Aggregation on finite ordinal scales by scale independent functions
Jean-Luc Marichal ∗ and Radko Mesiar ∗∗
∗ University of Luxembourg, Luxembourg
∗∗ Slovak Technical University, Bratislava (Slovakia)
IPMU 2004 – p.1
What is a finite ordinal scale ?
What is a finite ordinal scale ?
Two equivalent approaches :
IPMU 2004 – p.2
What is a finite ordinal scale ?
Two equivalent approaches :
• Symbolical approach Finite chain ( S, 4 )
S = { s 1 ≺ s 2 ≺ · · · ≺ s k }
What is a finite ordinal scale ?
Two equivalent approaches :
• Symbolical approach Finite chain ( S, 4 )
S = { s 1 ≺ s 2 ≺ · · · ≺ s k }
r r r r r -
s 1 s 2 · · · s k S
IPMU 2004 – p.2
What is a finite ordinal scale ?
Two equivalent approaches :
• Symbolical approach Finite chain ( S, 4 )
S = { s 1 ≺ s 2 ≺ · · · ≺ s k }
r r r r r -
s 1 s 2 · · · s k S
Example: evaluation of a product by a consumer
What is a finite ordinal scale ?
Two equivalent approaches :
• Numerical approach
Strictly increasing sequence of real numbers (defined up to order)
IPMU 2004 – p.3
What is a finite ordinal scale ?
Two equivalent approaches :
• Numerical approach
Strictly increasing sequence of real numbers (defined up to order)
r r r r r -
4 5 7 8 10
r r r r r -
−3 1 . 5 14 . 2 58 263
rating benchmarks %
Equivalence between both definitions
IPMU 2004 – p.4
Equivalence between both definitions
∃ isomorphism f : S → E ⊆ R such that
s i 4 s j ⇔ f (s i ) 6 f (s j )
Equivalence between both definitions
∃ isomorphism f : S → E ⊆ R such that
s i 4 s j ⇔ f (s i ) 6 f (s j )
...and defined up to an automorphism φ of E : f 0 = φ ◦ f
IPMU 2004 – p.4
Equivalence between both definitions
∃ isomorphism f : S → E ⊆ R such that
s i 4 s j ⇔ f (s i ) 6 f (s j )
...and defined up to an automorphism φ of E : f 0 = φ ◦ f
r r r r r - S
s 1 s 2 s 3 s 4 s 5
¡ ¡
¡ ª
@ @
@ R
f f 0 = φ ◦ f
Generally E = R or E = ]0 , 1[
IPMU 2004 – p.5
Generally E = R or E = ]0 , 1[
What if E = [0 , 1] ?
Generally E = R or E = ]0 , 1[
What if E = [0 , 1] ? Let (S, 4 ) be a finite ordinal scale
IPMU 2004 – p.5
Generally E = R or E = ]0 , 1[
What if E = [0 , 1] ? Let (S, 4 ) be a finite ordinal scale
f ( s i ) = 0 ⇒ i = 1
Generally E = R or E = ]0 , 1[
What if E = [0 , 1] ? Let (S, 4 ) be a finite ordinal scale
f ( s i ) = 0 ⇒ i = 1 f ( s i ) = 1 ⇒ i = | S |
IPMU 2004 – p.5
Generally E = R or E = ]0 , 1[
What if E = [0 , 1] ? Let (S, 4 ) be a finite ordinal scale
f ( s i ) = 0 ⇒ i = 1 f ( s i ) = 1 ⇒ i = | S |
We assume that f : S → E is endpoint-preserving
Generally E = R or E = ]0 , 1[
What if E = [0 , 1] ? Let (S, 4 ) be a finite ordinal scale
f ( s i ) = 0 ⇒ i = 1 f ( s i ) = 1 ⇒ i = | S |
We assume that f : S → E is endpoint-preserving
Consider two independent ordinal scales ( S, 4 S ) and ( T, 4 T )
