
Sales Delivery
Support
Reliability 1 4 5 5
After-Sales 1 3 1/2
Support
Speed of Delivery 1 1/3
Customization 1

• Fractional values indicate that the ‘column’ attribute is

most important.
– For example, ‘Speed of Delivery’ is only 1/3 as important as
‘Customization’.
16
 Comparing the machines for purchase
cost
• Same process is used to compare the manager’s relative
preferences for the machines with respect to each of the
lower-level attributes.
– For example, he will be asked to consider the purchase costs of the
machines and asked whether, in terms of purchase costs, the Aztec
and Barton are ‘equally preferred’. If he indicates that the Barton is
preferred he will then be asked whether it is ‘weakly preferred’,
‘strongly preferred’ or ‘extremely strongly preferred’ (with
intermediate responses allowed).

Aztec Barton Congres

s
Aztec 1 1/3 2
Barton 1 6
Congress 1
17
 Computational notations of AHP

• Pairwise comparison matrix

• Computing the vector of criteria weights

• Computing the matrix of option scores

• Checking the consistency

18
 Pairwise comparison matrix
• Pairwise comparisons to overcome our cognitive limits and
the impossibility of effectively comparing several
alternatives at the same time.

• Decomposing the original problem into many smaller sub-

problems
– allows the decision maker to consider two alternatives at a
time.

• A pairwise comparison matrix, A = (ajk)m×m; with ajk > 0

representing the importance of the jth criterion relative to the kth
criterion. If ajk > 1, then the jth criterion is more important than the kth
criterion, while if ajk < 1, then the jth criterion is less important than the
kth criterion

• ajk . akj = 1.

19
Pairwise comparison matrix - transformation

Costs Quality
Costs 1 5
Quality 1/5 1

Reliability After- Speed of Customization

Sales Delivery
Support
Reliability 1 4 5 4

After-Sales Support 1/4 1 3 1/2

Speed of Delivery 1/5 1/3 1 1/3

Customization 1/4 2 3 1

Aztec Barton Congress

Aztec 1 1/3 2
Barton 3 1 6
Congress 1/2 1/6 1 20
Computing the vector of criteria
weights

21
 Computing the vector of criteria
weights
• Sum the columns of the table and then divide each
number in the table by the total of its column.
– For example: the total of reliability column is 1.7. this
means that the four values in the reliability column
become 0.588, 0.147, 0.118 and 0.147

Reliability After- Speed of Customisation Average of

Sales Delivery row
Support
Reliability 0.588 0.545 0.417 0.686 0.559

After-Sales 0.147 0.136 0.250 0.086 0.155

Support
Speed of 0.118 0.045 0.083 0.057 0.076
Delivery
Customization 0.147 0.273 0.250 0.171 0.210

22
Computing the matrix of option scores

23
Computing the matrix of option scores

Aztec Barton Congress Average

of row
Aztec 0.222 0.222 0.222 0.222
Barton 0.667 0.667 0.667 0.667
Congress 0.111 0.111 0.111 0.111

Costs Quality Average

of row

Costs 0.833 0.833 0.833

Quality 0.167 0.167 0.167

24
Weights for the packaging machine
problem

25
 Ranking the options
• Once the weight vector w and the score matrix S have been
computed, the AHP obtains a vector v of global scores by
multiplying S and w, i.e.

v=S·w

• The ith entry vi of v represents the global score assigned by

the AHP to the ith option. As the final step, the option ranking
is accomplished by ordering the global scores in decreasing
order.

