Chapter 2

Understanding the Analytic Hierarchy Process

Abstract In this chapter, we will explain the fundamentals of the Analytic

Hierarchy Process. The reader is referred to the original Saaty’s (2012) discussion
of AHP or to Brunnelli’s (2015) for a theoretical introduction to the method. In this
book, AHP concepts will be explained from a practical point of view, using exam-
ples for greater clarity.

In this chapter, we will explain the fundamentals of the Analytic Hierarchy Process.
The reader is referred to the original Saaty’s (2012) discussion of AHP or to
Brunnelli’s (2015) for a theoretical introduction to the method. In this book, AHP
concepts will be explained from a practical point of view, using examples for greater
clarity.
To explain this method we will use a simple example1: Our goal is to purchase a
new car. Our purchase is based on different criteria such as cost, comfort, and safety
(the reader can think of many more but we will use only three for illustration pur-
poses). We could evaluate several alternatives but let us assume that we have only
two: Car 1 and Car 2. To analyze the decision of purchasing a car using the analytic
hierarchy process we should follow the next steps:
1. Develop a model for the decision: Break down the decision into a hierarchy of
goals, criteria, and alternatives.
2. Derive priorities (weights) for the criteria: The importance of criteria is com-
pared pairwise with respect to the desired goal to derive their weights. We then
check the consistency of judgments; that is, a review of the judgments is done in
order to ensure a reasonable level of consistency in terms of proportionality and
transitivity.
3. Derive local priorities (preferences) for the alternatives: Derive priorities or the
alternatives with respect to each criterion separately (following a similar process

1
For this chapter, it is recommended that the reader follows the calculations of this example using
a spreadsheet.
Electronic Supplementary Material The online version of this chapter (https://doi.org/
10.1007/978-3-319-68369-0_2) contains supplementary material, which is available to autho-
rized users.

© Springer International Publishing AG 2018 7

E. Mu, M. Pereyra-Rojas, Practical Decision Making using Super Decisions v3,
SpringerBriefs in Operations Research, https://doi.org/10.1007/978-3-319-68369-0_2
8 2 Understanding the Analytic Hierarchy Process

as in the previous step, i.e., compare the alternatives pairwise with respect to
each criterion). Check and adjust the consistency as required.
4. Derive overall priorities (model synthesis): All alternative priorities obtained are
combined as a weighted sum—to take into account the weight of each criterion—
to establish the overall priorities of the alternatives. The alternative with the
highest overall priority constitutes the best choice.
5. Perform sensitivity analysis: A study of how changes in the weights of the crite-
ria could affect the result is done to understand the rationale behind the obtained
results.
6. Making a final decision: Based on the synthesis results and sensitivity analysis,
a decision can be made.
At this point, the reader may feel a little intimidated by terms such as judgments,
priorities, pairwise comparison, and consistency; however, the following discussion
will clarify these topics.

2.1 Developing a Model

The first step in an AHP analysis is to build a hierarchy for the decision. This is also
called decision modeling and it simply consists of building a hierarchy to analyze
the decision.
The analytic hierarchy process (AHP) structures the problem as a hierarchy.
Figure 2.1 shows the hierarchy proposed for our example. Note that the first level of
the hierarchy is our goal; in our example, buying a car. The second level in the hier-
archy is constituted by the criteria we will use to decide the purchase. In our

Fig. 2.1 Decision hierarchy for buying a car

2.2 Deriving Priorities (Weights) for the Criteria 9

Table 2.1 Saaty’s pairwise Verbal judgment Numeric value

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

Table 2.2 Pairwise comparison matrix of criteria for buying a car

Buying a car Cost Comfort Safety
Cost
Comfort
Safety

example, we have mentioned three criteria: cost, comfort, and safety. The third level
consists of the available alternatives.2 In this case: Car 1 and Car 2.
The advantages of this hierarchical decomposition are clear. By structuring the
problem in this way it is possible to better understand the decision to be achieved,
the criteria to be used, and the alternatives to be evaluated. This step is crucial and
this is where, in more complex problems, it is possible to request the participation
of experts to ensure that all criteria and possible alternatives have been considered.
Also note that in complex problems it may be necessary to include additional levels
in the hierarchy such as sub-criteria.

