You are on page 1of 20

SERIE RESEARCH mEmORfMDn

QUAIITATIVE .MULTIPLE CRITERIA CHOICE ANALYSIS

THE DOMINANT REGIME METHOD Edwin Hinloopen P e t e r Nijkamp Researchmemorandum 1986-45 December 1986

VRIJE UNIVERSITEIT FACULTEIT DER ECONOMISCHE WETENSCHAPPEN A M S T E R D A M

QUALITATIVE MULTIPLE CRITERIA CHOICE ANALYSIS

THE DOMINANT REGIME METHOD

Paper presented at the Conference on Conflict Management, Amsterdam, April 1986

Free University Dept. of Economics Amsterdam RE/192/PN/hl

Edwin Hinloopen and Peter Nijkamp

o 3"
O HO

co

TC H I OJ g -

O Hl C D

>
I-

co
B
C D Cf O

B B
3
ct

C P> 1

o 3"
O
!-

P O

po
O

C_i.

C D

< C O
H-

co

C D ct P O H

<! DJ
M C ct HO

!- Ct
DJ ct

t-> H-

P O

3"

C D O ct

po co ra o ca

C D

< C O
v_x

C D

ra o 3 < C D
3 ct
M-

B
PJ ct H-

po t->
C D

<

(-

co o po
H C D C O

3*

co

P>

(- <i
C D

po
h" C O

HC O

< C D

s C D
ct

C D Ct

Ct i

O 3
--v

e.
ra ~i

B a P O
ra
ct
v_^

o.
C D C O

C O

H O

3 O3J

O Hl

ct HO C O

PJ O

co
!-

&
ct H1

(-> C D
C U

PJ HW C U

<i

O H. H-

O O

ct

po 3
ct

<!

C O HC O

ra

^H
ct C D *S
(-

C_i.

er

ct H-

er
H

c po
ct O
!-

co

3O

er
3

C D C D H)
t-1-

o. co -* *
>-i 3"
C D C O C D .O

c o * 3"
B
C D ct

VD C VJ1
v_^

3 B
C D ct

(-" < co
C O

1 o po 3
3 O ct

3"

3*
O

o o. a co
C U

!-

C O

o. co
H.

ol * H ct a co cO P O
T Hl M M

l-' vj

po

a co v_^ o T
ra
T
P O

O 3 C u H
C D M

!-

*-*
>

!-

po

B
C D C O C O

Ct C D HP O O -b
ct-

o *-4
P > ra 3"

Hi H>

o T
O

er

-i

po
fc

a <! H- C D
B
C O H-

<
B

H C D H-

C O ct HCt

o e ct
C D T

**_^ <
ct

C D

O 3 O
Hj

ra 3-

B
O -5 C D ct

&

C D

c H C

o ct

a C O
O
|_u

3"

!-
< D
C O
v_^

C D

e PJ
ct HHl

ct

co

!- C O

o O 3
O O T

C O H"
-

fu 3

<! po

H C D
a o 3
po
T P >
O

o
o

H-

P O ct

(-
PJ

er

C D

a H3
P O H"

< *- C O O
co
3 B
C D Ct

SO

C D

C D

er

CT o v-*- 3 P HO

vo -a

co 3

PJ

&
ct H1 PJ ct ct

B 3

C D PJ H"

3 H. Q T B
PJ ct HO

^-^
fu 3
O. ct

ct H-

B
C D ct

3
Hl O T C O PJ

<!

ffi o B H ra C D < B !-
N

*t

ra po D. co
K 3"

<

c c

O 3 po
M

3O

HO

co
3

ra o H
h-1 CD

PU

ct

C D ct O

C O
o C U
-i
C D

C C T


ct

3"

H-

<

a T
O

3H-

C D

fu

e C O
M H-

1 C D O C U
(-

T
C D

c+
C U ct H-

er co

P O C D

H' ct

B C

o<! oC O o 3" 3
C D T C D

H C D

o C o N (- e N B co << o (-" (- 3 o OQ T
ct ct

co 3

*<;
O

3*

po O

3*
O H-

o o ct
H-

H C D
!-

te

*
ra o co

O HO C D a T O
KC D

a M
HC D

ct

B C

po 3

3 B

Hl O T

a
po co
C D ct

a PJ
O C D

H ct
(-

i-1i-1S
C O C O O-

.O
C
PU M Hct PU ct HC D

o T rawHJ

-i C D C O C D

C D C U C O

ct HO

B po

cr1 B
B
C O O

H' C D

3 pu

3"
i-1
!->

H O
3"
OH-

t- -

i-"P U H1
C_i.

er 1
3

3 oq

(-
c+ 3"
C D

po 3

C D
B

o 3

-s
C D O Ct HO

a o Oi
C O C D X3

a M o < co o HOq co
O P O H M T

po

ra T i-1ct C D

O ct a i c o C D t> i >-t <t> C D ra H* c t Ct H- 3" C D <j C D T

3"

!-

ra

C D

o 3
C U C O

o *D C < C D
T H-

3 oq ra H. 3" C
O HM
<

tr < P
C O X) C D

T i-1' er
ct C D

3
yv

a c o ^.^ a < c o ra c t (- er D co C
HH-

Hct C D PJ

o 1

ra

ct H-

ra 3"
O

3"

a co v

Q. 3co
C D T C D ct O

P tC H

C D C U C O

a 1

ct

o 3

!- po
ct C D
C_i.

a
3

C D ct

a po

>
H1 ICI-

-s ol I-" H

C D

Hl

ra C D o T o cr \ffi B
co co
C D C D
V

C D H-

o * T
B
PJ ct

3"

co
M

B po T

3
ct HC O

ra1
Hl O T C O

& C D
ra
f*
t-"

oq
T C o
ct H' O

ra o H- 3
H-

H.

ra

PJ

ct
!-

ra xD po C
co

s co !- H O
M C O C O O

H C O

^/
HC O

3
O

Cr

co

3
O Hl PJ C O

a c t ^N C D -i o !-.. oq ra H> c H- o p- ra PJ
CD C O

ra. co 3

O C D t^- C O C O o C D B "O T C D H> O 3 o a c t T Ct (- O I-"- O H l Hl

3-

co

HDPJ ct ct C D ct HO

>-
ra
C O ct HO 3

O-

<! < p
G C c
H O

H> i-^

S-

C O

O-

oq
B 3

co C D
B
C D ct

B C u
TT C D O O

3 B
C

C D C D

ct

O H>
S CD M C D

3
ct

o 3
ct C D O T

3
C D PJ C O

H-

po
C D Ct

C D C O C O DJ

B O 3
C D ct

o >-1>
ct C D

o T

C U O ct
(-

3
ct

ra po
co co
o 3
C O C C O

3
C U H

o T
3

H
3

Ct CD

ct

!-' cr
O

yN

3
O Hl T C D
(-

<<!
B
C D C O

Q T B

s~^

CD

3-

3"

