The Analytic Hierarchy

Process
ARE 511 Construction
Maintenance Modeling
The analytic hierarchy process (AHP), developed by Thomas L. Saaty is
designed to solve complex problems involving multiple criteria.

The process requires the decision maker to provide judgments about the
relative importance of each of the criteria and then to specify a preference
for each decision alternative on each criterion.

The output of the AHP is a prioritized ranking indicating the overall
preference for each of the decision alternatives.

In order to introduce the AHP, we consider the problem faced by Dave
Payne. Dave is planning to purchase a new car. Alter a preliminary analysis
of the makes and models available, Dave has narrowed the list of decision
alternatives to three cars, which we will refer to as car A, car B, and car C. A
summary of the information Dave has collected about the cars has been
provided.
Car A Car B Car C
Price $13,100 $11,200 $9,500
MPG 18 23 29
Interior Deluxe Above average Standard
Body 4-door midsize 2-door sport 2-door compact
Radio AM/FM, tape AM/FM AM
Engine 6-cylinder 4-cylinder turbo 4-cylinder
Based on the information in the table - as well as his own personal feelings
resulting from driving each carDave decided that there were several
criteria that he needed to consider in making the purchase decision.
After some thought, he selected purchase price , miles per gallon ( MPG ),
comfort, and style as the four criteria to be considered. Quantitative data
regarding the purchase price and MPG criteria are provided directly in the
table.
However, measures of comfort and style cannot be specified so easily.
Dave will need to consider factors such as car interior, type of radio, ease
of entry and exit, seat-adjustment features etc., in order to determine the
comfort level for each car.
The style criterion will need to be measured in terms of Daves subjective
evaluation of each car.
Even when we deal with a criterion as easily measured as purchase price,
however, subjectivity becomes an issue whenever a particular decision
maker indicates his or her personal preferences. For instance, car A costs
$3600 more than car C; this difference might represent a great deal of
money to one person, but not very much money to another person. Thus,
whether car A is considered extremely more expensive than car C or only
moderately more expensive than car C is a subjective judgment that will
depend primarily on the financial status of the person making the
comparison. AHPs advantage is that it can handle situations in which the
subjective judgments of individuals constitute an important part of the
decision process
Developing the Hierarchy

The first step in the AHP is to develop a graphical representation of the
problem in terms of the overall goal, the criteria, and the decision
alternatives. Such a graph depicts the hierarchy for the problem.
The following figure shows the hierarchy for the car-selection problem.
Note that the first level of the hierarchy shows that the overall goal is to
select the best car. At the second level, we see that the four criteria
(purchase price, MPG, comfort, and style) will contribute to the
achievement of the overall goal. Finally, at the third level we see that each
decision alternative (car A, car B, and ear C) can contribute to each criterion
in a unique way.
The approach AHP takes is to have the decision maker specify his or her
judgments about the relative importance of each criterion in terms of its
contribution to the achievement of the overall goal.
At the next level, the AHP asks the decision maker to indicate a preference
or priority for each decision alternative in terms of how it contributes to
each criterion. For example, in the car-selection problem, Dave will need to
specify his judgment about the relative importance of each of the four
criteria. He will also need to indicate his preference for each of the three
cars relative to each criterion. Given information on relative importance
and preferences, a mathematical process is used to synthesize the
information and provide priority ranking of the three cars in terms of their
overall preference.
ESTABLISHING PRIORITIES USING THE AHP

In this section we will show how the AHP utilizes pair wise comparisons
to establish priority measures for both the criteria and the decision
alternatives.
The sets of priorities that need to be determined in the car-selection
problem are as follows:
1. The priorities of the four criteria in terms of the overall goal
2. The priorities of the three cars in terms of the purchase-price
criterion
3. The priorities of the three cars iii terms of the MPG criterion
4. The priorities of the three cars in terms of the comfort criterion
5. The priorities of the three cars in terms of the style criterion
In the following discussion we will demonstrate how to establish
priorities for the three cars in terms of the comfort criterion. The other
sets of priorities can be determined in a similar fashion.
Pair-wise Comparisons

Pair-wise comparisons are fundamental building blocks of the AHP.

In establishing the priorities for the three cars in terms of comfort, we will
ask Dave to state a preference for the comfort of the cars when the cars are
considered two at a time (pair wise). That is Dave will be asked to compare
the comfort of car A to car B, car A to car C, and car B to car C in three
separate comparisons.

The AHP employs an underlying scale with values from 1 to 9 to rate the
relative preferences for two items.