2.2 Deriving Priorities (Weights) for the Criteria

Not all the criteria will have the same importance. Therefore, the second step in the
AHP process is to derive the relative priorities (weights) for the criteria. It is called
relative because the obtained criteria priorities are measured with respect to each
other as we will see in the following discussion.
It is clear that when buying a car (as in other decisions), not all criteria are equally
important in a given time. For example, a student may give more importance to the
cost factor rather than to comfort and safety, while a parent may give more impor-
tance to the safety factor rather than to the others. Clearly, the importance or weight
of each criterion will be different and because of this, we first are required to derive
by pairwise comparisons the relative priority of each criterion with respect to each

2
Each criterion, alternative, and the goal are collectively referred as model elements.
10 2 Understanding the Analytic Hierarchy Process

Table 2.3 Pairwise comparison matrix with intensity judgments

Buying a car Cost Comfort Safety
Cost 1 7 3
Comfort 1/7 1 1/3
Safety 1/3 3 1

Table 2.4 Column addition

Buying a car Cost Comfort Safety
Cost 1.000 7.000 3.000
Comfort 0.143 1.000 0.333
Safety 0.333 3.000 1.000
Sum 1.476 11.000 4.333

Table 2.5 Normalized matrix

Buying a car Cost Comfort Safety
Cost 0.677 0.636 0.692
Comfort 0.097 0.091 0.077
Safety 0.226 0.273 0.231

of the others using a numerical scale for comparison developed by Saaty (2012) as
shown in Table 2.1.3
To perform the pairwise comparison you need to create a comparison matrix of
the criteria involved in the decision, as shown in Table 2.2.
Cells in comparison matrices will have a value from the numeric scale shown in
Table 2.1 to reflect our relative preference (also called intensity judgment or simply
judgment) in each of the compared pairs. For example, if we consider that cost is
very strongly more important than the comfort factor, the cost-comfort comparison
cell (i.e., the intersection of the row “cost” and column “comfort”) will contain the
value 7 as shown in Table 2.3. Mathematically this means that the ratio of the impor-
tance of cost versus the importance of comfort is seven (cost/comfort = 7). Because
of this, the opposite comparison, the importance of comfort relative to the impor-
tance of cost, will yield the reciprocal of this value (comfort/cost = 1/7) as shown in
the comfort-cost cell in the comparison matrix in Table 2.3. The rationale is intui-
tively obvious. For example, if in daily life we say that an apple A is twice as big as
apple B (A/B = 2), this implies that apple B is half the size of apple A (B/A = 1/2).
Similarly, if we consider that cost is moderately more important than safety (cost/
safety = 3), we will enter 3 in the cost-safety cell (using the scale from Table 2.1)
and the safety-cost cell will contain the reciprocal 1/3 (safety/cost = 1/3). Finally, if
we feel that safety is moderately more important than comfort, the safety-comfort

3
In this figure, the intermediate values 2, 4, 6, and 8 are used to address situations of uncertainty.
For example, when the decision-maker is in doubt whether to rate a pairwise comparison as “mod-
erately more important (3)” or “strongly more important (5),” a possible solution is to rate it as
“From moderately to strongly more important”; that is, a 4.
2.2 Deriving Priorities (Weights) for the Criteria 11

Table 2.6 Calculation of priorities: row averages

Buying a car Cost Comfort Safety Priority
Cost 0.677 0.636 0.692 0.669
Comfort 0.097 0.091 0.077 0.088
Safety 0.226 0.273 0.231 0.243

Table 2.7 Presentation of results: original judgments and priorities

Buying a car Cost Comfort Safety Priority
Cost 1.000 7.000 3.000 0.669
Comfort 0.143 1.000 0.333 0.088
Safety 0.333 3.000 1.000 0.243

