Vous êtes sur la page 1sur 19

SDA 8: The Analytic Hierarchy

Process
The importances of the criteria could be
approximated by the AHP using pairwise
comparisons:
T. L. Saaty, The Analytic Hierarchy Pro-
cess: Planning, Priority Setting, Re-
source Allocation, McGraw-Hill, New
York, 1980.
Suppose that the value function has the
form
1
v(y) =
q

i=1
w
i
y
i
If w
i
= 0, the corresponding outcome y
i
can be deleted from consideration. Thus,
we shall assume that w
i
> 0, i = 1, 2, . . . , q.
Dene the weight ratio by
w
ij
=
w
i
w
j
.
Note that, for any i, j, k indexes
w
ij
= w
1
ji
, w
ij
= w
ik
w
kj
.
2
Dene the matrix of weight ratios as W =
[w
ij
]
qq
:
_
_
_
_
_
_
_
_
_
_
_
w
1
w
1
w
1
w
2
w
1
w
3
. . .
w
1
w
q
w
2
w
1
w
2
w
2
w
2
w
3
. . .
w
2
w
q
. . . . . . . . . . . . .
w
q
w
1
w
q
w
2
w
q
w
3
. . .
w
q
w
q
_
_
_
_
_
_
_
_
_
_
_
Amatrix W is called consistent if its com-
ponents satisfy the equalities w
ij
= w
1
ji
,
w
ij
= w
ik
w
kj
for any i, j and k.
3
Observe that:
Since each row of W is a multiple of
the rst row, the rank of W is one, and
thus there is onlu one nonzero eigen-
value which is q.
This due to the fact that w
ii
= 1 and
that the sum of all eigenvalues is equal
to the trace of W (i.e.

q
i=1
w
ii
= q).
We can easily chack that
Ww = qw
therefore w must be the eigenvector of
W corresponding to the maximumeigen-
value q.
4
As a living system, human perception and
judgment are subject to change when the
information inputs or psychological states
of the decsion maker change.
A xed weight vector is difcult to nd.
Saaty proposed the following to overcome
this difculty:
Estimate or elicit the weight ratio w
ij
by
a
ij
and let A = [a
ij
]
qq
be the matrix of
components {a
ij
}.
Note that as each w
ij
> 0, we expect and
shall assume that all a
ij
> 0.
5
Furthermore, as w
ij
= w
1
ji
, Saaty sug-
gested that in practice, only a
ij
, j > i
need to be assessed.
Since A is found as an approximate for
W, when the consistency conditions are
almost satised for A, one would expect
that the normalized eigenvector correspond-
ing to the maximum eigenvector of A, de-
noted by
max
, will also be close to w.
Theorem1. The maximumeigenvalue,
max
,
of A is a positive real number.
Let w be the normalized eigenvector cor-
responding to
max
of A. Then w
i
> 0 for
all 1 i q.
6
Theorem 2. The maximum eigenvalue of
A satises the inequality

max
q.
Assume we have q objectives and we want
to construct a scale, rating these objec-
tives as to their importance with respect
to the decision, as seen by the analyst.
We ask the decision maker to compare the
objectives in paired comparisons.
If we are comparing objective i with ob-
jective j, we assign the values a
ij
and a
ji
as follows:
7
a
ij
= a
1
ji
If objective i is more important than
objective j then a
ij
gets assigned a num-
ber as follows:
Note that the above observation is valid
for any matrix which is consistent.
8
Intensity of
relative importance Denition
1 equal importance
3 weak importance
(of one over the other)
5 strong importance
7 demonstrated importance
over the other
9 absolute importance
2,4,6,8 intermediate
values between
Saatys scale of relative importances.
9
Example 1. Let us consider the following
matrix
A =
_

_
1 9 7
1
9
1
1
5
1
7
5 1
_

_
To nd
max
we solve
det[A I] = 0
that is,
10
det
_

_
1 9 7
1
9
1
1
5
1
7
5 1
_

_
= (1 )
3
3(1 ) + 9/35 + 35/9 = 0
The maximum solution is

max
= 3.21.
After normalization we get
11
research
growth
benefits
colleauges location reputation
A B
C
overall satisfaction with the job
level 1: focus
level 3: alternatives
criteria
w
1
= 0.77, w
2
= 0.05, w
3
= 0.17.
We illustrate Saatys method on a job se-
lection problem (3 alternatives compared
on 6 criteria)
Choice of job.
12
The question asked was, which of a given
pair of criteria is seen as contributing more
to overall satisfaction with a job and what
is the intensity or strength of the differ-
ence?
res. growth benets coll. location reputation priority
research 1 1 1 4 1 1/2 0.16
growth 1 1 2 4 1 1/2 0.19
benets 1 1/2 1 5 3 1/2 0.19
colleaug. 1/4 1/4 1/5 1 1/3 1/3 0.05
location 1 1 1/3 3 1 1 0.12
reputation 2 2 2 3 1 1 0.30
Pairwise comparison matrix of criteria.
The relative weights of criteria (priorities)
13
can be computed as normalized geometric
means of the rows (which are very close
to the eigenvector corresponding to the largest
eigenvalue of the matrix)
The geometric means are computed as
m
1
=
6
_
1 1 1 4 1 1/2
m
2
=
6
_
1 1 2 4 1 1/2
m
3
=
6
_
1 1/2 1 5 3 1/2
14
m
4
=
6
_
1/4 1/4 1/5 1 1/3 1/3
m
5
=
6
_
1 1 1/3 3 1 1
m
6
=
6

2 2 2 3 1 1
So, the relative weight (priority) of the
criterion research is obtained as
p
1
=
m
1
m
1
+ m
2
+ m
3
+ m
4
+ m
5
+ m
6
15
Then we compare the alternatives on each
of the criteria
research A B C priority
A 1 1/4 1/2 0.14
B 4 1 3 0.63
C 2 1/3 1 0.24
growth A B C priority
A 1 1/4 1/2 0.10
B 4 1 3 0.33
C 2 1/3 1 0.57
16
benets A B C priority
A 1 3 1/3
B 1/3 1
C 3 1 1
colleaug. A B C priority
A 1 1/3 5
B 3 1 7
C 1/5 1/7 1
17
location A B C priority
A 1 1 7
B 1 1 7
C 1/7 1/7 1
reputat. A B C priority
A 1 7 9 0.77
B 1/7 1 5 0.17
C 1/9 1/5 1 0.05
18
We obtain
0.16
_
_
_
0.14
0.63
0.24
_
_
_
+0.19
_
_
_
0.10
0.33
0.57
_
_
_
+
+0.30
_
_
_
0.77
0.17
0.05
_
_
_
=
_
_
_
0.40
0.34
0.26
_
_
_
A
B
C
So job A should be selected as the best
alternative.
19

Vous aimerez peut-être aussi