Researchers and experience have confirmed the 9-unit scale as a
reasonable basis for discriminating between the preferences for two items.
Verbal Judgment of Preference Numerical Rating
Extremely preferred 9
Very strongly to extremely 8
Very strongly preferred 7
Strongly to very strongly 6
Strongly preferred 5
Moderately to strongly 4
Moderately preferred 3
Equally to moderately 2
Equally preferred 1
In the car-selection example, suppose that Dave has compared the
comforts of car A with those of car B and is convinced that car A is more
comfortable.
Dave is then asked to state his preference for the comfort of car A
compared to that of car B using one of the verbal descriptions shown in
earlier table.
If he believes that car A is moderately preferred to car B, a value of 3 is
utilized in the AHP; if he believes that car A is strongly preferred, a value
of 5 is utilized; if he believes that car A is very strongly preferred, a value
of 7 is utilized; if he believes that car A is extremely preferred, a value of 9
is utilized. Values of 2, 4, 6, and 8 are the intermediate values for the scale.
A value of 1 is reserved for the case where the two items are judged to be
equally preferred.
Suppose that when asked his preference between cars A and B with
respect to the comfort criterion, Dave states that car A is between equally
and moderately more preferred than car B; the numerical measure that
reflects this judgment is 2.
Dave is then asked to provide his preference between car A and car C.
Suppose in this case he states that car A is very strongly to extremely more
preferred than car C; this corresponds to a numerical rating of 8.
Finally, Dave is asked to state his preference for car B compared to car C.
Suppose in this case he indicates that car B is strongly to very strongly
preferred to car C; the AHP would assign a numerical rating of 6
The Pairwise Comparison Matrix

In order to develop the priorities for the three cars in terms of the
comfort criterion, we need to develop a matrix of the pairwise comparison
ratings.

Since three cars are being considered, the pairwise comparison matrix will
consist of three rows and three columns. As shown below:
Comfort Car A Car B Car C
Car A 2 8
Car B 6
Car C
Note: In the pairwise comparison matrix, the value in row i and column j is
the measure of preference of the car in row i when compared to the car in
column j.
We see that the value in the matrix that corresponds to comparing car A
with car B is 2, the value that corresponds to comparing car A with cur C
is 8 and the value that corresponds to comparing car B with car C is 6.

In order to determine the remaining entries in the pairwise comparison
matrix, first note that when we compare any car against itself , the
judgment must be that they are equally preferred. Hence, the AHP assigns a
1 to all elements on the diagonal of the pairwise comparison matrix.

Given these entries, all that remains is to determine the rating for car B
compared to car A, car C compared to car A, and car C compared to car B.

Obviously, we could follow the same procedure and ask Dave to provide
his preferences for these pairwise comparisons. However, since we already
know that Dave has rated his preference for car A compared to car B as 2,
there is no need for him to make another pairwise comparison with these
two cars.
In fact, we will conclude that the preference rating for car B when
compared to car A is simply the reciprocal of the preference rating for car
A when compared to car B: 1/2.
Using this logic, the AHP obtains the preference rating of car B compared
to car A by computing the reciprocal of the rating of car A compared to
car B.
Using this inverse, or reciprocal, relationship, we find that the rating of car
C compared to car A is and the rating of car C compared to car B is .
Using these numerical values of preference, the complete pairwise
comparison matrix for the comfort criterion is shown in completed table.
Comfort Car A Car B Car C
Car A 1 2 8
Car B 1/2 1 6
Car C 1/8 1/6 1
Synthesis

Once the matrix of pairwise comparisons has been developed, we can
calculate what is called the priority of each of the elements being
compared. For example, we would now like to use the pairwise comparison
information to estimate the relative priority for each of the cars in terms of
the comfort criterion. This part of the AHP is referred to as synthesization.

The exact mathematical procedure required to perform this synthesization
involves the computation of eigenvalues and eigenvectors and is beyond
the scope of this text. However, the following three-step procedure
provides a good approximation of the synthesized priorities.
Procedure for Synthesizing Judgments

Step 1: Sum the values in each column of the pairwise comparison matrix.

Step 2: Divide each element in the pairwise comparison matrix by its
column total; the resulting matrix is referred to as the normalized pairwise
column.

Step 3: Compute the average of the elements in each row of the
normalized matrix; these averages provide an estimate of the relative
priorities of the elements being compared.