• If E = ]0 , 1[ , those scales have nothing in common
• If E = [0, 1] , those scales have fixed endpoints
IPMU 2004 – p.5
Aggregation on finite ordinal scales
Aggregation on finite ordinal scales
(a 1 , . . . , a n ) ∈ S n 7→ a ∈ S
IPMU 2004 – p.6
Aggregation on finite ordinal scales
(a 1 , . . . , a n ) ∈ S n 7→ a ∈ S
Define G : S n → S
( a 1 , . . . , a n ) ∈ S n 7→ G ( a 1 , . . . , a n ) ∈ S
Aggregation on finite ordinal scales
(a 1 , . . . , a n ) ∈ S n 7→ a ∈ S
Define G : S n → S
( a 1 , . . . , a n ) ∈ S n 7→ G ( a 1 , . . . , a n ) ∈ S
Example: n = 2 and | S | = 3
a 2 \ a 1 s 1 s 2 s 3 s 1 s 2 s 1 s 3 s 2 s 1 s 1 s 3 s 3 s 3 s 2 s 2
IPMU 2004 – p.6
Aggregation on finite ordinal scales
(a 1 , . . . , a n ) ∈ S n 7→ a ∈ S
Define G : S n → S
( a 1 , . . . , a n ) ∈ S n 7→ G ( a 1 , . . . , a n ) ∈ S
Example: n = 2 and | S | = 3
a 2 \ a 1 s 1 s 2 s 3
s 1 s 2 s 1 s 3
s 2 s 1 s 1 s 3
Aggregation on finite ordinal scales
Alternative approach
IPMU 2004 – p.7
Aggregation on finite ordinal scales
Alternative approach
Scale independent function F : E n → E
( x 1 , . . . , x n ) ∈ E n 7→ F ( x 1 , . . . , x n ) ∈ E
Aggregation on finite ordinal scales
Alternative approach
Scale independent function F : E n → E
( x 1 , . . . , x n ) ∈ E n 7→ F ( x 1 , . . . , x n ) ∈ E
Example: F = median
median(0 . 1 , 0 . 3 , 0 . 6) = 0 . 3
IPMU 2004 – p.7
Aggregation on finite ordinal scales
A ( E ) = automorphism group of E
Aggregation on finite ordinal scales
A ( E ) = automorphism group of E
Definition (Marichal and Roubens 1993)
F : E n → E is an invariant function if, for any φ ∈ A(E ) , we have
F [ φ ( x 1 ) , . . . , φ ( x n )] = φ [ F ( x 1 , . . . , x n )]
IPMU 2004 – p.8
Aggregation on finite ordinal scales
A ( E ) = automorphism group of E
Definition (Marichal and Roubens 1993)
F : E n → E is an invariant function if, for any φ ∈ A(E ) , we have
F [ φ ( x 1 ) , . . . , φ ( x n )] = φ [ F ( x 1 , . . . , x n )]
Examples:
median : yes
Aggregation on finite ordinal scales
A ( E ) = automorphism group of E
Definition (Marichal and Roubens 1993)
F : E n → E is an invariant function if, for any φ ∈ A(E ) , we have
F [ φ ( x 1 ) , . . . , φ ( x n )] = φ [ F ( x 1 , . . . , x n )]
Examples:
median : yes
arithmetic mean : no !
IPMU 2004 – p.8
Description of invariant functions
Description of invariant functions
Definition (Mesiar 2003, Bartlomiejczyk & Drewniak 2004) A subset I ⊆ E n is invariant if
x ∈ I ⇒ φ(x) ∈ I ∀ φ ∈ A(E )
IPMU 2004 – p.9
Description of invariant functions
Definition (Mesiar 2003, Bartlomiejczyk & Drewniak 2004) A subset I ⊆ E n is invariant if
x ∈ I ⇒ φ(x) ∈ I ∀ φ ∈ A(E )
I invariant is minimal if @ J invariant and J Ã I .
