You are on page 1of 22

The calculation in this Excel workbook is based on the wikipedia example available at the following loc

http://en.wikipedia.org/wiki/Talk:Analytic_Hierarchy_Process/Example_Leader
This Excel file is copyrighted by Dr. Kuanchin Chen 2012
Objective (or Goal):
Alternatives:
Criteria:

Choose a leader
Tom, Dick and Harry
Experience, Education, Charisma, and Age

ble at the following location:

Step 1: Pairwise comparison for "Experience"


Priority
Matrix 1

Experience Tom
Dick
Harry
Tom
1
1/4
4
Dick
4
1
9
Harry
1/4
1/9
1
CI: 0.01844777
CR: 0.0351721
Sum:

0.21717
0.71707
0.06577
1.00000

Priority values are th


the last matrix of ste

Step 2: Calculation of priority (i.e., normalized eigenvector)


Matrix 2

Experience Tom
Dick
Harry
Eigenvector Normalized Eigenvector
Tom
3 0.9444444
10.25
14.19444
0.21614
Dick
10.25
3
34
47.25000
0.71947
Harry
0.9444444 0.2847222
3
4.22917
0.06440

Matrix 3

Experience
Tom
Dick
Harry

Matrix 4

Tom
Dick
Harry
28.361111 8.5850694 93.6111
93.611111 28.361111 309.063
8.5850694 2.6003086 28.3611

130.55729
431.03472
39.54649

0.21718
0.71703
0.06579

Experience Tom
Dick
Harry
Tom
2411.6684
730.382 7963.15 11105.20367
Dick
7963.1533 2411.6684 26294 36668.57360
Harry
730.382 221.1987 2411.67 3363.24910

0.21717
0.71707
0.06577

Eige
sam
plac
add
eige
thes

Step 3: Calculation of consistency index (CI)


(1) Calculate the weighted rating for each row in matrix 1
Row 1:
0.66
Row 2:
2.18
Row 3:
0.20
(2) Approximation of Lambda(max)
Row 1:
3.0368955
Row 2:
3.0368956
Row 3:
3.0368955
Average:
3.0368955

Each row produces an appromixation of lambd


to be n (the total number of items in the comp
there three candidates to compare.
If any row lambda(max) is less than n, there m
ratings in the pairwise comparison matrix (i.e.

(3) Calculate consistency index (CI)


CI = (Lambda(max) - n)/ (n-1), where n is the number of elements that we compared in
CI =
0.0184478

Step 4: Calculation of consistency ratio (CR)


CR =

0.0351721

CR is calculated as CI divided by RI, where RI is the average CI from random matrices. RI


values may be difference across research studies. Here we use Alonso and Lamata's value
(0.5245) for 3 elements in the pair-comparison matrix (i.e., matrix 1).
CR > 0.10 indicates that there is a concern of inconsistency in pairwise comparison.

CR is calculated as CI divided by RI, where RI is the average CI from random matrices. RI


values may be difference across research studies. Here we use Alonso and Lamata's value
(0.5245) for 3 elements in the pair-comparison matrix (i.e., matrix 1).
CR > 0.10 indicates that there is a concern of inconsistency in pairwise comparison.

Note:

(1) Matrix 2 is matrix 1 squared. This involves the matrix algebra.


(2) Matrix 3 is matrix 2 squared. Continue creating matrices this way until "normalized eigenvalues
(3) An eigenvector is an array of values, each is called an eigenvalue.
Priority values are the "normalized eigenvalues" in
the last matrix of step 2 below.

vector)

Hint: How Matrix 2 is calculated.

malized Eigenvector

(1)
(2)
(3)
(4)
(5)
(6)

Matrix 2 is Matrix 1 squared.


Highlight the area where the "data values" of matrix 2 should be. In other
Enter "=MMULT(C4:E6, C4:E6)" without quotes in the top formula box abo
This creates a matrix manipulation formula in this form: {=MMULT(C4:E6,
Note that the formula is enclosed in {}. This is NOT the same thing as ent
The result should be in cells C11:E13.