To see how the synthesization process works for our example problem, we
carry out the procedure using the pairwise comparison matrix shown in
table.
Step 1: Sum the values in each column.
Comfort Car A Car B Car C
Car A 1 2 8
Car B 1/2 1 6
Car C 1/8 1/6 1
Column totals 13/8 19/6 15
Step 2: Divide each element of the matrix by its column total.
Comfort Car A Car B Car C
Car A 8/13 12/19 8/15
Car B 4/13 6/19 6/15
Car C 1/13 1/19 1/15
Note that all columns in the normalized pairwise comparison matrix now
have a sum of 1.
Step 3: Average the elements in each row. (The values in the normalized
pairwise comparison matrix have been converted to decimal form.)
Comfort Car A Car B Car B Car C
Car A 0.615 0.632 0.533 0.593
Car B 0.508 0.316 0.400 0.341
Car C 0.677 0.053 0.067 0.066
Total 1.000
This synthesis provides the relative priorities for the three cars with respect
to the comfort criterion. Thus, we see that ,considering comfort, the must
preferred car is car A (with a priority of 0.593). Car B (with a priority of
0.341) is second, followed by car C (with a priority of 0.066).
Consistency
A key step in the AHP is the establishment of priorities through the use of
the pairwise comparison procedure.
An important consideration in terms of the quality of the ultimate decision
relates to the consistency of judgments that the decision maker
demonstrated during the series of pairwise comparisons.
For example, consider a situation involving the comparison of three job
offers with respect to the salary criterion. Suppose that the following
pairwise comparison matrix was developed.
Salary Job 1 Job 2 Job 3
Job 1 1 2 8
Job 2 1/2 1 3
Job 3 1/8 1/3 1
The interpretation of the preference scores is that the preference for job 1
is twice the preference for job 2, and the preference for job 2 is three limes
the preference for job 3.
Using these two pieces of information ,we would logically concIude that
the preference for job 1 should be 2 x 3 = 6 times the preference for job 3.
The fact that the pairwise comparison matrix showed a preference of
instead of 6 indicates that some lack of consistency exists in the pairwise
comparisons.
However, it has to be realized that perfect consistency is very difficult to
achieve and that some lack of consistency is expected to exist in almost any
set of pairwise comparisons.
To handle the consistency question, the AHP provides a method for
measuring the degree of consistency among the pairwise judgments
provided by the decision maker, If the degree of consistency is acceptable,
the decision process can continue. However, if the degree of consistency is
unacceptable, the decision maker should reconsider and possibly revise the
pairwise comparison judgments before proceeding with the analysis.
The AHP provides a measure of the consistency of pairwise comparison
judgments by computing a consistency ratio.

This ratio is designed in such a way that values of the ratio exceeding 0.10
are indicative of inconsistent judgments; in such cases the decision maker
would probably want to reconsider and revise the original values in the
pairwise comparison matrix.

Values of the consistently ratio of 0.10 or less are considered to indicate a
reasonable level of consistency in the pairwise comparisons.

Although the exact mathematical computation of the consistency ratio is
beyond the scope of this text, an approximation of the ratio can be
obtained. We will illustrate this computational procedure for the car-
selection problem by considering Daves pairwise comparison for the
comfort criterion.
Estimating the Consistency Ratio

Step 1: Multiply each value in the first column of the pairwise comparison
matrix by the relative priority of the first item considered; multiply each
value in the second column of the matrix by the relative priority of the
second item considered; multiply each value in the third column of the
matrix by the relative priority of the third item considered. Sum the values
across the rows to obtain a vector of values labeled weighted sum. This
computation for the car-selection example is:
(
(
(

=
(
(
(

+
(
(
(

+
(
(
(

=
(
(
(

+
(
(
(

+
(
(
(

197 . 0
034 . 1
803 . 1
066 . 0
396 . 0
528 . 0
057 . 0
341 . 0
682 . 0
074 . 0
297 . 0
593 . 0
1
6
8
066 . 0
6 / 1
1
2
341 . 0
8 / 1
2 / 1
1
593 . 0
Weighted Sum Vector
Step 2: Divide the elements of the vector of weighted sums obtained in 1
by the corresponding priority value. For the car-selection example, we
obtain:
040 . 3
593 . 0
803 . 1
=
032 . 3
341 . 0
034 . 1
=
985 . 2
066 . 0
197 . 0
=
Step 3: Compute the average of the values computed in step 2; this average
is denoted by
max
. For the car-selection example, we obtain
019 . 3
3
985 . 2 032 . 3 040 . 3
max
=
+ +
=
1
max