H-

O.
PJ

(-<
O P O

P O ct HO

oq

3 c;
C D 1
!-

ra T

C
(-

< C O
H-

l- 1

3 3

<

<<

?r

C D

c 3

o. <<
er

C D.
Oq C D

s o c T M
ct H"

3
H"-

B
C D P O C D

*D C

3" O a co
x-^v

C U M

po T

O C D C O

-s % C U o. CO C D er
1

O
w

*< C O
HC O

< 1
er po co
C D

o ct
po

i-1 C D

< C U
!-

c ra

-b -4 C O

C D

<<:
ct C D

B 3

o t-"
C D O *1 H* ct C O T
"

3 O

3
C D

H
po er
M

ct M

3"

<-=!
ct

C O HO OQ 0 3" p ct & CD

3" ra co T
Hf-

ra 3"
PJ

po H1
I-

po co ct co T
i

3 oq
ct

<
B
PJ

O "J
Hl

ct PJ

3
T C D Oq P O T

cr

Hct C D T H-

C U

<

3
ct

<

er
M

co
C D O Hl C O

C D

B PJ C i-* D

ra

po
y

o ra- M ra co c 3- X C D c 3 3 ra co c: P o e Ht < C c PJ 3
ct

3"
O

c 3 O O

co

C D

ra 3"
O
H-

ct Hct

i- 1
M

H-

0 B
C D

<! ol PJ H
3
ct C D H-

3"

C D C O

*<
C 3
O. C D -3

C N
N

sC

a H3 oq
C D PJ

co
O Hl

C 3

<! PJ
M

ra C D o o ct
H-

C D ct

<! ra o (-
3
C D

raHB 3
C O C O

ffi 3

Cu

ct

po

C O O Hl C D P O

C o
O Hl PJ

!- a C
o 3
t-- t
PJ C

!-

co C

ra 31
pu co

o T
C O C O

a
B
C D ct

M C U M

er

Po
O

C U

<
C D C D

ra

C D Ct

Ht-

B
C D et O OC O

3 fu1 B co po
1

3"

M C po O 3

<! po

co

3"

-i
C D C O C D

C D OC D C O H-

3
O C D

ct O O H C O H.

C O

3"

3*
O

co ^
co

o.
3 P1 >
H-

3 Oq

ct C D T
!-

VO 3
k

er o
C O

<<;
H-

<!
B 3
ct HO

n
2 B
ct

C O

ra
-i
ct

Hl O

rato
PJ Oq "5 C D PJ ct

po 3

C D

C D

ct H-

H.
C_|.

c S oq po o
ct HO

a o *
o T O

o. C O
s~^

o
ct

Hii

O-

co ct
H-

H-

po

B P) O 3

c o 3
C O

h-> ct H- co 1 HPJ O

B C

o. C D ra H3

Oq 3"

&
B

<

WJ H) T

ct C O

ra 3-

H--

O 3 ca

&
ct H-

ra

o T
H-

ct H'
^ ^

PU M

co

3
-?>

>-* C D
T C D

ra co

a 3D C a co
C D ct

3
O

<<: C O
ct

ct O -J O O

o H>
ci-

H C D

PU

oq 3

O H)

O "S

co 1

c ra

3PJ ct

3*

T 1

Ct

^^ o *s

<i C D

o 3 a

ffi B
H> O -5.

<
o T

3"

C D l-1

ct C D S C O

B po

ty

C D -i C D

Q * ra B a1 O hH3" ct ra CD O ra 3" 3* C D o ra C-i. 3 l CD O ra ct c H Ho & H" CD CD T T a ra >- O H-PJ oq 3


C D
PJ M

ra 3"

o o C a !3
c C

<!

C D

!- 1

!-
O C D

co

ra
Ct H-

3 3

ra
ct

H.

!-

< C D

(- C D

C D ct

!-

<i 1 i

3 oq

-"

o s o 1

ct C D

er1
h-

K^

po

ffi B co

a C

c c

>->

c o
et P) et HHO

H-

M H

H M C D -S C D

P)

3 et oq 3"
C D 1
o

~J O

<

o G co T rt
C D C O C D

M>
T

3 oq

O O

O PJ H1

et
>-4

oq p) H- M B B
C D C O HC D 3 ct 13 J V H HO O G
Hrt

3
et

C D rt O

3"
C D C O C D

3 O Hrt ct co O S C D 3 O 3" B > e t cr (- (- CD 1 c t << C D G C O D Iet C D 3 C D << 3" 3 e t c o 3"


O

P> O O C D C O C O H-

Oq C D O

o a B 1 <! _ a CD CD C P) vO > c 3 Ct P T 00 o B o c t 3* H H- * H' O O C D H * ^


C O
!-

w-A

T C O rt C O O

(- B

3"

a c < o o
3
et C D O ct

ct H

C O

et O

pi

a o H
O

O
3"
O

a
3" T
C D C D

C D

H H3

p >

CO

<! C D
CT C D O O

O. O H O O pi Oq c t HC D O oq ( HO H

O 't

rt

P)
et HC D H-

co c o
-4

P> 3 CL ct

<i

& Ct ^-^ 3 P) <j .Q c t p) H p O C 3" r t O c t *r 3 ct H P CD ! - " 3 > H O O. 3 * 3" * 3 C D e C D H - M) C D


^'

VO -^1 -3

n
C D H H-

p >

VO

oo 0 0 oq o CD * 3

<o

C D H-

C D

C O HM

O O HO CD

et

e C D

3"

ct C D O

C D H-

<

1
M
(-

C O

3" B
C D HX C D

Q
B 3
C D

o - -i
CT

ct C D C O C D >

o Hc o !-

oq B 3O

p >

p)

-4 O T

3- B

C H ct H-

D ~J -q
v>

W HC D ct
H-1 D.