Table 2.8 Consistency indices for a randomly generated matrix

n 3 4 5 6
RI 0.58 0.9 1.12 1.24

cell will contain the value 3 and the comfortsafety cell will have the reciprocal 1/3.
Once all these judgments are entered in the pairwise comparison matrix (Table 2.2),
we obtain the results shown in Table 2.3.
Note in the comparison matrix of Table 2.3 that when the importance of a crite-
rion is compared with itself; for example, cost versus cost, comfort versus comfort,
or safety versus safety; the input value is 1 which corresponds to the intensity of
equal importance in the scale of Table 2.1. This is intuitively sound because the ratio
of the importance of a given criterion with respect to the importance of itself will
always be equal.
At this stage, you can see one of the great advantages of the analytic hierarchy
process: its natural simplicity. Regardless of how many factors are involved in mak-
ing the decision, the AHP method only requires to compare a pair of elements at any
time; something that, because of our pair anatomy (e.g., two hands), we have done
for centuries. Another important advantage is that it allows the inclusion of tangible
variables (e.g., cost) as well as intangible ones (e.g., comfort) as criteria in the deci-
sion. The comparison matrix (Table 2.3) shows the pairwise relative priorities for
the criteria. We now need to calculate the overall priorities or weights of the criteria.
There are two methods available for this purpose: the exact and the approximate.
Although we will not show the exact method in detail here, the general idea is
very simple. Raise the comparison matrix to powers (e.g., raise the matrix to the
power of two, raise the resulting matrix to the power of two again, and so forth) a
few times until all the columns become identical. This is called the limit matrix. At
this point, any of the matrix columns constitutes the desired set of priorities. This
calculation can be done in a spreadsheet but it is currently done very easily using
AHP-based software packages.
12 2 Understanding the Analytic Hierarchy Process

Table 2.9 Prioritization results

Buying a car Cost Comfort Safety Priority
Cost 1.000 7.000 3.000 0.669
Comfort 0.143 1.000 0.333 0.088
Safety 0.333 3.000 1.000 0.243

Table 2.10 Priorities as factors

Buying a car Cost Comfort Safety
Criteria weights -> 0.669 0.088 0.243
Cost 1.000 7.000 3.000
Comfort 0.143 1.000 0.333
Safety 0.333 3.000 1.000

As we aim to explain roughly the elements of the AHP method, we will rather
use the approximate method in our example due to its simplicity. However, keep in
mind that this method provides a valid approximation to the overall weights only
when the comparison matrix has a very low inconsistency.4
The approximate method requires the normalization of the comparison matrix;
i.e., add the values in each column (Table 2.4).
Next, divide each cell by the total of the column (Table  2.5). The normalized
matrix is shown in Table 2.5.
From this normalized matrix, we obtain the overall or final priorities (Table 2.8)
by simply calculating the average value of each row (e.g., for the cost row: 0.677 + 
0.636 + 0.692)/3 = 0.669).
Although there is no standardized way of presenting the results, showing the
comparison matrix with the original judgments (Table 2.4) along with the calculated
priorities (obtained in Table 2.6) is a useful way to see the judgments and priorities
at the same time, as it can be seen in Table 2.7.
According to the results in Table 2.7, it is clear that—for this example—we give
more importance to the cost criterion (0.669), followed by safety (0.243). The com-
fort factor has a minimum weight (0.088) in our purchasing decision. Another
important observation is that the pairwise comparison of criteria, through questions
such as to purchase your car, what is more important: cost or comfort?, allows us to
derive, based on our preferences, the final priorities or weights for the criteria. That
is, the priorities are not assigned arbitrarily but are derived based on our judgments
and preferences. These priorities have mathematical validity, as measurement val-
ues derived from a ratio scale, and have also an intuitive interpretation. From
Table 2.7 we can interpret that cost has 66.9% of the overall importance of the cri-
teria, followed by safety with 24.3% and comfort (8.8%), respectively.