Eigenvalues in matrix 4 are pretty much the


same as those in matrix 3 to the 4th decimal
place. So, it is reasonable to believe that any
additional matrices won't create any
eigenvalues substantially different from
these.

appromixation of lambda(max), which should be very close


ber of items in the comparison). In our case, n is 3 because
s to compare.
x) is less than n, there may be a problem of inconsistent
comparison matrix (i.e., matrix 1).

ts that we compared in matrix 1.

andom matrices. RI
o and Lamata's value

se comparison.

andom matrices. RI
o and Lamata's value

se comparison.

ormalized eigenvalues" do not change.

2 should be. In other words, highlight C11:E13.


e top formula box above the worksheet. Press SHIFT+CTRL+ENTER when done.
rm: {=MMULT(C4:E6, C4:E6)}
the same thing as entering the curly braces manually.

Step 1: Pairwise comparison for "Education"


Priority
Matrix 1

EducationTom
Dick
Harry
Tom
1
3
1/5
Dick
1/3
1
1/7
Harry
5
7
1
CI: 0.032444
CR: 0.0618566062
Sum:

0.18839
0.08096
0.73064
1.00000

Step 2: Calculation of priority (i.e., normalized eigenvector)


Matrix 2

Matrix 3

Matrix 4

Matrix 5

EducationTom
Tom
3.00
Dick
1.38
Harry
12.33

Dick
7.40
3.00
29.00

Harry
0.83
0.35
3.00

EducationTom
Tom
29.44
Dick
12.63
Harry
114.05

Dick
68.43
29.44
265.27

Harry
7.58
3.26
29.44

EducationTom
Tom
2595.35
Dick
1115.33
Harry
10065.47

Dick
6039.28
2595.35
23422.01

Harry
669.20
287.58
2595.35

EducationTom
Tom
20207450.93
Dick
8684030.72
Harry
78370111.40

Dick
47022066.84
20207450.93
182364645.10

Harry
5210418.43
2239146.04
20207450.93

Eigenvector
11.23
4.73
44.33

105.45
45.33
408.75

9303.83
3998.26
36082.82

72439936.20
31130627.68
280942207.42

Step 3: Calculation of consistency index (CI)


(1) Calculate the weighted rating for each row in matrix 1
Row 1:
0.58
Row 2:
0.25
Row 3:
2.24
(2) Approximation of Lambda(max)
Row 1:
3.06488758
Row 2:
3.06488758
Row 3:
3.06488758
Average:
3.06488758
(3) Calculate consistency index (CI)
CI = (Lambda(max) - n)/ (n-1), where n is the number of elements that we compared in
CI =
0.03244379

Step 4: Calculation of consistency ratio (CR)


CR =

0.061856606

nvector)
Normalized eigenvector
0.186227
0.078503
0.735271

0.188455
0.081012
0.730533

0.188394
0.080961
0.730645

0.188394
0.080961
0.730645

This time it takes matrix 5 for the eigenvalues


to remain stable. Again, compare eigenvalues
in matrices 4 and 5 to see if they are the
same to the 4th decimal place.

nts that we compared in matrix 1.

Step 1: Pairwise comparison for "Chrisma"


Priority
Matrix 1

Chrisma Tom
Dick
Harry
Tom
1
5
9
Dick
1/5
1
4
Harry
1/9
1/4
1
CI: 0.035633
CR: 0.067936428

Sum:

0.74287
0.19388
0.06325
1.00000

Step 2: Calculation of priority (i.e., normalized eigenvector)


Matrix 2

Chrisma Tom
Tom
3.00
Dick
0.84
Harry
0.27

Dick
12.25
3.00
1.06

Harry
38.00
9.80
3.00

Matrix 3

Chrisma Tom
Tom
29.69
Dick
7.73
Harry
2.52

Dick
113.61
29.69
9.67

Harry
348.05
90.89
29.69

Chrisma Tom
Tom
2638.87
Dick
688.72
Harry
224.69

Dick
10110.94
2638.87
860.90

Harry
30992.42
8088.75
2638.87

Chrisma Tom
Tom
20890849.59
Dick
5452327.40
Harry
1778761.68

Dick
80044275.45
20890849.59
6815409.25

Harry
245354733.15 346289858.19
64035420.36
90378597.35
20890849.59
29485020.52

Matrix 4

Matrix 5

Eigenvector
53.25
13.64
4.33

491.35
128.31
41.88

43742.23
11416.34
3724.46

Step 3: Calculation of consistency index (CI)


