Appendix A

Practical Questions Related to AHP

Modeling

1. What is the best kind of decision problems for AHP?

While AHP can be used in a wide number of decision-making problems, AHP is
traditionally used in selection, prioritization, and forecasting. AHP assumes that the
decision-makers know or will come up with, individually or collectively, implicitly
or explicitly, the criteria or objectives and alternatives associated with the decision.
AHP is also particularly useful for situations in which we have both tangible and
intangible criteria to consider in the decision.
2. How many hierarchies are needed to perform AHP Analysis?
When working with a single type of stakeholder, one hierarchy may be enough (or
4 if you perform a BOCR analysis); however, when working with different types of
stakeholders, a hierarchy for each perspective may be needed. In any case, there are
no rules about the number of hierarchies to analyze a problem.
3. How many criteria are needed for the AHP hierarchy?
Saaty’s scale intensity, as well as AHP as a whole, is based on the findings from
cognitive science that suggest that a person’s working memory capacity is in the
order of 7 ± 2; that is between 5 and 9 elements. This suggests that 5–9 criteria
should be the ideal. If you have more than that you may consider grouping some of
them into an overall criterion and creating sub-criteria for it (e.g., cost can group
sub-criteria such as acquisition cost and maintenance cost). An important step in the
process, which is not usually properly addressed, is the importance of modeling the
problem with a correct hierarchy. If the criteria are incomplete or they are not
clearly defined and different from each other, the model will not be a good fit for the
decision at hand and any decision obtained this way will be sub par.
4. How many levels should an AHP hierarchy have?
The same rationale from the previous question can be applied here. While there is
not a limit to the number of levels in a hierarchy, you may want to keep it within the
7 ± 2 limit, if possible. One way to do this is by decomposing the problem into a
set of hierarchies rather than using a big gigantic hierarchy.

© The Author(s) 2017 105

E. Mu and M. Pereyra-Rojas, Practical Decision Making,
SpringerBriefs in Operations Research, DOI 10.1007/978-3-319-33861-3
106 Appendix A: Practical Questions Related to AHP Modeling

5. Does AHP eliminate Cognitive Bias problems?

While cognitive biases may certainly affect the judgments we make when com-
paring elements in the model, the visibility and transparency of the decision-making
process allows us to detect potential biases much more easily, in particular during
the sensitivity analysis.
6. In a nutshell, what are the advantages of using AHP?
In terms of advantages, the most important ones are: (a) the ability of structuring a
problem in a way that is easily manageable, (b) making the decision criteria explicit
and the decision-making process transparent as a whole, (c) deriving priorities
through a rigorous mathematical process using ratio scales, (d) allowing measuring
and comparison of tangible and intangible elements and (e) allowing easy sharing
of the decision-making process for feedback and buy-in.
7. What are the potential limitations of using AHP?
Based on our experience in the use of AHP, the following limitations have been
found: (a) the comparison process may be long if the decision is complex (b) the
comparison judgment may be unreliable if the participants are not fully engaged in
the process (c) the decision-making transparency may be counter-productive for
managers who are interested in manipulating the results (d) group decision-making
may make difficult to handle consistency problems.
Appendix B
AHP Basic Theory

We present here, for the purpose of completeness, the basics of the AHP theory.1
While the theoretical fundamentals were presented by Saaty (2012). Brunnelli
(2015) and Ishizaka and Nemery (2013) also do a good job of presenting the AHP
theoretical fundamentals in a very accessible way. AHP methodology requires the
following steps: first, development of the hierarchy (goal, criteria, and alternatives);
second, assessing relative weights of the criteria; third, assessing the alternatives
relative priority with respect to criteria and finally, calculating the overall priorities.
These steps will be explained with a simple model (Fig. B.1).

Development of the Hierarchy

In a basic AHP hierarchy, we may consider three levels (as shown in Fig. B.1): the
goal, the criteria2 and the alternatives.

Assessing Criteria Relative Importance

In the AHP example shown in Fig. B.1, the C1–C3 criteria are used to evaluate the
alternatives. However, not all the criteria have the same importance for the
decision-makers. It could be that for one institution C3 has greater importance than
C2. In AHP, the criteria need to be compared pairwise with respect to the goal to
establish their relative importance using an intensity scale developed for this pur-
pose as shown in Fig. B.2.

1
This appendix is optional and some basic knowledge of linear algebra and vector notation is
required.
2
In more complex hierarchies, the criteria may have sub-criteria and it is also possible that
alternatives may have sub-alternatives.
© The Author(s) 2017 107
E. Mu and M. Pereyra-Rojas, Practical Decision Making,
SpringerBriefs in Operations Research, DOI 10.1007/978-3-319-33861-3
108 Appendix B: AHP Basic Theory

Fig. B.1 Basic AHP model example

Relative Intensity Importance Explanation

Both criteria are equally
1 Equal
important
One criterion is moderately more
3 Moderately
important than the other
One criterion is strongly more
5 Strong
important than the other
One criterion is very strongly
7 Very Strong
more important than the other
One criterion is extremely more
9 Extreme
important than the other
Intermediate
2,4,6,8 Compromise is needed
Values

Fig. B.2 Intensity scale for criteria pairwise comparison

Using the scale from Fig. B.2 we will ask questions such as: With respect to the
purpose of this decision, which is more important criterion “C3” or “C2”? If we
consider that C3 is moderately more important than C2 we are mathematically
stating C3/C2 = 3 (using the scale from Fig. B.3). Notice that this judgment auto-
matically implies that the comparison of C2 with C3 will yield the ratio C2/C3 = 1/3.
This constitutes the reciprocity rule that can be expressed mathematically as
Cij = 1/Cji where i and j are any element (i corresponds to the row and j refers to the
column) in the comparison matrix.
Appendix B: AHP Basic Theory 109

These judgments are recorded in a comparison matrix as shown in Fig. B.3.