Ct

H"

^-^ X o rt T
-J

C D

Pi et H-

C D t- 1

O T HC O

< C D

C O
K-*'

C D O >"4 et

, ^
C O C D C D
h*

oq
HO pi H

C D ct

*<
T C D O C D rt l-1

n
P>

3"

o 3

O T C D

C D

O T O T O. H3 P> M H3 i> O -i ct

ca
H1 C D

Pi et HO P> C O

a. H-

ffi o B T
C D

<! C D
C D PJ M rt HO

&
O "-4

HT C D C O

P >

CT C D

3- oq 3 3 3" H- r t rt 1 *- 3r 3 CD e oq C D

o o T
rt

s e C D
H-

oq 3"
rt H" 3

< HO

ff P >

c oo
C O

CT S B CD O C D C rt C D

O
C D Pi ct ct O

oq B

EC

31
C D

o ~i

3 M O O

C O O

o o

(-" c o
C U

C D 3" P 3 G pi

c o "S 3

t-1-

o ^ C D
3
C D rt P> 3 C D

" O s C D HC O C O C O C D

o o
3
rt

T C O

B c o p) ^_- O O 7\ C D P > 3
*3 C D C O O HC D rt H"4
H-

HT M

>->> tu

c 3 o

o M
HO

C D

p) T

CT C D O C C O C D

*< a o T3
H

P)

3 O H-

o (-< C D
C O O

P >

et C D

e
P)

!-
0 w

P >

= s
C D rt O

P >

o C D
C O C O H-

*-*
et H-

co c *-^ o 3

3"

vO 00

3" *

CT C D o et o 3C 3 C D 1 C D H" 3 CD rt

a.

(-

a
p>

3
PJ M

<
P>
O O 3

a P)
!-

T3 rt C O H> C D O C 3 P) i CT I-" o C D p >

c o

B rt C O C D 3" 3 CD X * e <> o P) c o M H- -t> oo C D !- H- P) O T o O N C D ^^ r t Ct 3 P > P > P 1 CD r t > et B I-" C a H-

o.
OP) ct

HO T Ht

B <! C

C P > 3 B
C D ct O (X C O 3* P> C O CT C D C D 3 G C D
t- O

H1 et HC D

co CD H- 3 W C

*_

c
3

o1
pi C O T C D C D C U H C D G

O G C O

3- 3

C D et

rt

a P!
*

3* P O > a 3" c o
C D G et O ct

a1 ho
*5

*-> ^ ^ co P < ct > rt P >


!- O
3
C D O ct C D

o *D CD C H- n
Q_ 9
rt C O O P> 3 CT C D P> C O C O C D C O C D G O

H-

< C D
1

C D et

a B o e ha H3
Pi H H-

3" c o O o < C U G H B

C D 3 O

ct

31
C D T3 C D T

1
-4 HO et
!- Ct

<

3"

ct

O "-b et

O H" I-1

C u

3 p1 >
i-

a
3" P) C O C D C D T

cr
H-

3-

<< et
H--

O < 3 rt
-

O C D C O
ct

et Ho H C D O T
HCt

^-^ C O
C D C D

H3

C D rt

PJ H1 C C D

<!

N _ ^

3-

3"
C D C O

*->* cr P > c o C D G
O

C D P> ct ct C D

(-

C D ct

3"

et C D T
!-

C D -i

sD C
-4 C D T

C
O

oq 3
C D ct

3"
O

a ct r
P> rt H3

B
Ct

C D O

3O
Q. C O C D C D W P> P> ct

P 3 >

*-

o
C D -4 C D T C D

o C D
C O C O H<J C D <3 Pi M C CD O T Het C D T
(-

a
* ^ N

3" ^ .
.**v

PI
M

rt C O

ffi 3
C D

HO C C O

O C D

ffi 3
B
O

C D ct O

3 P > ca

3 (-
ct O 3

-*
35 C D H3^ rt

o H^
CD et 3" C D M C D C O C O O O T -i C D O et et S C D C U rt 1

< O
OQ
V*

H-

C D rt O O.

(-" (- et <<
pj

3*
C D T C D

CT C rt OQ O O

CD HT

<: (- o O
O

(- i-'-

3 p1 > P l- c > o
C D ct O

M> O T et HO 3

ot r
cr
C D ct C D

c o
et rt HC O ct

C D
C_i.

-s

<! C D
T P> H O

>-3

.a
H'S C D 3 ct C O

P >

3- 3* B
C D
N

3 G B Pi et

o.
rt O

c o
C D O rt HO 3

o -J
C D C O C D

oq
V

C D CX

ffi 3D CD B C 3" 3
C O

(-

o p>
H 1 rt

rt C D C O C O 3" ct 3 H- H>-4 . O H" D C D C C O

B p)

C D T HP> O 3" O HO C D

o O a a

3-

C O C D C D

o a
3
C D rt O

<
H-

<

c o

c o

/v

<

C D C D

l-1

c o
ct 3 ?r C O ct O ct C D

* v 00 co v_^

ca

C D C D

C D M -4 ct P> 3 Q.

3"

c oq o p)
^ - y

a s C D
-4 C D T C D

C D et

O C D

ca

3"
C O
^-s

o.
*^N

*< < * + C D
B
HC D ct O VO -3

&

C D
Ct

HM H

3
ct

C o
C D -4 H1 C D O rt
H*

3"
O

3- ^ p>
et C D C O C D

*
C O

f-"C_i.

C O C D C D

o C D

v-> C D
!-

O C D C D ct 3" O

3 3

CT C O C D G O 3
!-

P)

TT

3"
C D ct

C O H>

rt

o
s-^

ro O co lP) C D O -1 " *i I I

P >

3 3* P CD o > C D rt B 3 !-
O

o1 hC D

3 P > 3* C D 3

a.
CT

3"

s~^

et c r
(-

pi
et 1

c o o
3"
P)

P J a B 3 o CD -!> et 3" O ct T O P > 3* a r t 3 P c > o O.


rt et 3" C D HT et 1 C D C U rt

s PI
ct o c
C D 3 CT T O C D TT P>

* -^ Q. ca H- % * C D > -3 O
N P) C O

B 3" P > a CD H3 c CT M o P < >


O CT P) C O C D G O C O H-

3*

-i

P)

P 3 c 3 o 0. o. 3*
T3 C D

C D C O O -4 H>-4

a c o ^^ CT
P >
C O C D 0.
O 3

c o o
-4

CT C D G C D O ct C D G ct O

oq
ct

3"
C D T C D

rt 3" C D

cr
C rt

+
H1

PJ

C D C O C O

3"
~i

C D ct

g P >
ct C D M

sa HC_i.