(1) Calculate the weighted rating for each row in matrix 1
Row 1:
2.28
Row 2:
0.60
Row 3:
0.19
(2) Approximation of Lambda(max)
Row 1:
3.071265313
Row 2:
3.071265313
Row 3:
3.071265313
Average:
3.071265313
(3) Calculate consistency index (CI)
CI = (Lambda(max) - n)/ (n-1), where n is the number of elements that we compared in
CI =
0.035632656

Step 4: Calculation of consistency ratio (CR)


CR =

0.067936428

envector)
Normalized eigenvector
0.74766
0.191576
0.060764

0.742732
0.193959
0.063309

0.742867
0.193882
0.063252

0.742867
0.193882
0.063252

ents that we compared in matrix 1.

Step 1: Pairwise comparison for "Age"


Priority
Matrix 1

Age
Tom
Dick
Harry
Tom
1
1/3
5
Dick
3
1
9
Harry
1/5
1/9
1
CI: 0.014532
CR: 0.027706
Sum:

0.26543
0.67163
0.06294
1.00000

Step 2: Calculation of priority (i.e., normalized eigenvector)


Matrix 2

Age
Tom
Dick
Harry

Tom
3.00
7.80
0.73

Dick
1.22
3.00
0.29

Harry
13.00
33.00
3.00

Matrix 3

Age
Tom
Dick
Harry

Tom
28.07
71.00
6.65

Dick
11.09
28.07
2.63

Harry
118.33
299.40
28.07

Age
Tom
Dick
Harry

Tom
Dick
Harry
2362.36
933.63
9962.46 13258.45 0.265433
5977.47 2362.36 25207.99 33547.82 0.671625
560.18
221.39
2362.36
3143.93 0.062941

Matrix 4

EigenvectNormalized eigenvector
17.22 0.264776
43.80 0.673386
4.02 0.061838

157.49 0.265443
398.47 0.671605
37.35 0.062952

Step 3: Calculation of consistency index (CI)


(1) Calculate the weighted
Row 1:
Row 2:
Row 3:

rating for each row in matrix 1


0.80
2.03
0.19

(2) Approximation of Lambda(max)


Row 1:
3.029064
Row 2:
3.029064
Row 3:
3.029064
Average:
3.029064

Each row produces an appromixation of lambda(m


close to be n (the total number of items in the com
3 because there three candidates to compare.
If any row lambda(max) is less than n, there may b
inconsistent ratings in the pairwise comparison ma

(3) Calculate consistency index (CI)


CI = (Lambda(max) - n)/ (n-1), where n is the number of elements that we compared in
CI =
0.014532

Step 4: Calculation of consistency ratio (CR)


CR =

0.027706

d eigenvector

mixation of lambda(max), which should be very


er of items in the comparison). In our case, n is
dates to compare.
ss than n, there may be a problem of
irwise comparison matrix (i.e., matrix 1).

that we compared in matrix 1.

Step 1: Pairwise comparison for all criteria


Matrix 1

Criteria Experience
Experience
1
Education
1/4
Charisma
1/3
Age
1/7
CI: 0.039473

Education
Charisma
4
3
1
1/3
3
1
1/3
1/5
CR: 0.0447789844

Step 2: Calculation of priority (i.e., normalized eigenvector)


Matrix 2

Criteria Experience
Experience
4.00
Education
1.04
Charisma
2.13
Age
0.44

Education
19.33
4.00
9.00
1.84

Charisma
8.73
2.02
4.00
0.94

Matrix 3

Criteria Experience
Experience
72.58
Education
16.72
Charisma
35.70
Age
7.40

Education
308.63
71.56
152.41
31.59

Charisma
147.38
34.06
72.81
15.03

Matrix 4

Criteria Experience
Experience
20840.85
Education
4816.77
Charisma
10274.45
Age
2128.56

Education
88941.97
20556.50
43848.18
9084.03

Charisma
42405.31
9800.80
20905.72
4331.03

Matrix 5

Criteria Experience
Experience 1724877141.32
Education 398657012.84
Charisma 850359036.64
Age
176168972.27