=
n
n
CI

Step 4: Compute the consistency index (CI), which is defined us follows:
Where n= the number of
items being compared.
For the car-selection example with n = 3, we obtain
010 . 0
2
3 019 . 3
=

= CI
Step 5: Compute the consistency ratio (CR), which is defined as follows
RI
CI
CR =
n 3 4 5 6 7 8
RI 0.58 0.9 1.12 1.24 1.32 1.41
where RI, the random index, is the consistency index of a randomly
generated pairwise comparison matrix, It can be shown that RI depends on
the number of elements being compared and takes on the following values:
Thus, for our car- example with n = 3 and RI = 0.5 we obtain the following
consistency ratio
017 . 0
58 . 0
01 . 0
= = CR
Other Pairwise Comparisons for the Car-Selection Example

In continuing with the AHP analysis of the car-selection problem, we need
to use the pairwise comparison procedure to determine the priorities of the
three cars in terms of the purchase price, MPG, and style criteria.
This requires that Dave express pairwise comparison preferences for the
cars, considering each of these criteria one at a time. Daves preferences are
summarized in the pairwise comparison matrices shown.
Price Car A Car B Car C
Car A 1 1/3 1/4
Car B 3 1 1/2
Car C 4 2 1
MPG Car A Car B Car C
Car A 1 1/4 1/6
Car B 4 1 1/3
Car C 6 3 1
Style Car A Car B Car C
Car A 1 1/3 4
Car B 3 1 7
Car C 1/7 1
The interpretation of the numerical values in the earlier tables is the same
as the interrelation of the preference values we observed for the comfort
criterion. For example, consider the comparison of car A and car B in
terms of the purchase price criterion. Car B ($11,200) is considered more
preferable than car A ($13,100).
In fact, the pairwise comparison matrix shows Daves preference for car B
is three times greater than his preference for car A in terms of purchase
price. Similarly, car A is only as preferred as car B. Recall that the
pairwise comparison matrix is set up to show the preference of the item in
row i when compared to the item in column j
Following the same synthesis procedure that we used for the comfort
criterion, the priority vectors for these criteria can be computed. The result
of this synthesis is shown below.
(
(
(

557 . 0
320 . 0
123 . 0
(
(
(

639 . 0
274 . 0
087 . 0
(
(
(

080 . 0
655 . 0
265 . 0
Price MPG Style
In interpreting these priorities we see that car C is the most preferable in
terms of purchase price (0.557) and miles per gallon (0.639). Car B is the
most preferable in terms of style (0.655).
No car is the most preferred with respect to all criteria. Thus, before a
final decision can be made, we must assess the relative importance of the
criteria.
In addition to the pairwise comparisons for the decision alternatives, we
must use the same pairwise comparison procedure to set priorities for all
four criteria in terms of the importance of each in contributing toward the
overall goal of selecting time best car.
To develop this final pairwise comparison matrix, Dave would have to
specify how important he thought each criterion was compared to each of
the other criteria.
In order to do this, six pairwise judgments have to be made: purchase price
compared to MPG; purchase compared to comfort; purchase price
compared to style; MPG compared to comfort; MPG compared to style;
and comfort compared to style.
For example, in the pairwise comparison of the purchase price and MPG
criteria, Dave indicated that purchase price was moderately more important
than MPG. Using the AHP 9-point numerical rating scale, a value of 3 was
recorded to show the higher importance of the purchase-price criterion.
The summary of the pairwise comparison matrix preferences for the four
criteria is shown in table below.
Criterion Price MPG Comfort Style
Price 1 3 2 2
MPG 1/3 1
Comfort 4 1
Style 4 2 1
The synthesization process described earlier in this section can now be
used to convert the pairwise comparison information into the priorities for
the four criteria. The results obtained are as follows
Criteria Priorities
Price 0.398
MPG 0.085
Comfort 0.218
Style 0.299
We see that the purchase price (0.398) has been identified as the highest-
priority or most important criterion in the car-selection decision. Style
(0.299) and comfort (0.218) rank next in importance. MPG (0.085) is a
relatively unimportant criterion in terms of the overall goal of selecting the
best car.
Using The AHP To Develop An Overall Priority Ranking