Notice that the judgment diagonal, given that the importance of a criterion com-
pared with itself (Cij/Cij), will always be equal and is 1 in the comparison matrix.
Also, only the comparisons that fill in the upper part of the matrix (shaded area) are
needed. The judgments in the lower part of the comparison matrix are the recip-
rocals of the values in the upper part, as shown in Fig. B.3.
Another important consideration when completing the comparison matrix is the
extent to which it respects the transitivity rule. If the importance of C1/C2 = 1/5,
and the importance of C2/C3 = 1/3, then it is expected that C1/
C3 = (1/5)
(1/3) = 1/15. In other words, Cij = Cik
Ckj where Cij is the com-
parison of criteria i and j. However, this is not the case in Fig. B.3 where C1/C3 = 1
as indicated by the decision-maker. This means there is some inconsistency in this
matrix of judgment as will be explained next.

Checking Consistency of Judgments

Any comparison matrix that fulfills the reciprocity and transitivity rules is said to be
consistent. The reciprocity rule is relatively easy to respect, whenever you elicit the
judgment Cij you make a point of recording the judgment Cji as the reciprocal value
in the comparison. However, it is much harder to comply with the transitivity rule
because of the use of English language verbal comparisons from Fig. B.2 such as
“strongly more important than,” “very strongly more important than,” “extremely
more important than,” and so forth.
Deriving criteria weights in AHP only makes sense if the comparison matrix is
consistent or near consistent, and to assess this Saaty (2012) has proposed a con-
sistency index (CI) as follows:

CI ¼ ðkmax  NÞ=ðN  1Þ

where kmax is the matrix maximal eigenvalue. This is used to calculate the con-
sistency ratio defined as:

CR ¼ CI=RI

where RI is the random index (the average CI of 500 randomly filled matrices
which is available in published tables). CR less than 10 % means that the incon-
sistency is less than 10 % of 500 random matrices. CR values of 0.1 or below
constitute acceptable consistency.
110 Appendix B: AHP Basic Theory

Fig. B.3 Pairwise compar- C1 C2 C3 Weights

ison matrix C1 1 1/5 1 0.481
C2 5 1 1/3 0.114
C3 1 3 1 0.405

C. R. = 0.028

For the comparison matrix used in our example analysis, CR can be calculated as
being 0.028, which constitutes an acceptable consistency and means that we can
proceed to calculate the priorities (weights) for our criteria comparison matrix
shown in Fig. B.3.3

Deriving Criteria Weights

The vector of priorities (or weights) p for the criteria matrix, given that it is
consistent, is calculated by solving the equation (Ishizaka and Nemery 2013):

Cp ¼ np

where n is the matrix dimension of C, the criteria matrix, and p = (p1, p2, … pn).
Saaty (2012) demonstrated that for a consistent matrix, the priority vector is
obtained by solving the equation above. However, for an inconsistent matrix, this
equation is no longer valid. Therefore, the dimension n is replaced by the unknown
k. The calculation of k and p is constituted by solving the eigenvalue problem
Cp = kp. Any value k satisfying this equation is called an eigenvalue and p is its
associated eigenvector. Based on Perron theory, a positive matrix has a unique
positive eigenvalue called the maximum eigenvalue kmax. For perfectly consistent
matrices, kmax = n; otherwise the difference kmax − n is a measure of the incon-
sistency. Software packages4 calculate the eigenvector5 associated to the maximum
eigenvalue by elevating the comparison matrix to successive powers until the limit
matrix, where all the columns are equal, is reached. Any column constitutes the
desired eigenvector. The calculated priorities, using this eigenvalue method, for our
tentative criteria comparison matrix is shown in the rightmost column (under the
heading Weights) in Fig. B.3.

3
Given the extensive availability of commercial (e.g., Decision Lens, Expert Choice) and freely
available software (e.g., SuperDecisions, MakeItRational), we do not show the calculations here
but simply report the consistency reported by the software package.
4
In our applications, the open software SuperDecisions was used to perform the comparison matrix
calculations to obtain the eigenvector (criteria and sub-criteria weights) as well as ensuring that
C.R. was less or equal 0.1 (SuperDecisions 2014).
5
Naturally, there is the question if the eigenvalue is still valid for inconsistent matrices. Saaty
(2012) justified this using perturbation theory which says that slight variations in a consistent
matrix imply only slight variations of the eigenvector and eigenvalue (Ishizaka and Nemery 2013).
Appendix B: AHP Basic Theory 111

References

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

Ishizaka, A., & Nemery, P. (2013). Multi-criteria decision analysis: Methods and software. West
Sussex, UK: John Wiley and Sons.
Saaty, T. L. (2012). Decision making for leaders: The analytic hierarchy process for decisions in a
complex world (Third Revised Edition ed.). Pittsburgh: RWS Publications.