O O

<

?T
0}

c o

31

oq

a o
-

o.

f O O ct

C U C O

o.

3 et C D 1
*-

3
O ct G P) H

o. H>-4 -4 C D T C D

<l

p)

c o H) 3 c t1 CD P 3 -5 >
C D 1

<

H P)
ct HC D

C D

rt

<

3
while next various new elements and extensions of this method will be developed (section 3 ) . For pedagogical reasons our presentation will be based on some numerical examples. It will be concluded that the regime method is able to encapture a wide variety of qualitative multiple criteria choice problems based on both ordinal and mixed ordinal-cardinal data regarding both the characteristics (or impacts) of a choice option and the weights (or priorities) of a choice maker. 2. Principles of the Regime Method The regime method for qualitative multiple criteria choice analysis

is based on the following considerations; - the technique should not use methodologically unpermitted operations (for instance, summation or multiplication of ordinal numbers) - the technique should maker - the technique should be easily applicable on a computer - the application of the unambiguous solution, identified. In the sequel Suppose we alternatives of this regime method should in principle lead to an so that always a dominant choice option is be as much accessible as possible to a choice

section,

the

essence and structure of the

regime method will be further described. have a discrete choice problem with I choice options or criteria j information we have is composed of qualitative ranking of all I i (i=1,...,1), characterized by J judgement

(j1 , . f J) The basic .

data regarding the ordinal value of all J judgement criteria for all I choice options. In particular we assume a partial choice options for each matrix can be constructed: criterion j , so that the following <v effect

E -

*u
(2.1)

ir

e u

The entry e^j (i=1,...,I; j=1,...,J) represents thus the rank order of alternative i according to judgement criterion j . Without loss of generality, we may assume a rank order characterized by the condition 'the higher, the better', in other words: if ejj > e i i j , then choice option i is preferabie i' for judgement criterion j . As there is usually not a single dominating alternative, we need additional information on the relative importance of (some of) the judgement criteria. In case of weighting methods this information is given by means of preference weights attached to the successive criteria. If we deal with ordinal information, the weights are represented

by means of rank orders W ( j = 1 , . . . , J ) in a weight vector w: j w - (wlf...,wj)T (2.2)

Clearly, i t i s again assumed that W > wji implies that c r i t e r i j on j i s regarded as more important than j ' . Next, the regime method uses a pairwise comparison of a l l choice options, so t h a t then the mutual comparison of two choice options i s not influenced by the presence and effects of other a l t e r n a t i v e s . Of course, the eventual rank order of any two a l t e r n a t i v e s i s co-determined by remaining a l t e r n a t i v e s (cf. the independence of i r r e l e v a n t a l t e r n a t i v e s problem). In order t o explain the mechanism of the regime method, we w i l l f i r s t define the concept of a regime. Consider two a l t e r n a t i v e choice options i and i ' . If for c r i t e r i o n j a c e r t a i n choice option i i s b e t t e r than i ' ( i . e . s i ; L j - e j j - e i t j > 0 ) , i t should be noted t h a t in case of ordinal Information, the order of magnitude of s i i ' j i s n o t r e l e v a n t , but only i t s sign. Consequently, if a ' j = sign s ^ f j = +1, then a l t e r n a t i v e i i s b e t t e r than i f for c r i t e r i o n j . Otherwise, <JJ - 1 , or (in case of t i e s ) 0ii>jO. By making such a pairwise comparison for any two a l t e r n a t i v e s i and i ' for a l l c r i t e r i a j ( j = 1 , . . . , J ) , we may construct a Jxl regime vector r ^ i , defined a s :
r

i i , = ! (ii'1<Jii'j)T,

V i , i ' , i'*i

(2.3)

Thus, the regime vector contains only + and - signs (or in case of ties also 0 signs), and reflects a certain degree of (pairwise) dominance of choice option i with respect to i' for the unweighted effects for all J judgement criteria. Clearly, we have altogether 1(1-1) pairwise comparisons, and hence also 1(1-1) regime vectors. These regime vectors can be included in an Jxl(l-1) regime matrix R:

R = 1 1 r 13 ...ru r2-|. . . rx1.. .r I( z^ ) *4 .

(2.4)

1^1

P-2

It is evident that, if a certain reg: ne vector r - t would only oon^^ tain + signs, alternative i would absolutely dominate i'. Usually however a regime vector contains both + and - signs, so that then additional information in the form of the weights vector (2.2) is required.

5
In order to treat ordinal information on weights, the assumption is now made here that the ordinal weights wj (j=1,...,j) are a rank order representation of an (unknown) underlying oardinal stochastic T weight vector w*= (w*,...,w*) with max{w*}= 1, w* > 0, .. The ordi 1 -J . j j j nal ranking of the weights is thus supposed to be consistent with the quantitative information incorporated in an unknown cardinal vector w*; in other words: wj > wj i Wj* > Wjt*. Next, we assume * that the weighted dominance of choice option i with regard to i' can be represented by means of the following stochastic expression based on a weighted summation of cardinal entities (implying essentially a additive linear utility structure): J v. . .1. e . ., .w* r (2.5) -il' j-1 n'j-j If Xii' i s positive, choice option i is dominant with respect to i'. However, in our case we do not have information on the cardinal value of Wj*, but only on the ordinal value of Wj (which is assumed to be consistent with w-j*). Therefore, we introduc a certain probability p^i for the dominance of i with respect to i': Pii' P^OD (vii' > ) and define as an aggregate probability measure:
p =

(2.6)

i Vi.SiP'

(2 7)