4
Inconsistency will be explained later in our discussion.
2.3 Consistency 13

Table 2.11 Calculation of weighted columns

Buying a car Cost Comfort Safety
Cost 0.669 0.617 0.729
Comfort 0.096 0.088 0.081
Safety 0.223 0.265 0.243

Table 2.12 Calculation of weighted sum

Buying a car Cost Comfort Safety Weighted sum
Cost 0.669 0.617 0.729 2.015
Comfort 0.096 0.088 0.081 0.265
Safety 0.223 0.265 0.243 0.731

2.3 Consistency

Once judgments have been entered, it is necessary to check that they are consistent.
The idea of consistency is best illustrated in the following example: If you prefer an
apple twice as much than a pear and a pear twice as much than an orange, how much
would you prefer an apple with respect to an orange? The mathematically consistent
answer is 4. Similarly, in a comparison matrix criteria, if we provide a value of 2 to
the first criterion over the second and assign a value of 3 to the second criterion with
respect to the third one, the value of preference of the first criterion with respect to
the third one should be 2 × 3 = 6. However, if the decision-maker has assigned a
value such as 4, 5, or 7, there would be a certain level of inconsistency in the matrix
of judgments. Some inconsistency is expected and allowed in AHP analysis.
Since the numeric values are derived from the subjective preferences of individu-
als, it is impossible to avoid some inconsistencies in the final matrix of judgments.
The question is how much inconsistency is acceptable. For this purpose, AHP cal-
culates a consistency ratio (CR) comparing the consistency index (CI) of the matrix
in question (the one with our judgments) versus the consistency index of a
random-like matrix (RI). A random matrix is one where the judgments have been
entered randomly and therefore it is expected to be highly inconsistent. More spe-
cifically, RI is the average CI of 500 randomly filled in matrices. Saaty (2012) pro-
vides the calculated RI value for matrices of different sizes as shown in Table 2.8.
In AHP, the consistency ratio is defined as CR where CR = CI/RI. Saaty (2012)
has shown that a consistency ratio (CR) of 0.10 or less is acceptable to continue the
AHP analysis. If the consistency ratio is greater than 0.10, it is necessary to revise
the judgments to locate the cause of the inconsistency and correct it.
Since the calculation of the consistency ratio is easily performed by computer
programs, we limit ourselves here to producing an estimate of this value as
follows:
14 2 Understanding the Analytic Hierarchy Process

Table 2.13 Calculation of λmax

Weighted sum Priority
2.015/ 0.669 = 3.014
0.265/ 0.088 = 3.002
0.731/ 0.243 = 3.005
Total 9.021
Divide total by 3 to obtain 3.007
Lambdamax =

Table 2.14 Comparison with Cost Car 1 Car 2

respect to cost
Car 1.000 7.000
1
Car 0.143 1.000
2
Sum 1.143 8.000

(a) Start with the matrix showing the judgment comparisons and derived priorities
(Table 2.7) which is reprinted for convenience in Table 2.9.
(b) Use the priorities as factors (weights) for each column as shown in Table 2.10.
(c) Multiply each value in the first column of the comparison matrix in Table 2.10
by the first criterion priority (i.e., 1.000 × 0.669 = 0.669; 0.143 × 0.669 = 0.096;
0.333 × 0.669 = 0.223) as shown in the first column of Table 2.11; multiply each
value in the second column of the second criterion priority; continue this pro-
cess for all the columns of the comparison matrix (in our example, we have
three columns). Table  2.11 shows the resulting matrix after this process has
been completed.
(d) Add the values in each row to obtain a set of values called weighted sum as
shown in Table 2.12.
(e) Divide the elements of the weighted sum vector (obtained in the previous step)
by the corresponding priority of each criterion as shown in Table 2.13. Calculate
the average of the values from the previous step; this value is called λmax.

λmax = ( 3.014 + 3.002 + 3.005 ) / 3 = 3.007.

(f) Now we need to calculate the consistency index (CI) as follows:

CI = ( λmax − n ) / ( n − 1)

where n is the number of compared elements (in our example, n = 3).

