AHP

Analytic hierarchy process

Introduction
• The analytic hierarchy process (AHP), developed by
Thomas L. Saaty is designed to solve complex
multicriteria decision problems.
• AHP requires the decision maker to provide
judgments about the relative importance of each
criterion and then specify a preference for each
decision alternative using each criterion.
• The output of AHP is a prioritized ranking of the
decision alternatives based on the overall
preferences expressed by the decision maker.

1
Structure of AHP process
• The structure of AHP is hierarchical, and there are
basic and more complex structure:
• The basic structure:
• Overall Goal (eg. Selection the best car)
• Criteria (eg. Price, Comfort, fuel consumption…)
• Alternative (eg. Mazda 5, Citroen C4 picasso, Renault
Senic, Opel Zafira…)

GOAL

CRITERIA 1 CRITERIA 2 CRITERIA k

ALTERNATIVE 1 ALTERNATIVE 2 ALTERNATIVE n

2
 • The complex structure:
• goal, criteria, alternatives
• goal, criteria, sub-criteria, alternatives
• goal, scenarios, criteria (sub-criteria), alternatives
• goal, factors, scenarios, criteria (sub-criteria),
alternatives
• goal, factors, scenarios, criteria (sub-criteria), groups
of alternatives, alternatives

Establishing priorities using pairwise

comparisons approach
• Comparison of criteria and alternatives in pairs is the
fundamental building blocks of analytic hierarchy process.
• The result of the comparison is a calculation of the ratio of
the comparison
• Pairwise comparison performs person or group that is
responsible for preparing the decision.
• To determine the priorities it will require to state how
important each criterion (or alternative) is relative to each
other criterion when the criteria are compared two at a
time (pairwise)

3
• In this model, we can also compare the subjective
qualities, like tastes and personal preferences, and
express them in numbers.
• The calculation is based on the matrices and includes
the calculation of eigen vectors (C*) and eigenvalues
(λmax)
• There is software that can help us to employ AHP
method without mathematical calculation (Expert
Choice)

Pairwise comparison matrix

• Suppose we have n alternatives and m criteria.
• First, we take only one criterion
• We evaluate or compare alternatives within that
criterion.
• The principle of comparison in pairs consist of
comparing each alternative to each, in pairs, for each
criterion

4
Saaty’s scale for pairwise comparisons

Verbal Judgement Numerical rating

Extremely more important 9 or 8
Very strongly more important 7 or 6
Strongly more important 5 or 4
Moderately more important 3 or 2
Equally important 1

• Within first criterion we consider first alternative in

order to determine its advantage over all other
alternatives.
• The results of comparation, which is based on Saaty’s
scale, can be denoted as a1j, j = 1,2, ... n.
• Factor a1k denotes the preferrece of the first
alternative over the k-th alternative.
• Then, we take the second alternative and determine
its advantage over the third and all other alternatives.
• We continue our comparation until we compare the
last two alternatives.
• We have to repeat this comparation for every criterion

5
• If, for example, within one criterion, we consider the
5th alternative strongly more important than the 2nd
alternative, this means that a52=5* (or 4*), and in the
same time this means that a25=1/5
• General rule is: aij=1/aji
• This also means that aii=1* (The comparation of the
alternative to itself is equal to one)

* - These numbers are from Saaty’s scale

• In this way we get the aij components of pairwise

comparison matrix using each criterion
• Based on this matrix we will calculate the vectors of priority
of the alternatives for every single criterion. (Note: there
will be as many vectors as there are criteria)
• This calculation is called syntesization
• The same principle is applied to the comparison of the
criteria between themselves
• From the obtained vectors a final ranking of alternatives can
be achieved

6
 Example
• The company needs to make a decision about
purchasing a new machine for manufacturing.
• For this decision, the management decided to
employ AHP method.
• The selection is based on four criteria: speed of the
machine (how fast it can produce one piece of
product), service availability, reliability (potential
failures), and price.
• After the initial selection, management will have to
choose one of these three machines which
characteristics are provided in the following table:

Alt./ Crit. Speed Reliability Service availability Price

Failures happen In the premises
2000
Mashine A after 5000 hours where the machine 10.000 €
units/hour
of operation is used
Failures happen The mashine will
2500
Mashine B after 3000 hours have to be brought 12.000 €
units/hour
of operation to a near city
Failures happen The mashine will
1500
Mashine C after 4000 hours have to be brought 8.000 €
units/hour
of operation to a distant city

7
a) Compare the criteria with sufficient consistency - forming the
pairwise comparison matrix of the criteria (use Saaty’s scale)
b) Based on information from the Table, compare the alternatives
within each criterion with sufficient consistency - forming the
pairwise comparison matrices of the alternatives for each criterion
(use Saaty’s scale)
c) Calculate the vector of priority of the criteria among themselves (k)
d) Calculate vectors of priority of alternatives for each criterion (v1, v2,
...)
e) Calculate the overall priority vector (p), and provide the comment
of the obtained results
f) Conduct a "what-if" analysis (sensitivity analysis) for the criterion
"Price", and give the appropriate explanations
g) Calculate the degree of consistency of the pairwise judgemens of
alternatives within criterion "speed" and explain the obtained
results

a) Forming the pairwise comparison

matrix of the criteria
Service
Speed Reliability Price
availability
Speed 1 2 5 1/3
Reliability 1/2 1 4 1/4
Service
1/5 1/4 1 1/5
availability
Price 3 4 5 1

8
 b) forming the pairwise comparison matrices of the
alternatives for each criterion
Comparation of the alternatives within Comparation of the alternatives within
criterion „Speed” criterion „Reliability”
Mash. A Mash. B Mash. C Mash. A Mash. B Mash. C
Mash. A 1 1/3 4 Mash. A 1 6 4
Mash. B 3 1 7 Mash. B 1/6 1 1/3
Mash. C 1/4 1/7 1 Mash. C 1/4 3 1
Comparation of the alternatives within Comparation of the alternatives within
criterion „Service availability” criterion „Price”
Mash. A Mash. B Mash. C Mash. A Mash. B Mash. C
Mash. A 1 4 8 Mash. A 1 3 1/3
Mash. B 1/4 1 4 Mash. B 1/3 1 1/6
Mash. C 1/8 1/4 1 Mash. C 3 6 1

c) The vector of priority of the criteria

among themselves (k)
• The vector of priority of the criteria among themselves (k) contains
the the weights of each criterion and it is calculated based on their
pairwise comparison
• Calculation procedure:
1. Sum the values in each column of the pairwise comparison matrix
2. Divide each element in the pairwise comparison matrix by its
column total; the resulting matrix is referred to as the normalized
pairwise comparison matrix
3. Compute the average of the elements in each row of the
normalized pairwise comparison matrix; these averages provide
the vector of priority of the criteria among themselves (k)

9
 Pairwise comparison matrix of the criteria
Service
Speed Reliability Price
availability
Speed 1 2 5 1/3
Reliability 1/2 1 4 1/4
Service
1/5 1/4 1 1/5
availability
Price 3 4 5 1
Sum 4.700 7.250 15.000 1.783

Normalized pairwise comparison matrix of the The vector of priority

of the criteria among
criteria themselves (k)

Service
Norm. matr. Speed Reliability Price
availability k

Speed 0.213 0.276 0.333 0.187 0.252

Reliability 0.106 0.138 0.267 0.140 0.163

Service
0.043 0.034 0.067 0.112 0.064
availability

Price 0.638 0.552 0.333 0.561 0.521

10
d) The vectors of priority of alternatives
for each criterion (v1, v2, ...)
• In the same manner (likewise the vector k) the
vectors of priority of alternatives for each criterion
are calculated
• We will have as many vectors as there are criteria.
• Number of elements in each vector will be as many
as there are alternatives.
• These vectors are also known as vectors of local
priority of alternatives.