Then it is easily seen that p _ is the average probability that j alternative i is higher valued than any other alternative. Consequently, the eventual rank order of choice options is then determined by the rank order (or the order of magnitude) of the p^'s. However, the crucial problem here is to assess p^i and p^. This implies that we have to make an assumption about the probability distribution function of both the Wj*'s and of the SijM's. In view of the ordinal nature of the Wj's, it is plausible to assume for the whole relevant area a uniform density function for the Wj*'s. The motive is that, if the ordinal weights vector w is interpreted as originating from a stochastic weight vector w*,- there is without any prior information no reason to assume that a certain numerical value of _w* has a higher probability than any other value. In other words, the weitonts vector w* can adopt with equal probability each value that is in agreement with the ordinal information implied by w. This argument is essentially based on the 'principle of insufficint reason', which also constitutes the foundation stone for the so-called Laplace criterion in case of decision-making under uncertainty (see Taha, 1976). However, if due to prior information in

co
HX C D

a H"
3

CD "4 "4 CD

/ /

o o C D
C O C O HC D M

B P>
et
H-

H-

<5 "O C U 3" C D T


3

O T H'
I-

3 O

3
C D 03

C O

ca
H-

ct

*%
et O ct

&
et
H-

O
rt / C O /O

o 3
-i C D

o o e B C D
co
C O
!-

3 P) 1

o
t

ct C O C D O ct
H-

O 3P>

B H

cr
C D

g
T H

B CL C Hu

ct

B
C D C U C O

ctH-

H'

a ^ C D
(-"
H-

<-4 -4

o
IS C D C D C O C D

p.
3

<:

C u

3 3

^ *D C
3" P >
C D

^
et

C D C O

o
et
h*

et HC D C O

o
et HC D C O

H/ et

/**

Ct

ct

a p>
P)

o T
O

Oq PJ H-

P) oq P> <i P >


3 oq

ffi T
C D O C D C O

'S cu T
HC D et C D C O
.\

co
H1

cr c

et O

l---

o
H
H

X) T

H H

pi

o PI
O

! <
ct O ct

cr C D
3 C D Oq H
H^-

c
C D
H-

cr o 3

< <
C D CL C D C U M et

C O et

e C U
et
(-

3" P>
Ct

H-

C D C O C O

/ ** / |J" / P

/ <o

cr O < C!
ffi B
0) ct

3
>

p]

C O

C
4= CD -"

cr C O

C 3 o O C O T D O CL et e t H -

* >- !- <; O
pi H M

Pi H
l-"-

co H a C t C D er oH
p> H
H

*:

<

o 3 cr c t oU PJ C cr cr cr s H- pi P) Q.
H
et
H-

cr

<
H-

*< cr a !B H-4 x> >-4 H"


C D T C D ct H" ct C D T HC D C O ct

C O C D

HH

C D
Hct

et O

co

O. C D

& o 3
ct

3 P>
CJ H-

ct C D T X)

S C D
C D P> C O

ct ct

T 3

^
et

et

Cr Oq p> H >
H C D

Cr
C D

cr

!-

co O
T
H-

co C D
CL

C D

ct ct

*-> O
3

i-"-

C H O

3 C 0) D

(- et
3*
H-

P)

H-

<<
(-

C O et -i

!-> << < B a> HB C


X C D M ct
H-

X C D CL
H"

3C D -3

PJ

C D ct C D

H C D
3
C U

!- ct
Cr
C D T O CL H3 P> M O T

< o
CL !-

H^-

C u

er
pi ct
H-

* *
Ha

a M
H-

a p)
!-

a CD cr "-(> a a a cr O P) (- M> ^ CD T T o
H H P) H C O

H-

ca
et
H-

C D

C D

3
PJ ct
H-

co e t

e
o o
3

ca
C D

H-1

- O
et I-1 O "3 HCL

>-4

o.
B

o p>
.

&

C U CT C D

o -i
3

"
ca
ct O

C D

cr co
CD

CD CO

ot C
!-<

e o. < ! < C D

P>-4

o T
3 C u
ct
!-

o- o <S
T

l- 1 -

!- 3

e o 3*
PJ et ct

H1-

O H-

o -J o cr

-i 3

CD O C D C O

B > T

O ct O

a *
Q

3
O

<<

cr

HHCJ.

CD C O
et-

<

o pi co C D
-4

tion tion

CT

Cu

O C O OJ

a
i-"-

a H
C D O
H"et

n a O tet 3(D

et H-Cr

o "3

3
ct

3
pi t->
CL

ro

fu 3

3C D
't O H H O

a C u
et
H-

O 3
s

ffi 3
ct
H-

<-t>

1 HCD ( - Ho et

cr

o Cr
et
H.

C u

co o

s a Hl-"

o h"
O C D O

C D
3 H1
i-

II O

er C D
CL H*-4 -4 C D C D

p)

3 P)

co

o 3
O

a P)

H 3*

o pi
HH-

C D

a
j=

P)
ct

C D *3

O 3
i-*

!->
C D

o
cr
C D

! <
O.

H-

s:
C D
y->-

P>
ct
H-

p>
O

co
PJ
*-4 -i CD

<
-3 et H O

o 3
et

C O

< P)
CT I-1 C D C O JC HO

ca
3

31

Q "

o et
H-

C D C O C D
V

"-4 Oq

C O

cr co C D ot e
h-

3 B O HX
H-

et C D C D C O et C D

!- 3
Oq C O O 3" C D

3"

o o P)

g ct
3C D

P >

cr ct co

S -4
HO
H-

!- -s co H e t P)
-i
H-

*
!-<

a h"'
3 P>
(-
v^>

co
T

(- C O
C u

-i

-s

co
O

C D

C O

3 3 0 B

cr C
et H-

et

o> P
3

s
O O

co c o ez
B
h*

TT

O C D C O H-

o B
o C U
-J
*-

o et
O

C D C O

o
H-

cr C D
3
et

C O

-i
C D H

O C D

P>
3 Q.

O 3
-Cr

P > 3
(U (-

& C
CD

< ! !- < ct
P)
C P)
et
H-

H 3*

C D ct T C D C O

C D *3
H-

*D C
3-

C D "3 C D O

o 3

cr

3
C D

cr

C D

T C D
(-

3
O

C O

et

S
C D

co

c o

HX C D

<; <!
co Hco

(-

-s aH
3 C u
H

X o1 "O
H-

P >
C D

P)

o 3
M-

(-" C O
C D

O 3
PJ

cr CD cr co cr Hc CD o C O et
-i
pi

O PJ H
-4 O T

P) < S
C D P)

M>

P)

aH
Ca et H-

et

o -3
C O

C D O
Ct

o 3
O
>-4

!- -i CD O
3
O

C O et

PJ

3 O.

a
H-

H 3*
C D C O C D

o O
O

a
H-

et 3ct
H-

3
C O C D O

a T)
T O

^ 1
Q

3
OH-

M -4 ct

o O
C O

3
h*

>-ti

H
t->-

o T ct M

p>

3
ct

a
3

*-*
>-4 C D T C D

C O 3C D
CL

o C D
"-4 *-4
!-

o % o 3
a co
C D CD C U C O C O CJ B C D
CL

3 3
3 CJ

co fcr
H> !-

!-

(-

O *3 1

3
Oq

C D C O

cu cr o C D Cr

O
H-

3*

C D

o >- H *. CD co

O C D C O

Cr
C D

co C D

(-
ct

H--

C D 3 et

co C o
C D

<: <

3 1

C D T 1

3.