Therefore,

CI = ( λmax − n ) / ( n − 1) = ( 3.007 − 3 ) / ( 3 − 1) = 0.004

(g) Now we can calculate the consistency ratio, defined as:

2.4 Deriving Local Priorities (Preferences) for the Alternatives 15

CR CI / RI

Therefore,

CR CI / RI 0.004 / 0.58 0.006

CI is the consistency index calculated in the previous step with a value of 0.004.
RI is the consistency index of a randomly generated comparison matrix and is avail-
able to the public in tables (Table 2.8). In other words, RI is the consistency index
that would be obtained if the assigned judgment values were totally random. It is
possible to show (this is beyond the scope of this book) that the value of RI depends
on the number of items (n) that are being compared (see expected values shown in
Table 2.8). It can be seen that for n = 3, RI = 0.58. Using these values for CI and RI,
it can be calculated that

CR 0.004 / 0.58 0.006

Since this value of 0.006 for the proportion of inconsistency CR is less than 0.10,
we can assume that our judgments matrix is reasonably consistent so we may con-
tinue the process of decision-making using AHP.

2.4 Deriving Local Priorities (Preferences)

for the Alternatives

Our third step consists of deriving the relative priorities (preferences) of the alterna-
tives with respect to each criterion. In other words, what are the priorities of the
alternatives with respect to cost, comfort, and safety, respectively? Since these pri-
orities are valid only with respect to each specific criterion, they are called local
priorities to differentiate them from the overall priorities to be calculated later.
As indicated, we need to determine the priorities of the alternatives with respect
to each of the criteria. For this purpose, we do a pairwise comparison (using the
numeric scale from Table 2.1) of all the alternatives, with respect to each criterion,
included in the decision-making model. In a model with two alternatives, it is
required to make only one comparison (Alternative 1 with Alternative 2) for each
criterion; a model with three alternatives would require to make three comparisons
(Alternative 1 with Alternative 2, Alternative 2 with Alternative 3, and Alternative 1
with alternative 3) for each criterion; and so on. There will be as many alternative
comparison matrices as there are criteria.
In our example, we only have two alternatives: Car 1 and Car 2 and we have three
criteria. This means that there will be three comparison matrices corresponding to
the following three comparisons:
With respect to the cost criterion: Compare Car 1 with Car 2.
With respect to the comfort criterion: Compare Car 1 with Car 2.
16 2 Understanding the Analytic Hierarchy Process

Table 2.15 Preference with respect to cost

Cost Car 1 Car 2 Priority
Car 1 0.875 0.875 0.0875
Car 2 0.125 0.125 0.125

Table 2.16 Results with respect to cost

Cost Car 1 Car 2 Priority
Car 1 1.000 7.000 0.875
Car 2 0.143 1.000 0.125

Table 2.17 Comparison with Comfort Car 1 Car 2

respect to comfort
Car 1 1.000 0.200
Car 2 5.000 1.000
Sum 6.000 1.200

Table 2.18 Preference with respect to comfort

Comfort Car 1 Car 2 Priority
Car 1 0.167 0.167 0.167
Car 2 0.833 0.833 0.833

With respect to the safety criterion: Compare Car 1 with Car 2.

We can do these comparisons through a series of questions as shown below with
sample answers.
Comparison Question 1: With respect to the cost criterion, which alternative is
preferable: Car 1 or Car 2?
For our example, let us assume that we prefer very strongly (using the scale in
Table 2.1) the Car 1 over the Car 2. This means that in the Car 1–Car 2 cell (i.e., the
cell intersected by the row “Car 1” and the column “Car 2”) of our comparison
matrix regarding cost alternatives (Table  2.14), we assign a value of 7 (value
assigned using the scale from Table  2.1) to reflect our preference Similarly, we
assign the reciprocal reverse 1/7 in the Car 2–Car 1 cell in the table.
By normalizing the matrix and averaging the rows we obtain the priorities (or
preferences) for each of the alternatives (Table 2.15) with respect to cost.
Because these priorities apply only to the cost criterion, they are called local
priorities with respect to cost. The results are summarized for convenience as shown
in Table 2.16.
Comparison Question 2: With respect to the comfort criterion, which alternative
is preferable: Car 1 or Car 2?
Assume that Car 2 is strongly preferred over the Car 1; that is, we assign a value of 5
(using scale from Table 2.1) in the cell Car 2–Car 1 in our comparison matrix regarding
comfort alternatives and the reciprocal reverse 1/5 in the Car 1–Car 2 cell (see Table 2.17).
2.4 Deriving Local Priorities (Preferences) for the Alternatives 17