A matrix that summarizes the priorities for each car in terms of each
criterion is given below. This matrix is referred to as the priority matrix.
The overall priority for each decision alternative is obtained by summing
the product of the criterion priority times the priority of the decision
alternative with respect to that criterion. Recall that the criterion priorities
were found to be 0.398 for purchase price, 0.085 for MPG, 0.218 for
comfort, and 0.299 for style.
Price MPG Comfort Style
Car A 0.123 0.087 0.593 0.265
Car B 0.320 0.274 0.341 0.655
Car C 0.557 0.639 0.066 0.080
Thus, the computation of the overall priority for car A is as follows:
Overall car A priority = 0.398(0.123) + 0.085(0.087) + 0.218(0.593) +
0.299(0.265)
= 0.265
Repeating this calculation for cars B and C provides their overall priorities
as follows:
Overall car B priority = 0.398(0.320) + 0.085(0.274) + 0.218(0.341) +
0.299(0.655)
= 0.421
Overall car C priority = 0.398(0.557) + 0.085(0.639) + 0.218(0.066) +
0.299(0.080)
= 0.314
Ranking these priority values, we have the following AHP ranking of the
decision alternatives:
Alternative Priority
Car B 0.421
Car C 0.314
Car A 0.265
Total 1.000
These results provide a basis for Dave to make a decision regarding the
purchase of a car. Based on the AHP priorities, Dave should select car B.
If Dave believes that the judgments that he has made regarding the
importance of the criteria and his preferences for the cars in terms of the
criteria are valid, then the AHP priorities show that car B is the preferred
car.
USING EXPERT CHOICE TO IMPLEMENT THE AHP

Expert Choice (EC), a software package marketed by Decision Support
Software, provides a user-friendly procedure for implementing the AHP on
a microcomputer. We now provide an introduction to this software package
by showing how it cart be used to compute the priorities for the car-
selection problem.
Expert Choice enables the user to simply construct a graphical
representation of the hierarchy. For example, to create the hierarchy for the
car-selection example, the user selects the option to develop a new
application; what appears on the computers monitor is a request to define
the overall goal.
After the user defines the overall goal, a rectangular box, or node, appears
on the screen, with the goal description written directly above it.
The user selects the EDIT command and then the INSERT option;
another rectangular box or node appears below the goal node, and the user
now types the name of a criterion, which will be entered inside the box.

This process continues until all four criterion nodes have been specified.
The figure given shows the partial hierarchy appearing on the computer
screen after the four criteria have been specified.
In the figure we see that in addition to the names of each criterion , the
criterion nodes also contain the decimal value of 0.250.

This value represents the initial weight, or priority, given to each criterion
at the start of the EC session.

The user can now continue the process of using the EDIT command with
the INSERT option to define the decision alternative nodes associated with
each of the criterion nodes.

In the following figure we show the result of defining the decision
alternative nodes for the price criterion; now that since there are three
alternatives, the initial priorities are set at 0.333.
A similar set of decision alternatives is then identified for each of the
other three criteria.
Once the user has developed the complete hierarchy for the problem, he
or she can focus on any particular part of time hierarchy through time use
of the REDRAW command.
In fact, to show the detail displayed in the earlier figure, all we did was to
point to the price node (using the arrows on the keyboards numeric key
pad) and then type R for redraw.
Our intent here is not to attempt to show you how to use EC but merely
to let you develop some appreciation for the ease with which the analysis
can be performed using this software package.
Now that the hierarchy has been input to EC, we are ready to begin
developing the pairwise comparisons needed to establish priorities for the
decision alternatives.
In order to illustrate the type of approach used, we moved back to the
goal node with EC and then selected time COMPARE command by typing
C.
After selecting the option to make comparisons based on the importance
of the decision criteria, the EC system begins to go through the pairwise
comparison analysis.
One portion of this analysis, which shows the approach used by EC to
establish the comparative importance between the purchase price and MPG
criteria, is shown in following figure.
Note that this figure indicates to time EC system that price is moderately
more important than MPG. This process continues until all the entries in
the pairwise comparison matrix for criteria have been developed.
The synthesization process is then performed to compute the priorities for
the criteria.
The process of entering pairwise preferences for the cars relative to each
of the criteria was then performed in a similar manner.
The overall decision was then arrived at by entering the command S which
is an abbreviation for synthesizing; this command is used only when we
have entered all the data for the pairwise comparison matrices and want to
obtain an overall prioritization of the decision alternatives.
The priorities that were obtained after synthesization
The following figure shows the results obtained. Note that the results
indicate that the final priority for car B, the most preferable, is 0.422.
The EC system is a very helpful software package in performing the
multiple-criteria decision analysis of the AHP.
In addition to providing the overall priorities for the decision alternatives,
EC has the capability of doing what if types of analyses, where the
decision maker can begin to learn how the overall priorities for the decision
alternative are affected by changes in the preference input data.