Description of invariant functions
Definition (Mesiar 2003, Bartlomiejczyk & Drewniak 2004) A subset I ⊆ E n is invariant if
x ∈ I ⇒ φ(x) ∈ I ∀ φ ∈ A(E ) I invariant is minimal if @ J invariant and J Ã I . Example: E = [0, 1] and n = 2
IPMU 2004 – p.9
Description of invariant functions
Definition (Mesiar 2003, Bartlomiejczyk & Drewniak 2004) A subset I ⊆ E n is invariant if
x ∈ I ⇒ φ(x) ∈ I ∀ φ ∈ A(E ) I invariant is minimal if @ J invariant and J Ã I . Example: E = [0, 1] and n = 2
6
¡ ¡ ¡ ¡
r r
x 2
Description of invariant functions
Proposition (Ovchinnikov 1998, Mesiar 2003)
F : E n → E is an invariant function if and only if, for any minimal invariant subset I ⊆ E n ,
IPMU 2004 – p.10
Description of invariant functions
Proposition (Ovchinnikov 1998, Mesiar 2003)
F : E n → E is an invariant function if and only if, for any minimal invariant subset I ⊆ E n ,
• either F | I = inf E (if inf E ∈ E )
Description of invariant functions
Proposition (Ovchinnikov 1998, Mesiar 2003)
F : E n → E is an invariant function if and only if, for any minimal invariant subset I ⊆ E n ,
• either F | I = inf E (if inf E ∈ E )
• or F | I = sup E (if sup E ∈ E )
IPMU 2004 – p.10
Description of invariant functions
Proposition (Ovchinnikov 1998, Mesiar 2003)
F : E n → E is an invariant function if and only if, for any minimal invariant subset I ⊆ E n ,
• either F | I = inf E (if inf E ∈ E )
• or F | I = sup E (if sup E ∈ E )
• or ∃ i ∈ {1 , . . . , n } such that F | I = P i (coord. proj.)
Scale independent vs. invariant functions
IPMU 2004 – p.11
Scale independent vs. invariant functions
Proposition
F : E n → E is invariant
E n −−−→ F E
Scale independent vs. invariant functions
Proposition
F : E n → E is invariant m
E n −−−→ F E
IPMU 2004 – p.11
Scale independent vs. invariant functions
Proposition
F : E n → E is invariant m
∀ ( S, 4 ) , ∃ G : S n → S ,
E n −−−→ F E
n G
Scale independent vs. invariant functions
Proposition
F : E n → E is invariant m
∀ ( S, 4 ) , ∃ G : S n → S ,
∀ f : S → E ,
E n −−−→ F E
f
x
x
f
S n −−−→ G S
IPMU 2004 – p.11
Scale independent vs. invariant functions
Proposition
F : E n → E is invariant m
∀ ( S, 4 ) , ∃ G : S n → S ,
∀ f : S → E ,
F [f (a 1 ), . . . , f (a n )] = f [G(a 1 , . . . , a n )]
E n −−−→ F E
f
x
x
f
n G
Scale independent vs. invariant functions
Proposition
F : E n → E is invariant m
∀ ( S, 4 ) , ∃ G : S n → S ,
∀ f : S → E ,
F [f (a 1 ), . . . , f (a n )] = f [G(a 1 , . . . , a n )]
E n −−−→ F E
f
x
x
f
S n −−−→ G S G represents F in ( S, 4 )
IPMU 2004 – p.11
Scale independent vs. invariant functions
Proposition
F : E n → E is invariant m
∀ ( S, 4 ) , ∃ G : S n → S ,
∀ f : S → E ,
F [f (a 1 ), . . . , f (a n )] = f [G(a 1 , . . . , a n )]
E n −−−→ F E
f
x
x
f
n G
Scale independent vs. invariant functions
Proposition
F : E n → E is invariant m
∀ ( S, 4 ) , ∃ G : S n → S ,
∀ f : S → E ,
F [f (a 1 ), . . . , f (a n )] = f [G(a 1 , . . . , a n )]
E n −−−→ F E
f
x
x
f
S n −−−→ G S G represents F in ( S, 4 )
G is uniquely determined : G ( a ) = f − 1 ( F [ f ( a )])
IPMU 2004 – p.11
Example
On R 2 :
Example
On R 2 :
- 6
¡ ¡ ¡ ¡ ¡ ¡ ¡
x 1 x 2
x 1
x 2
F (x 1 , x 2 ) = x 1 ∧ x 2
IPMU 2004 – p.12
Example
On R 2 :
- 6
¡ ¡ ¡ ¡ ¡ ¡ ¡
x 1 x 2
x 1
x 2
- 6
¡ ¡ ¡ ¡ ¡ ¡ ¡
a 1 a 2
s s s
s s s
s s s
s 1 s 2 s 3
s 1 s 2
s 3
s 1 s 1 s 1 s 1
s 1
s 2 s 2
s 2 s 3
F (x 1 , x 2 ) = x 1 ∧ x 2 G(a 1 , a 2 ) = a 1 ∧ a 2
Example
On R 2 :
- 6
¡ ¡ ¡ ¡ ¡ ¡ ¡
x 1 x 2
x 1
x 2
- 6
¡ ¡ ¡ ¡ ¡ ¡ ¡
a 1 a 2
s s s
s s s
s s s
s 1 s 2 s 3
s 1 s 2
s 3
s 1 s 1 s 1 s 1
s 1
s 2 s 2
s 2 s 3
F (x 1 , x 2 ) = x 1 ∧ x 2 G(a 1 , a 2 ) = a 1 ∧ a 2
F and G are always isomorphic !