Table 2.19 Results with respect to comfort

Comfort Car 1 Car 2 Priority
Car 1 1.000 0.200 0.167
Car 2 5.000 1.000 0.833

Table 2.20 Comparison with Safety Car 1 Car 2

respect to safety
Car 1 1.000 0.111
Car 2 9.000 1.000
Sum 10.000 1.111

Table 2.21 Preferences with respect to safety

Safety Car 1 Car 2 Priority
Car 1 0.100 0.100 0.100
Car 2 0.900 0.900 0.900

Table 2.22 Results with respect to safety

Safety Car 1 Car 2 Priority
Car 1 1.000 0.111 0.100
Car 2 9.000 1.000 0.900

By normalizing the matrix and averaging the rows we obtain the local priorities
(or preferences) for each one of the alternatives (Table 2.17) with respect to com-
fort. See Table 2.18.
The results are summarized in Table 2.19.
Comparison Question 3: With respect to the safety criterion, which alternative is
preferable: Car 1 or Car 2?
For our example, let us say the Car 2 is extremely preferable to Car 1 with respect
to this criterion. These judgments are entered numerically (using scale from
Table 2.1) in the respective cells in Table 2.20.
By normalizing the matrix and averaging the rows we obtain the local priorities
(or preferences) for each one of the alternatives (Table 2.21) with regard to safety.
The results are summarized in Table 2.22.
Notice that having only two alternatives to compare with respect to each criterion
simplifies the calculations with respect to consistency. When there are only two ele-
ments to compare (in our example, Car 1 and Car 2), the respective comparison
matrices (Tables 2.14, 2.17, and 2.20) will always be consistent (CR = 0). However,
consistency must be checked if the number of elements pairwise compared is 3 or
more.
We can summarize the results of this step by indicating that if our only criterion
were cost, Car 1 would be our best option (priority = 0.875 in Table 2.16); if our
18 2 Understanding the Analytic Hierarchy Process

Table 2.23 Local priorities (or preferences) of the alternatives with respect to each criterion
Alternatives Cost Comfort Safety
Car 1 0.875 0.167 0.100
Car 2 0.125 0.833 0.900

Table 2.24 Local priorities table as a base

Cost Comfort Safety
Car 1 0.875 0.167 0.100
Car 2 0.125 0.833 0.900

Table 2.25 Preparation for weighing of priorities

Cost Comfort Safety
Criteria weights -> 0.669 0.088 0.243
Car 1 0.875 0.167 0.100
Car 2 0.125 0.833 0.900

Table 2.26 Calculation of overall priorities

Cost Comfort Safety Overall priority
Criteria weights -> 0.669 0.088 0.243
Car 1 0.585 0.015 0.024 0.624
Car 2 0.084 0.074 0.219 0.376

only criterion were comfort, our best bet would be the Car 2 (0.833 priority in
Table 2.19) and finally, if our sole purchasing criteria were safety, our best option
would be the Car 2 (0.900 priority in Table 2.22). As previously indicated, the pri-
orities (preferences) of the alternatives, with respect to each criterion, are called
local priorities (or preferences). The summary of the local priorities for each alter-
native is shown in Table 2.23.

2.5 Derive Overall Priorities (Model Synthesis)

Up to this point we have obtained local priorities which indicate the preferred alter-
native with respect to each criterion. In this fourth step, we need to calculate the
overall priority (also called final priority)5 for each alternative, that is, priorities that
take into account not only our preference of alternatives for each criterion but also
the fact that each criterion has a different weight. Given that we are using all the
values provided in the model, this step is called model synthesis.
We start the calculation of the overall priority using the local priority of each
alternative as the starting point (Table  2.23, also repeated for convenience as
Table 2.24).