Education
Charisma
7361236219.10 3509648555.69
1701343458.99 811156908.41
3629066436.00 1730245762.60
751834080.42 358455199.09

Step 3: Calculation of consistency index (CI)


(1) Calculate the weighted rating for each row in matrix 1
Row 1:
2.26
Row 2:
0.52
Row 3:
1.11
Row 4:
0.23
(2) Approximation of Lambda(max)
Row 1:
4.1184180243
Row 2:
4.1184180243
Row 3:
4.1184180243
Row 4:
4.1184180243
Average:
4.1184180243

(3) Calculate consistency index (CI)


CI = (Lambda(max) - n)/ (n-1), where n is the number of elements that we compared in
CI =
0.0394726748

Step 4: Calculation of consistency ratio (CR)


CR =

0.0447789844

Priority
Age
7
3
5
1
Sum:

0.54757
0.12656
0.26995
0.05593
1.00000

ed eigenvector)
Age
41.00
9.42
21.33
4.00

Eigenvector
Normalized eigenvector
73.07 0.54848
16.47 0.12366
36.46 0.27372
7.21 0.05415

Age
696.37
160.98
342.79
71.22

1224.95
283.32
603.70
125.23

0.54754
0.12664
0.26985
0.05598

352526.40
81476.65
173794.38
36005.04

0.54757
0.12656
0.26995
0.05593

Age
16580872423.72 29176634339.83
3832203994.37
6743361374.61
8174318253.26 14383989488.50
1693474383.96
2979932635.75

0.54757
0.12656
0.26995
0.05593

Age
200338.28
46302.59
98766.03
20461.42

of elements that we compared in matrix 1.

Leader
Criterion
Tom
Experience
Education
Charisma
Age

Criterion Weight Leader's weight Weighted Score


0.547569239
0.2171656128
0.1189132094
0.1265552844
0.1883940966
0.0238422685
0.2699499225
0.7428666222
0.2005367871
0.0559255542
0.2654333434
0.0148445068
Sum
0.3581367717 <-- SECOND BEST

Dick

Experience
Education
Charisma
Age

0.547569239
0.1265552844
0.2699499225
0.0559255542

0.7170650349
0.080961232
0.1938816334
0.6716254505
Sum

0.3926427554
0.0102460717
0.0523383319
0.0375610255
0.4927881846 <-- BEST

Harry

Experience
Education
Charisma
Age

0.547569239
0.1265552844
0.2699499225
0.0559255542

0.0657693523
0.7306446714
0.0632517444
0.062941206
Sum

0.0360132742
0.0924669442
0.0170748035
0.0035200218
0.1490750437 <-- LAST

<-- SECOND BEST

Matrix multiplication
Matrix 1

Experience Tom
Tom
1
Dick
4
Harry
1/4

Dick
1/4
1
1/9

Harry
4
9
1

Our goal here is to multiple matrix 1 by itself (i.e., matrix 1 squared). There are two ways to do it.
(1) Solution 1 - Excel MMULT function
Experience Tom
Dick
Tom
3.00
0.94
Dick
10.25
3.00
Harry
0.94
0.28

Harry
10.25
34.00
3.00

(2) Solution 2 - Manual calculation


Experience Tom
Dick
Harry
Tom
3.00
0.94
10.25
Dick
10.25
3.00
34.00
Harry
0.94
0.28
3.00

Step 1: highlight the destination


If needed, format these
Step 2: type "=MMULT(C3:E5, C3
Don't press ENTER yet.
Step 3: press CTRL+SHIFT+ENTE
The result is "{=MMULT
This is NOT the same t

See WikiPedia article at http://en.w


Read especially the "Non-technica

two ways to do it.

hlight the destination cells (i.e., C11:E13)


needed, format these cells as decimal numbers. Otherwise, you will see fractions.
e "=MMULT(C3:E5, C3:E5)" without quotes in formula box above the worksheet.
on't press ENTER yet.
ss CTRL+SHIFT+ENTER to turn the formula into a matrix formula.
he result is "{=MMULT(C3:E5,C3:E5)}" without quotes.
his is NOT the same thing as just adding { } around the formula.

ia article at http://en.wikipedia.org/wiki/Matrix_multiplication
ally the "Non-technical example" section.

You might also like