IPMU 2004 – p.12
Important remark
Considering G : S n → S is not equivalent to considering
F : E n → E invariant
Important remark
Considering G : S n → S is not equivalent to considering F : E n → E invariant
Example: n = 2 and E = R
IPMU 2004 – p.13
Important remark
Considering G : S n → S is not equivalent to considering F : E n → E invariant
Example: n = 2 and E = R
• 4 invariant functions F : R 2 → R
Important remark
Considering G : S n → S is not equivalent to considering F : E n → E invariant
Example: n = 2 and E = R
• 4 invariant functions F : R 2 → R
F = min, max, P 1 , P 2
IPMU 2004 – p.13
Important remark
Considering G : S n → S is not equivalent to considering F : E n → E invariant
Example: n = 2 and E = R
• 4 invariant functions F : R 2 → R
F = min, max, P 1 , P 2
• | S | |S|
2discrete functions G : S 2 → S
Important remark
Considering G : S n → S is not equivalent to considering F : E n → E invariant
Example: n = 2 and E = R
• 4 invariant functions F : R 2 → R
F = min, max, P 1 , P 2
• | S | |S|
2discrete functions G : S 2 → S
Note: If F invariant then
F (x 1 , . . . , x n ) ∈ { x 1 , . . . , x n } ∪ {inf E, sup E }
IPMU 2004 – p.13
Continuous invariant functions
Continuous invariant functions
Proposition (Ovchinnikov 1998, Marichal 2002) F : E n → E is a continuous invariant function iff
IPMU 2004 – p.14
Continuous invariant functions
Proposition (Ovchinnikov 1998, Marichal 2002) F : E n → E is a continuous invariant function iff
• either F ≡ inf E (if inf E ∈ E )
Continuous invariant functions
Proposition (Ovchinnikov 1998, Marichal 2002) F : E n → E is a continuous invariant function iff
• either F ≡ inf E (if inf E ∈ E )
• or F ≡ sup E (if sup E ∈ E )
IPMU 2004 – p.14
Continuous invariant functions
Proposition (Ovchinnikov 1998, Marichal 2002) F : E n → E is a continuous invariant function iff
• either F ≡ inf E (if inf E ∈ E )
• or F ≡ sup E (if sup E ∈ E )
• or ∃ a lattice polynomial L : E n → E such that F = L
Continuous invariant functions
Proposition (Ovchinnikov 1998, Marichal 2002) F : E n → E is a continuous invariant function iff
• either F ≡ inf E (if inf E ∈ E )
• or F ≡ sup E (if sup E ∈ E )
• or ∃ a lattice polynomial L : E n → E such that F = L
Example of lattice polynomial : L ( x ) = ( x 1 ∨ x 2 ) ∧ x 3
IPMU 2004 – p.14
Continuous invariant functions
Proposition (Ovchinnikov 1998, Marichal 2002) F : E n → E is a continuous invariant function iff
• either F ≡ inf E (if inf E ∈ E )
• or F ≡ sup E (if sup E ∈ E )
• or ∃ a lattice polynomial L : E n → E such that F = L
Example of lattice polynomial : L ( x ) = ( x 1 ∨ x 2 ) ∧ x 3
Smooth discrete functions
IPMU 2004 – p.15
Smooth discrete functions
Definition (Godo and Sierra 1988)
G : S n → S is smooth if,
Smooth discrete functions
Definition (Godo and Sierra 1988) G : S n → S is smooth if,
∀ a, b ∈ S n , with a = b except on i th coordinate
IPMU 2004 – p.15
Smooth discrete functions
Definition (Godo and Sierra 1988) G : S n → S is smooth if,