5
It is customary to refer to overall (also called general or final) priorities of the alternatives when
they are calculated with respect to the whole model, that is, after the synthesis process.
2.6 Sensitivity Analysis 19

Table 2.27 Synthesis of the model

Cost Comfort Safety Overall priority
Criteria Weights -> 0.669 0.088 0.243
Car 1 0.875 0.167 0.100 0.624
Car 2 0.125 0.833 0.900 0.376

Next, we need to take into consideration the weights of each criterion (from 2.8)
and for this purpose they are inserted in the table as shown in Table 2.25.
For example, the cost criterion has a priority (or weight) of 0.669 and the Car 1
has a local priority (or preference) of 0.875 relative to cost; therefore, the weighted
priority, with respect to cost, of the Car 1 is: 0.669 × 0.875 = 0.585. A similar cal-
culation is necessary to obtain the Car 1 weighted priorities with respect to comfort
and safety criteria. The resulting matrix is shown in Table 2.26. Finally, the overall
priority of the Car 1 is obtained by adding these results along the row. This procedure
is repeated for each of the alternatives being evaluated. The overall priorities of the
alternatives are shown in the rightmost column of Table 2.26.
The calculations for each alternative are shown below and the results are pre-
sented in Table 2.27 following the convention of showing the local priorities (cells)
and the weights for each criterion (at the top of each column). This process is called
the model synthesis (see Table 2.27).
In other words

Overall Priority of the car1 : 0.875 × 0.669 + 0.167 × 0.088 + 0.100 × 0.243 = 0.624
Overall Priority of the car 2 : 0.125 × 0.669 + 0.833 × 0.088 + 0.900 × 0.243 = 0.376

Now we can list the alternatives ordered by their overall priority or preference as
follows:

Alternatives Overall priority

1. Car 1 0.624
2. Car 2 0.376

In other words, given the importance (or weight) of each buying criteria (cost,
comfort, and safety), the Car 1 is preferable (overall priority = 0.624) compared to
the Car 2 (overall priority = 0.376).

2.6 Sensitivity Analysis

The overall priorities will be heavily influenced by the weights given to the respec-
tive criteria. It is useful to perform a “what-if” analysis to see how the final results
would have changed if the weights of the criteria would have been different. This
process is called sensitivity analysis and constitutes the fifth step in our AHP meth-
odology. Sensitivity analysis allows us to understand how robust is our original
20 2 Understanding the Analytic Hierarchy Process

Table 2.28 Original scenario—synthesis of the model

Cost Comfort Safety Overall priority
Criteria 0.669 0.088 0.243
weights ->
Car 1 0.875 0.167 0.100 0.624
Car 2 0.125 0.833 0.900 0.376

Table 2.29 Scenario—all criteria have the same weight

Cost Comfort Safety Overall priority
Criteria Weights -> 0.333 0.333 0.333
Car 1 0.875 0.167 0.100 0.380
Car 2 0.125 0.833 0.900 0.619

Table 2.30 Third scenario: cost weight leading to equal preferences of the alternatives
Cost Comfort Safety Overall priority
Criteria 0.500 0.250 0.250
weights
Car 1 0.875 0.167 0.100 0.504
Car 2 0.125 0.833 0.900 0.496