Structure of the Standard Regime Method

The regime method was o r i g i n a l l y developed for purely q u a l i t a t i v e information on multidimensional choice problems, and aimed even at designing a ohoice evaluation method which did not need the use of a computer. However, i t turned out that in case of many c r i t e r i a (7 or more), of the presence of t i e s , or of the presence of mixed data, computer assistance was necessary. For simple examples ( i . e . , with a low number of c r i t e r i a ) , however, numerical i l l u s t r a t i o n s can without l o s s of generality d i r e c t l y be used t o explain the basic steps of the regime algorithm. In our case we assume the following choice problem. A decision-maker has to make a choice out of 3 a l t e r n a t i v e commodities (goods, plans, p r o j e c t s , e t c ) , which are characterized by 4 a t t r b u t e s (features, impacts, e t c ) , measured in a s t r i c t l y ordinal sense ( i . e . , without t i e s ) . Then we may assume the following effect matrix E: eriterion
alternative^ 1

E -

1 2 3

3 2 1

2 2 1 3

3 1 2 3

4 1 3 2

(3.1)

Furthermore, we assume the following weight vector w:


w =* (wj = ( 4 wj? W wi})"^ 3 3 "2

)T

(3.2)

which of course also implies the following consistency condition for the cardinal weights: w-| * > W2* > W3* > W4*. A pairwise comparison of the information in E leads to the following regime matrix R:

V1X \!n \JU fe h\ !-.


R = + + + + + + + + -

(3.3)

For instance, if we take regime r-|2 it is easily seen that:


y_i2=wj*+W2*-W3*- w\*

(3.4)

and: P12
=

prob ( _ ] > 0) v-2 = prob{(wi*+w?*-W3*-w4*)>0}

(3.5)

8 The question is now whether we can make any valid statement regarding the value of p-|2> In this case the previous question is easy to answer, given the information implied by (3-2), viz. w-| > W 2 > 3 > W4 > 0. Thus it can directly be derived that P12 1. If we next take the choice alternatives 1 and 3 we can easily derive v^-j3 by means of (3.3). i.e., y - 3 =w-| *-W2*-W3 * -wij * _] (3 6)

In t h i s case, a p r i o r i no unambiguous statement regarding the value of p-j3 can be made, unless we use the p r o b a b i l i t y approach outlined in the previous section. If we thus assume t h a t a l l Wj*s are uniformly d i s t r i b u t e d , we have to identify the r e l a t i v e s i z e of the four-dimensional hyperplane for which condition (3.6) holds. The r e l a t i v e s i z e of the various hyperplanes which make up the envelopes of the information embodied in the weight vector may thus be regarded as a p r o b a b i l i t y measure for the dominance of the a l t e r n a t i v e concerned. Of course, one has t o take i n t o account the standardization condition that max{w-|*, W2*, W3*, W4*}=1 and w-|*0, W2*0, W3%0, W4*S0 (see for further d e t a i l s Hinloopen, 1985, and Hinloopen and Smyth, 1985). Then by using conditions (3-2) and (3.6) in addition to the standardization condition, we can e a s i l y derive the value P13 in case of a uniform d i s t r i b u t i o n :
p 1 3 = 1/6 (3.T)

Finally, w e will compare choice options 2 and 3- Then w e have: V23=w-| *-W2*-W3 *+wi( * In t h i s case, we can e a s i l y derive t h a t : P23 = 1/2 Now we can directly derive the total dominance by means of (2.7), i.e.,
P1 = 1 / 2 ( 1 + 1/6) =7/1 2 p 2 - 1/2( + f / 2 ) = 1 / 4 P3 1/2(5/6+1Y2)=2/3

(3.8)

(3.9) of each choice option

1
final

(3.10)

Thus, in our illustrative example the following alternatives results: alternative 3 > alternative 1 > alternative 2

ranking of

9
It is evident that the foregoing example can easily be generalized in a formal notation, but as this notation is more cumbersome than illustrative we suffice to conclude here that the regime method provides a fairly direct and unambiguous solution to a strictly qualitative multiple criteria choice problem. 4. Ties and Mixed-Data in the Regime Analysis additional problems caused by the presence of for both the effects e^j and the weights WJ

In this section the ties and mixed data will be dealt with.

4.1. Ties in the effect matrix If the effect matrix contains ties (i.e., ejj = ei'j i n other words: equal rank orders of two alternatives i and i' for a specific criterion j ) , then the additional problems can easily be solved. This can easily be illustrated by including ties for the first criterion in the effect matrix (3.1):

-*^gri t er i on a l t er na'ErT'e--. 1
1 2 2 2 T

2 2 1 3

3 1 2

4 1

3
2

(4.1)

In t h i s case only the regime r-| 2 w i l l change, i . e . , r 1 2 - (0, +, - , - ) T


so that: Z.12=W2*-w_2*-v^4* On the basis of (4.3), we can easily derive uniform probability distribution - that: P12 - 1/2 (4.3) by assuming again a (4.2)

(4.4)

The existence of t i e s has c l e a r l y consequences order of plans, as in the present case we have:
P1 = 1 / 2 ( 1 / 2 + 1/6)= P2 = 1 / 2 ( 1 / 2 + 1/2)= P3 = 1 / 2 ( 5 / 6 + 1 / 2 ) = 1/3 1/2 2/3

for

the

final

rank

(4.5)

10
so that choice options 1 and 2 have changed position in the eventual rank order. Clearly, without any difficulty this procedure can be directly generalized for the existence of multiple ties. 4.2. Ties in the weights vector Ties in the weight vector (i.e., Wj - w-j') imply a different situation regarding the evaluation of values of the choice criteria. It is evident, that also the existence of ties in the weight vector will not affect the regime vector, but no doubt the probabilities P ' will alter. This will be illustrated by using again the same effect matrix (3.1). whereas the weight vector is assumed to be equal to: w. - ( 4 3 2 2 )! (4.6)