∀ a, b ∈ S n , with a = b except on i th coordinate
where |ind( a i ) − ind( b i )| = 1
Smooth discrete functions
Definition (Godo and Sierra 1988) G : S n → S is smooth if,
∀ a, b ∈ S n , with a = b except on i th coordinate where |ind( a i ) − ind( b i )| = 1
⇒ ¯ ¯ ¯ ind[G(a)] − ind[G(b)] ¯ ¯ ¯ 6 1
IPMU 2004 – p.15
Smooth discrete functions
Definition (Godo and Sierra 1988) G : S n → S is smooth if,
∀ a, b ∈ S n , with a = b except on i th coordinate where |ind( a i ) − ind( b i )| = 1
⇒ ¯ ¯ ¯ ind[G(a)] − ind[G(b)] ¯ ¯ ¯ 6 1
Example: n = 2 and | S | = 3
a 2 \ a 1 s 1 s 2 s 3 s 1 s 2 s 1 s 3
s s s s
Smooth discrete functions
Definition (Godo and Sierra 1988) G : S n → S is smooth if,
∀ a, b ∈ S n , with a = b except on i th coordinate where |ind( a i ) − ind( b i )| = 1
⇒ ¯ ¯ ¯ ind[G(a)] − ind[G(b)] ¯ ¯ ¯ 6 1
Example: n = 2 and | S | = 3
a 2 \ a 1 s 1 s 2 s 3 s 1 s 2 s 1 s 3 s 2 s 1 s 1 s 3
s 3 s 3 s 2 s 2 Not smooth !
IPMU 2004 – p.15
Smooth discrete functions
Proposition
An invariant function is continuous iff it is represented
only by smooth discrete functions
A more general case
Two finite ordinal scales (S, 4 S ) and (T, 4 T ) ( a 1 , . . . , a n ) ∈ S n 7→ a ∈ T
IPMU 2004 – p.17
A more general case
Two finite ordinal scales (S, 4 S ) and (T, 4 T ) ( a 1 , . . . , a n ) ∈ S n 7→ a ∈ T
Define G : S n → T
( a 1 , . . . , a n ) ∈ S n 7→ G ( a 1 , . . . , a n ) ∈ T
A more general case
Two finite ordinal scales (S, 4 S ) and (T, 4 T ) ( a 1 , . . . , a n ) ∈ S n 7→ a ∈ T
Define G : S n → T
( a 1 , . . . , a n ) ∈ S n 7→ G ( a 1 , . . . , a n ) ∈ T
Example: n = 2 , | S | = 3 , and | T | = 5 a 2 \ a 1 s 1 s 2 s 3
s 1 t 4 t 1 t 2 s 2 t 4 t 3 t 5
s 3 t 2 t 1 t 1
IPMU 2004 – p.17
A more general case
Alternative approach:
Scale independent function F : E n → R
A more general case
Alternative approach:
Scale independent function F : E n → R Definition
F : E n → R is weakly invariant if, for any φ ∈ A(E ) , there is a strictly increasing ψ φ : R → R such that
F [ φ ( x 1 ) , . . . , φ ( x n )] = ψ φ [ F ( x 1 , . . . , x n )]
IPMU 2004 – p.18
Description of weakly invariant functions
Description of weakly invariant functions
Proposition
F : E n → R is weakly invariant if and only if, for any minimal invariant subset I ⊆ E n , there exist
IPMU 2004 – p.19
Description of weakly invariant functions
Proposition
F : E n → R is weakly invariant if and only if, for any minimal invariant subset I ⊆ E n , there exist
i) i ∈ {1, . . . , n }
Description of weakly invariant functions
Proposition
F : E n → R is weakly invariant if and only if, for any minimal invariant subset I ⊆ E n , there exist
i) i ∈ {1, . . . , n }
ii ) g : R → R, constant or strictly monotonic
IPMU 2004 – p.19
Description of weakly invariant functions
Proposition
F : E n → R is weakly invariant if and only if, for any minimal invariant subset I ⊆ E n , there exist
i) i ∈ {1, . . . , n }
ii ) g : R → R, constant or strictly monotonic such that