26
 Scores for the three machines
• All the paths that lead from the top of the hierarchy to the Aztec
option are identified. All the weights in each path are then
multiplied together and the results for the different paths
summed, as shown below:
– Aztec scores less well on the more important attributes so its overall
score is relatively low.
• Score for Aztec = 0.833 × 0.875 × 0.222
+ 0.833 × 0.125 × 0.558
+ 0.167 × 0.569 × 0.167
+ 0.167 × 0.148 × 0.286
+ 0.167 × 0.074 × 0.625
+ 0.167 × 0.209 × 0.127 = 0.255

Aztec 0.255
Barton 0.541
Congress 0.204

27
 Checking the consistency
• When many pairwise comparisons are performed, some
inconsistencies may typically arise.
– Assume that 3 criteria are considered, and the decision maker
evaluates that the first criterion is slightly more important than the
second criterion, while the second criterion is slightly more important
than the third criterion. An evident inconsistency arises if the
decision maker evaluates by mistake that the third criterion is
equally or more important than the first criterion.

• Consistency index (CI): obtained by first computing the scalar

x (λmax) as the average of the elements of the vector whose jth
element is the ratio of the jth element of the summed vector
A·w to the corresponding element of the vector w.

• A perfectly consistent decision maker should always obtain CI=0,

but small values of inconsistency may be tolerated. In particular,
if

28
 Checking the consistency
• RI is the Random Index, i.e. the consistency index
when the entries of A are completely random. The
values of RI for small problems (m ≤ 10) are shown
below:

Number of Rows in
2 3 4 5 6 7 8 9 10
A
RI 0 .58 .90 1.12 1.24 1.32 1.41 1.45 1.49

• Saaty notes that the maximum value of CR that should be

allowed is .10, or 10%.

• Note that if the DM is perfectly consistent, then λmax be equal to

m, resulting in a CR value of 0.
29
 Calculating the CI
Reliability After- Speed of Customization
Sales Delivery
Support
Weights 0.559 0.155 0.076 0.210

Reliability 1 4 5 4

After-Sales Support 1/4 1 3 1/2

Speed of Delivery 1/5 1/3 1 1/3

Customization 1/4 2 3 1

• Multiply the weights to the columns. Then, sum each row

of the resulting table:
Reliability After- Speed of Customization Sums
Sales Delivery
Support

Reliability 0.559 0.620 0.380 0.840 2.399

After-Sales Support 0.140 0.155 0.228 0.105 0.628
Speed of Delivery 0.112 0.052 0.076 0.070 0.309
Customization 0.140 0.310 0.228 0.210 0.88830
 Calculating the CI
Reliability After- Speed of Customization
Sales Delivery
Support
Weights 0.559 0.155 0.076 0.210

Reliability 1 4 5 4

After-Sales Support 1/4 1 3 1/2

Speed of Delivery 1/5 1/3 1 1/3

Customization 1/4 2 3 1

• Multiply the weights to the columns. Then, sum each row

of the resulting table:

Sums Weights Ratio

Reliability 2.399 0.559 4.291
After-Sales Support 0.628 0.155 4.056
Speed of Delivery 0.309 0.076 4.078
Customization 0.888 0.210 4.221
Average 4.161 (λmax) 31
ratio
 Calculating the CI

• Consistency Index:

• CI = 4.161 – 4 / 4 – 1 = 0.054

• Consistency ratio: 0.054/0.9 = 0.06 < 0.1

32
 Strengths of the AHP
• Some users may be more comfortable
expressing preferences in words rather than
numbers
• Simplicity of pairwise comparisons
– DMs can focus on every small parts of the problem.
Only to attributes have to be considered at any one
time.
• Redundancy allows consistency to be checked
– Any inconsistencies in the judgements are checked
automatically in AHP.
• Versatility
– Wide range of applications of the AHP is evidence of
its versatility.
33
 Criticisms of the ‘standard’ AHP

• Conversion from verbal to numeric scale

– If you indicate that A is weakly more important than B the
AHP will assume that you consider A to be three times
more important, but this may not be the case.
• Problems of 1 to 9 scale
– if A is considered to be four times more important than B,
and B is four times more important than C, then to be
consistent A should be judged to be 16 times more
important than C, but this is not possible.
• No. of comparisons required may be large
– Depends on the number of alternatives and attributes.

34