The treatment of ties will first be illustrated by comparing alternatives 1 and 2. In this case the regime matrix is still equal to (3.3), whilst V12 remains also unchanged, as can be seen from (3.4). It is clear that in this case the same result emanates, i.e., P12 - 1 (4.7)

If we next compare choice options 1 and 3, vf 3 and p-j3 do not _ change either in comparison with (3.6). However, in this case we may substitute the fact that W3* = W4* into (3.6), so that the new condition becomes: V13 w | * - W2* - 2W3* By pursuing next the same stochastic distribution, we find the result: P13 - 1/3 Finally, we will compare alternatives becomes:
v

(4.8) analysis by means of a uniform

(4.9) 2 and 3. In this case V23

23 w 1 * ~ w 2 * ~ w 3 * + w 4 * - wi - w 2 *

(4.10)

Then it is evident that P23=1, as condition (4.10) is always satisfied, given the initial condition (3-2). Consequently, we may find the following final results:

11 p-l = P2 P3 = 1/2(1 + 1/3)- 2/3 1/2(0 + 1 ) - 1/2 1/2(2/3+ 0 )= 1/3

(4.11)

so that now the f i n a l ranking of a l t e r n a t i v e s becomes: alternative 1 > alternative 2 > alternative 3 Thus, t i e s may exert a s i g n i f i c a n t impact on the eventual rank order of choice options. Finally, i t i s worth mentioning that also a s i t u a t i o n of combined t i e s in both the effect matrix and the weight vector can be handled in the same way. 4.3. Mixed data In case of mixed data in either the effect matrix or the weight vector, the regime method has to be significantly adjusted, irrespective of the presence of ties. First we will consider a situation of mixed data in the weight vector, so that part of the weights is ordinal and another part cardinal in nature. Then we impose the condition that all stochastic weights Wj* are standardized as follows: w* ir-r

X. =

(4.12)

It is easily seen that in this case Xj < 1 , while the highest . value of Xj is always equal to 1. The motive for this specific way of standardizing the weights is that (since Wj* are uniformly distributed) also the vector _ = (Xj , ... ,X_j)T X is uniformly distributed. Next, we will consider the presence of mixed data in the effect matrix. Also in this case the (stochastic qualitative) differences S.*'J are assumed to be uniformly distributed. In order to be able to compare the differences across different criteria, s*ii'j is also standardized, i.e., dilM s*., . ' - ^ max{e l j f e l f J } (4.13)

Note that d ^ i j is also uniformly distributed, either on the interval (0, 1) (if lijSeitj) or on the interval (-1, 0) (if iij^i'j)-

12
In order to compare now 2 alternatives i and i', we define - instead of Vj^ from (2.5) - a new stochastic variable z^i as follows: z.. = 1 . d.., . A . , -ii' j=1 -II'J -j while next p^r is according to (2.6) defined as: Pt = prob (zjj_i>0) (4.14)

(4.15)

The remaining part of the procedure i s then simular t o t h a t described in section 2. The various steps of the abovementioned exposition w i l l now be i l l u s t r a t e d by means of a simple numerical example. Assume the following effect matrix:

while w is assumed to be equal to: w - ( 2 1 )T (4.17)

-D

1 2

(4.16)

The r e s u l t i n g regime matrix R i s :


(4.18) If we want to compare choice option 1 with 2, we find: X12 =H 1 * ~ w 2 * . (4.19)

Consequently, we may derive: P12 = prob (yj2 > 0) = prob {(w-j* - w2*) > 0} =1

(4.20)

since we know from (4.17) that w-j > w2 and hence also w-| * > w_2*. In addition, we know that jd-j 2 1 and -dj 2 2 a r e uniformly d i s t r i b u t e d on the i n t e r v a l ( 0 , 1 ) . Note that X- =1 _ ] and X2 i s _ uniformly d i s t r i b u t e d on ( 0 , 1 ) . Therefore, we derive t h a t : i l 2 = l l 2 , 1 + l l 2 , 2 12.
By using next our Standard procedure ultimately: for (4.21) uniform densities, we find

13 p 1 2 = prob ( z 1 2 > 0) = 3/4 The proof of (4.22) runs as f o l l o w s . prob (zj2>0) = = prob{(z12>0|(d12,1>-dl2,2)J-Pr,ob(d12Ji>-d12>2) +prob{ (zj 2 >01 ( d ! 2 , 1 <~\2,2)}.prob(dj
2,1

(4.22)

<-di 2 , 2 ) (4.23)

As both _ - 2 1 and -^12,2 a r e d|, interval (0,1), we know that:

uniformly

distributed

on

the

prob(d12, l>-di2,2)=Pr,ob(di2,l<-di2,2)=1 /2 (4.23) while also the following condition holds: prob{z_12 > 0|(d12,i > -di2,2>J
=1

(4.25) 2^

Next, in order to calculate prob{z_-]2>0 | di 2<-^12 _ we define: 112,2 - -012,2/012,2 It is also easily seen that: ^12,1 < "412,2 Consequently, _3i2,l is uniformly (0,1). Next we may c a l c u l a t e 12= 112= -12.2- 3.12,1 + ^12,2 A2 = - d i 2 , 2 ( i l . 2 , 1 - A2) This i m p l i e s that: distributed

(4.26)

(^-27) on the interval

(4-28)

prob{z-|2 > 0 | ( d j 2 , 1 < _ ^ 1 2 , 2 ) J = = p r o b { - d i 2 , 2 ( 1 1 2 , 1 " i 2 ) > 0 } = 1/2 Thus, i n c o n c l u s i o n : prob(z 1 2 > 0) = 1/2 + 1/2 . 1/2 = 3/4
Q.E.D.