F | I = g ◦ P i
Description of weakly invariant functions
Proposition
F : E n → R is weakly invariant if and only if, for any minimal invariant subset I ⊆ E n , there exist
i) i ∈ {1, . . . , n }
ii ) g : R → R, constant or strictly monotonic such that
F | I = g ◦ P i (+ some extra conditions)
IPMU 2004 – p.19
Scale independent vs. weakly invariant functions
Scale independent vs. weakly invariant functions
Proposition
F : E n → R is weakly invariant
E n −−−→ F R
IPMU 2004 – p.20
Scale independent vs. weakly invariant functions
Proposition
F : E n → R is weakly invariant m
E n −−−→ F R
Scale independent vs. weakly invariant functions
Proposition
F : E n → R is weakly invariant m
∀ ( S, 4 ) , ∃ ( T, 4 ) & G : S n → T ,
E n −−−→ F R
S n −−−→ G T
IPMU 2004 – p.20
Scale independent vs. weakly invariant functions
Proposition
F : E n → R is weakly invariant m
∀ ( S, 4 ) , ∃ ( T, 4 ) & G : S n → T ,
∀ f : S → E , ∃ g f : T → R ,
E n −−−→ F R
f
x
x
g
fG
Scale independent vs. weakly invariant functions
Proposition
F : E n → R is weakly invariant m
∀ ( S, 4 ) , ∃ ( T, 4 ) & G : S n → T ,
∀ f : S → E , ∃ g f : T → R ,
F [ f ( a 1 ) , . . . , f ( a n )] = g f [ G ( a 1 , . . . , a n )]
E n −−−→ F R
f
x
x
g
fS n −−−→ G T
IPMU 2004 – p.20
Scale independent vs. weakly invariant functions
Proposition
F : E n → R is weakly invariant m
∀ ( S, 4 ) , ∃ ( T, 4 ) & G : S n → T ,
∀ f : S → E , ∃ g f : T → R ,
F [ f ( a 1 ) , . . . , f ( a n )] = g f [ G ( a 1 , . . . , a n )]
E n −−−→ F R
f
x
x
g
fG
Scale independent vs. weakly invariant functions
Proposition
F : E n → R is weakly invariant m
∀ ( S, 4 ) , ∃ ( T, 4 ) & G : S n → T ,
∀ f : S → E , ∃ g f : T → R ,
F [ f ( a 1 ) , . . . , f ( a n )] = g f [ G ( a 1 , . . . , a n )]
E n −−−→ F R
f
x
x
g
fS n −−−→ G T G represents F in (S, 4 )
G and g f are uniquely determined
IPMU 2004 – p.20
Continuous weakly invariant functions
Continuous weakly invariant functions
Proposition (Yanovskaya 1989, Marichal 2002)
F : E n → R is continuous and weakly invariant if and only if there exist
IPMU 2004 – p.21
Continuous weakly invariant functions
Proposition (Yanovskaya 1989, Marichal 2002)
F : E n → R is continuous and weakly invariant if and only if there exist
i ) L : E n → E , lattice polynomial
Continuous weakly invariant functions
Proposition (Yanovskaya 1989, Marichal 2002)
F : E n → R is continuous and weakly invariant if and only if there exist
i ) L : E n → E , lattice polynomial ii) g : E → R, constant
or continuous and strictly monotonic
IPMU 2004 – p.21
Continuous weakly invariant functions
Proposition (Yanovskaya 1989, Marichal 2002)
F : E n → R is continuous and weakly invariant if and only if there exist
i ) L : E n → E , lattice polynomial ii) g : E → R, constant
or continuous and strictly monotonic such that
F = g ◦ L
Continuous weakly invariant functions
Proposition
A continuous weakly invariant function is represented only by smooth discrete functions
IPMU 2004 – p.22