The vector of priority of alternatives for

criterion „Speed”
Comparation of the alternatives within criterion „Speed”
Mash. A Mash. B Mash. C
Mash. A 1 1/3 4
Mash. B 3 1 7
Mash. C 1/4 1/7 1
Sum 4.250 1.476 12.000

Normalized pairwise comparison matrix Vector of

priority of
Mash. A Mash. B Mash. C alternatives (v1)
Mash. A 0.235 0.226 0.333 0.265
Mash. B 0.706 0.677 0.583 0.656
Mash. C 0.059 0.097 0.083 0.080

11
The vector of priority of alternatives for
criterion „Reliability”
Comparation of the alternatives within criterion „Reliability”
Mash. A Mash. B Mash. C
Mash. A 1 6 4
Mash. B 1/6 1 1/3
Mash. C 1/4 3 1
Sum 1.417 10.000 5.333

Normalized pairwise comparison matrix Vector of

priority of
Mash. A Mash. B Mash. C alternatives (v2)
Mash. A 0.706 0.600 0.750 0.685
Mash. B 0.118 0.100 0.063 0.093
Mash. C 0.176 0.300 0.188 0.221

The vector of priority of alternatives for

criterion „Service availability”
Comparation of the alternatives within criterion „Service avail”
Mash. A Mash. B Mash. C
Mash. A 1 4 8
Mash. B 1/4 1 4
Mash. C 1/8 1/4 1
Sum 1.375 5.250 13.000

Normalized pairwise comparison matrix Vector of

priority of
Mash. A Mash. B Mash. C alternatives (v3)
Mash. A 0.727 0.762 0.615 0.702
Mash. B 0.182 0.190 0.308 0.227
Mash. C 0.091 0.048 0.077 0.072

12
The vector of priority of alternatives for
criterion „Price”
Comparation of the alternatives within criterion „Price”
Mash. A Mash. B Mash. C
Mash. A 1 3 1/3
Mash. B 1/3 1 1/6
Mash. C 3 6 1
Sum 4.333 10.000 1.500

Normalized pairwise comparison matrix Vector of

priority of
Mash. A Mash. B Mash. C alternatives (v4)
Mash. A 0.231 0.300 0.222 0.251
Mash. B 0.077 0.100 0.111 0.096
Mash. C 0.692 0.600 0.667 0.653

e) The overall priority vector (p)

• Vectors of priority of alternatives for each
criterion (vectors v1, v2,…) represents the local
priority of alternatives.
• The overall priority of alternatives (vector p) is
calculated by combining vectors v1, v2,… into one
matrix and multiplying that matrix with vector of
priority of the criteria among themselves (vector
k)

13
 The overall priority vector (p)

Matrix of local priorities vector of The overall

of the alternatives priority of priority of
the criteria alternatives
V1 V2 V3 V4 k p
0.265 0.685 0.702 0.251 0.252 0.354
0.656 0.093 0.227 0.096 × 0.163 = 0.245
0.080 0.221 0.072 0.653 0.064 0.401
0.521

The comment of the obtained results

• If we compare the alternatives, for example, within the
criterion „Speed" (the vector v1), the best alternative is
machine B (0.656), then machine A (0.265), and finally the
machine C (0,080)
• However, if we compare the alternatives in the context of all
the criteria simultaneously (vector p), the best alternative is
machine C (0.401), then machine A (0.354), and then machine
B (0,245)
• This result can be significanty different if the the weight
(importance) of each criterion changes.
• The change of the result by changing the weight of the
criteria are shown in sensitivity analysis or in "what-if"
analysis

14
Sensitivity analysis ("what-if" analysis)
• In this example the input data for calculation of
overall priority was the assessment of the relative
importance of the criteria and the assessment of
local priorities of the alternatives.
• It can be assumed that these estimates can vary in
certain ranges without affecting the final ranking the
decision alternatives
• Sensitivity analysis is performed in order to see to
what extent the changes in the input data will reflect
to the final ranking of the alternatives.

f) "what-if" analysis (sensitivity analysis) for

the criterion "Price"
• Procedure:
• At the X-axis in 0 and in 1 draw two vertical lines
• On the vertical line at point 1 apply the prioritiy of
alternatives within the criterion "Price" (that would be 0.251,
0.096, 0.653 values)
• The priority (weight) of the criterion "Price" apply to the X-
axis (this is 0.521) and at this point raise another vertical line
• On this vertical line apply overall priority of the alternatives
(that would be 0.354, 0.245, 0.401)
• Connect the correspodent points and you will get the graph
that would look like graph presented on the following slide:

15
Explanation of the obtained result
• The obtained graph shows that:
1. If the weight (importance) of the criteria "price" increase, the
„attractiveness” of alternative „Machine C” will also increase, and
the „attractiveness" of the alternatives "Machine A" and „Mashine
B” will decrease, and vice versa
2. If the weight (importance) of the criteria "price" decrease, the
„attractiveness” of alternative „Machine C” will decrease, and the
„attractiveness” of the alternatives "Machine A" and „Mashine B”
will increase
3. If the weight of criteria "Price" is between 0,340 and 0,448, the rank
order of alternatives is: „Mashine A” (first place), „Machine C"
(second), and „Machine B" (third place).
4. If the weight of criteria "Price" drop below the 0,340-level, the
rank order of alternatives will be: „MachineA”, „Machine B”, and
„Machine C”.

16
Consistency
• By measuring the consistency we examine whether we were consistent
enough in the process of comparing the alternatives and criteria.
• For example, if criterion A compared to criterion B has a numerical rating
of 3, and if criterion B compared to criterion C has a numerical rating of 2,
perfect consistency of criterion A compared to criterion C would have a
numerical rating of 3 × 2 = 6.
• If the A to C numerical rating assigned by the decision maker was 4 or 5,
some inconsistency would exist among the pairwise comparison.
• However, if the A to C numerical rating was assigned 1/3 (this means that
C is better than A), this wold be the major violation of the consistency, and
all process of decision making will be meaningless
• With numerous pairwise comparisons, perfect consistency is difficult to
achieve, but it is aloud to have some degree of inconsistency

• Consistency is measured by consistency ratio (CR) that can be

calculated by dividing the computed consistency index (CI) with the
consistency index of of a randomly generated pairwise comparison
matrix (RI)
• =
• The procedure for computation the consistency index (CI):
• Step 1
o Multiply each value in the first column of the pairwise comparison
matrix by the priority of the first item;
o Multiply each value in the second column of the pairwise
comparison matrix by the priority of the second item, and continue
this process for all columns of the pairwise comparison matrix.
o Sum the values across the rows to obtain a vector of values labeled
“weighted sum.” (or c)

17
• Step 2
o Divide the elements of the weighted sum vector (c) obtained
in step 1 by the corresponding priority for each criterion. With
this we will obtain vector c*
• Step 3
o Compute the average of the values found in step 2. This
average is eigenvalue of the matrix and it is denoted λmax
o Step 4
o Compute the consistency index (CI) as follows:
o =
o where n is the number of items being compared.

• The value of RI depends on the number of items

being compared and is given as follows:

n RI
3 0,58
4 0,9
5 1,12
6 1,24
7 1,32
8 1,41

18
 • if the consistency ratio (CR) is 0.10 or less, the
consistency of the pairwise comparisons is
considered reasonable,
• If it is greater than 0.10 this indicates an
inconsistency in the pairwise judgements and the
obtained results are worthless

The degree of consistency of the pairwise

judgemens of alternatives within criterion "speed"
Comparation of the alternatives within criterion „Speed”
Mash. A Mash. B Mash. C
Mash. A 1 1/3 4
Mash. B 3 1 7
Mash. C 1/4 1/7 1
Vector of priority
of alternatives (v1) 0.265 0.656 0.080

Step 1
1 1/3 4 0.803 0.803
0.265 × 3 + 0.656 × 1 + 0.080 × 7 = 2.011 c = 2.011
1/4 1/7 1 0.240 0.240

19
• Step 2
. . .
• = 3.030 = 3.065 = 3.000
. . .

3.030
∗
• = 3.065
3.000

• Step 3
. . .
• λ = = 3.032

• Step 4
.
• = = = 0.0158

• Consistency index RI for n=3 is RI = 0,58 (from the

Table – slide 36)
,
• Consistency ratio = = = 0,0272
,
• Because the pairwise comparisons show CR< 0,1 we
can conclude that the degree of consistency is
acceptable

20