(4.29)

(4.30)

14 Next, we will assume that we have cardinal information on'one of the criteria, say criterion 1. Let us assume then the following effect matrix:

E=

20 8
2

1 2 2J
8

(4.31) = 0.6,(4.32)

Then d 1 2 J so t h a t :

prob(z 1 2 > 0) = prob {(0.6 + d-|2,2 *2) > * = 0 . 6 - 6 in 0.6 = 0.9

(4.33)

The proof of the latter calculation can easily be given. It is already known that A 2 and cii2,2 a r e independently _ uniformly distributed on (0,1). Then we have: prob (0.6 + d-|2,2 lZ > ) = Pi"ob(di2,2 h2> -0.6) = 1-prob(-d-|2,2 lZ > 0 > 6 ) Since we know that ~di2,2 a n d distributed on (0,1), we find: 1- prob(-^2,2 hZ
>

^2

are

independent uniformly

-6) =

} } dy dx= 1- 0.4 - 0.6 In 0.6 = 0.9 0.6 1/x 5. Concluding Remarks

The foregoing analysis has demonstrated that for discrete choice problems which are marked by complete (or partial) uncertainty in the form of ordinal (or mixed) information the dominant regime method may be an operational tooi. It leads to a probability statement regarding the choice of alternatives, and in so doing it leads usually to a unique solution (which is a major advantage compared to other 'soft' multiple criteria choice methods - like the concordance method -, which often do not lead to an unambiguous solution). Also its ability to deal with both qualitative information (including ties) and mixed information, make? it a powerful vehicle for evaluation analysis, not only in the field of public choice theory but also in the field of consumer theory and marketing analysis. Various empirical applications (e.g., housing market, transportation and physical planning) have also demonstrated its usefulness in practical choice situations.

15
References Delft, A. van, and P. Nijkamp, Multicriteria Analysis and Regional Decision-Making, Martinus Nijhoff, The Hague/Boston, 1977.

Hinloopen, 'E., P. Nijkamp and P. Rietveld, Qualitative Discrete Multiple Criteria Choice Models in Regional Planning, Regional Science and Urban Economios, vol. 13, 1983, pp. 77-102. Hinloopen, E., De Regime Methode, M.A. Thesis, Interfaculty Actuariat and Econometrics, Free University, Amsterdam, 1985 (mimeographed). Hinloopen, E., and A.W. Smyth, A Description of the Principles of a New Multicriteria Evaluation Technique, The Regime Method, in: Proceedings Colloquium Vervoersplanologisch Speurwerk 1985, Delft (1985b), pp. Keeney, R., and H. Raiffa, Decisions with Multiple Objectives, Preferences and Value Tradeoffs, Wiley, New York, 1976. Kmietowicz, Z.W., and A.D. Pearman, Decision Knowledge, Gower, Aldershot, 1981. Lancaster, K., Consumer Demand, Columbia 1971. Theory and Incomplete

University

Press, New York,

Lootsma, T.A., Saaty's Priority Theory and the Nomination of a Senior Professor in Operation Research, European Journal of Operational Research, 4, 1980, pp. 380-388. Mastenbroek, P., and J.H.P. Paelinck, Qualitative Multicriteria Analysis - Applications to Airport Location, Environment and Planning A, 9 (8), 1977, pp. 883-895. Nijkamp, P., and H. Voogd, New Multicriteria Methods for Physical Planning by means of Multidimensional Scaling Techniques, in: Haimes, Y. and Kindier, J. (eds.), Water and Related Land Resource Systems, Pergamon Press, Oxford, 1981, pp. 19-30. the Unmea-

Nijkamp, P., H. Leitner and N. Wrigley (eds.), Measuring surable, Martinus Nijhoff, Dordrecht, 1985.

Rietveld, P., Multiple Objective Decision Methods and Regional Planning , North-Holland Publ. Co., Amsterdam, 1980. Saaty, T.L., A Scaling Method for Priorities in Hierarchical Structures, Journal of Mathematical Psychology, 15, 1977, pp. 234-281. Taha, H.A., Operations Research, MacMillan, New York, 1976. Voogd, H., Multicriteria Evaluation for Urban and Regional Planning, Pion, London, 1983. .. _,. . ---

1981-1

E. Vogel v.ing

A quarterly econometrie model for the Price Formution of Coffee on the World Market DeiMdd i i Supply of Natural Rubber, Part I id

1981-18

F.C. Palm and J.M. Sneek

1981-2 1981-3

H.P. Smit R. Vos

1981-19 The political Economy of the Republic of Korea; A proposal for a model framework of an open economy in the ESCAP-region, with emphasis on the Role of the State Structural Econometrie Model ing and Time Series Analysls - Twards an Integrated Apoaoh Urban Impact Analysls in a Spatial Context: Methodologie and Case Study 1981-23 Primaire ling exporten en ekonomiese ontwikke1981-21 1981-7 D. A. Kodde Het genereren en evalueren van voorspellingen van omzet en netto winst: een toegepast kwantitatieve benadering Financile vraagstukken onder onzekerheid 1982-2 1981-9 P. van Dijck and H. Verbruggen A Con3taht-Market-Shares Analysls of ASEAN Manufactured Exports to the European Communlty 1982-3 1981-10 P. Nljkamp, H. de Graaff and E. Sigar International Conflict Analysls 1981-11 1981-12 1981-13 P. Nljkamp A.J. Mathot S.P. van Duin en P.A. Cornelis L'Utili3ation une Voiture du Crdit lors de 1'Achat d' 1982-1 A Multldlmenslonal Analysls of Regional Infrastructure and Economie Development

P. Nijkamp and P. Rietveld H. Blommestein and P. Nijkamp P. Nljkamp and P. Rietveld F. Brouwer and P. Nijkamp A. Klelnknecht

1981-20

1981-1

F.C. Palm

1981-21

1981-22 1981-5 P. Nljkamp in co-op. with H. v. Handenhoven and R. Janssen R. Ruben

1981-6

Hldde P. Smit

1982-1

Peter Nljkamp

1981-8

B. Out

J.M. Sneek

F.E. Schippers

Piet van Helsdingen maart 1982

Onderzoek naar levensomstandigheden en opvattingen over arbeid bij mensen zonder werk, deel I Dlsaggregate Context Models of Cholse In a Spatial

1982-5

Peter Nljkamp Jaap Spronk Ruerd Ruben (ed.) H.W.M. Jansen mei 1982

1981-11

W. van Lierop and P. Nljkamp Htdde P. Smit F.C. Palm

1982-6 1982-7

1981-15 1981-16

The World Vehlcle Market Structural Econometrie Model ing and Time Series Analysls: An Integrated Approaoh Linear Regression Using Both Temporally Aggregated and Temporally Disaggregated Data 1982-9

1982-8 mei

J. Klaassen and H. Schreudr

1981-17

F.C. Palm and Th.E. Nijman

F. Brouwer and P. Nijkamp