decision and what are the drivers (i.e., which criteria influenced the original results).
This is an important part of the process and, in general, no final decision should be
made without performing sensitivity analysis.
Note that in our example (Table 2.27), the cost is of great importance (priority
0.669) and given that the Car 1 has a high local priority (0.875) for this single crite-
rion, undoubtedly this influences the final result favorably for the Car 1. The ques-
tions that we can ask ourselves at this stage are: What would be the best alternative
if we change the importance of the criteria? What if we give the same importance to
all the criteria? And, what if we give more importance to safety or we consider it to
be as important as the cost? and so on.
To perform a sensitivity analysis, it is necessary to make changes to the weights
of the criterion and see how they change the overall priorities of the alternatives. To
exemplify this we will analyze the following scenarios: (a) when all the criteria have
the same weight and (b) what weight is needed for the cost criterion to lead to a tie
in the overall priorities of the alternatives? (This is a logical question since we know
cost is very important in the original analysis and Car 1 scores very high in this
criterion causing Car 1 to be the best preference.) Table 2.28 is the original model
synthesis (Table 2.27) where the most preferable option is the Car 1 (0.624).
Table 2.29 shows the case where all 3 criteria have the same weight (0.333). In
this second scenario, the final choice no longer has the Car 1 but the Car 2 (0.619)
as the best option. This is because the Car 2 wins in all criteria except for cost. By
lowering the weight of cost (from 0.669 in the original scenario to 0.333 in the sec-
ond stage), its cost disadvantage is not as noticeable.
References 21

This also suggests that the alternatives switch preferences when cost weighs in the
range of 0.333–0.669. To calculate the break-even point we can try different weights
for the cost and find that when the cost weighs approximately 0.5 of the overall crite-
rion importance, the Car 1 and the Car 2 have the same preference for practical pur-
poses. That is, both alternatives are equally preferred as shown in Table 2.30.

2.7 Making a Final Decision

Once the above steps have been completed, it is now possible to make a decision.
This constitutes the last step in our AHP analysis. For this, it is necessary to com-
pare the overall priorities obtained and whether the differences are large enough to
make a clear choice. It is also necessary to analyze the results of the sensitivity
analysis (Tables 2.28, 2.29, and 2.30). From this analysis, we can express our final
recommendation as follows: If the importance of the cost is more than 50% of the
overall importance of the criteria in the decision, the best alternative is the Car 1
(Table 2.28); however, if the importance of cost is much less than 50%, the Car 2 is
the best decision (from Tables 2.29 and 2.30).

2.8 Conclusion

It is important to note that, contrary to the common belief, the system does not
determine the decision we should make, rather, the results should be interpreted as
a blueprint of preference and alternatives based on the level of importance obtained
for the different criteria taking into consideration our comparative judgments. In
other words, the AHP methodology allows us to determine which alternative is the
most consistent with our criteria and the level of importance that we give them
(Videos 2.1, 2.2, 2.3 and 2.4).
Although AHP calculations can be done using electronic spreadsheets, the
appearance of software packages such as Expert Choice (2015) in the late eighties
and Super Decisions (2015) and Decision Lens (2015) later on has made AHP math-
ematical calculations very easy to deal with. A survey of current AHP software
packages is beyond the scope of this book. However, the reader is referred to
Ishizaka and Nemery (2013) for a partial survey of AHP software packages.

References

Brunnelli, M. (2015). Introduction to the analytic hierarchy process. Cham: Springer.

Decision Lens. (2015). Decision Lens. Retrieved from: http://www.decisionlens.com
Expert Choice. (2015). Expert Choice. Retrieved from: http://www.expertchoice.com
22 2 Understanding the Analytic Hierarchy Process

Ishizaka, A., & Nemery, P. (2013). Multi-criteria decision analysis: Methods and software. West
Sussex, UK: Wiley.
Saaty, T. L. (2012). Decision making for leaders: The analytic hierarchy process for decisions in a
complex world (3rd revised ed.). Pittsburgh: RWS Publications.
Super Decisions. (2015). Super Decisions. Retrieved from: http://www.superdecisions.com

Suggested Reading6

Developing a Model https://mix.office.com/watch/17icbrnswidq0.

Deriving Priorities (weights) for the Criteria https://mix.office.com/watch/4odxenri07nm.
Deriving Local Priorities (preferences) for the Alternatives https://mix.office.com/
watch/1idaxl30c6o5o.
Deriving Overall Priorities https://mix.office.com/watch/ztkx3ea8lki8.

6
These videos are the same as those provided in the supplementary material section but the links
below are provided as alternative links and for redundancy purposes.
http://www.springer.com/978-3-319-68368-3