Multiple objectives an example

Consider the problem of deciding upon treatment for a patient

with esophageal cancer. The problem can be (schematically)
modelled as follows:
L.E.
Q.L.
X

endo
prosthesis

Multi-attribute utility theory

(very superficial in textbook!)

surgery

X
X

X
radio
therapy
chemo
therapy

X
X

X
X
X

X
X

When a decision problem concerns multiple objectives,

captured by multiple attributes, consequences are no longer
simple; this is apparent from the above consequence matrix.
164 / 401

Multiple objectives another example

165 / 401

The multiattribute utility function assessment

The City of Utrecht is considering four different sites (A,B,C and

D) for a new electric power generating station.
The objectives of the city are to
minimise the cost of building the station;
minimise the acres of land damaged by building it.

Let X1 , . . . , Xn , n 2, be a set of attributes associated with the

consequences of a decision problem.
The utility of a consequence (x1 , . . . , xn ) can be determined from
direct assessment: estimate the combined utility

Factors influencing the objectives include the land type at the

different sites, the architect and construction company hired, the
cost of material and machines used, the weather, etc.
Costs, however, are estimated to fall between e 15 million and
e 60 million; between 200 and 600 acres of land will be
damaged.
The possible consequences of the decision alternatives are
captured by two attributes. We therefore need to determine a
two-attribute utility function: u(Cost, Acres) (or u(C, A)).
166 / 401

u(x1 , . . . , xn ) over the given values of all n attributes;

decomposed assessment:
1 estimate n conditional utilities ui (xi ) for the given values of
the n attributes;
2 compute u(x1 , . . . , xn ) by combining the ui (xi ) of all
attributes:
u(x1 , . . . , xn ) = f [u1 (x1 ), . . . , un (xn )]

167 / 401

Direct assessment an example

Notation
Let X be an attribute and Y a set of n 1, n > 0, attributes.
Compare the following:

u(X, Y )

UX (X)
(possibly rescaled)

600
15

Acres
Cost

60 200

Assign a utility of 0 to the worst consequence (60, 600), and a

utility of 1 to the best consequence (15, 200).

(15, 200)

(50, 300)

(1 p)

The utility of, for example, consequence (50, 300) can be

determined from:
p
1.0

if n = 1 then u(X) is a utility function for X in a onedimensional decision problem

(60, 600)

To find a good representation through direct assessment,

utilities must be assessed for a substantial number of points. 168 / 401

u(X, yi )
if n > 1 then
u(X, Y ) is an n-attribute utility function;
u(X, yi ) is a subutility function for X given a fixed
value-assignment yi to attributes Y ;
uX (X) is a conditional utility function for X: a (possibly)
re-scaled function u(X, yk ) for some no longer explicit
yk .
169 / 401

Different functional forms

An example

Let X and Y be two attributes (generalisation to n > 2 attributes is

straightforward). Consider the utility function u(X, Y ) for X and Y .

The City of Utrecht decides to model the utility function

u(Cost, Acres) as an additive function.
Using standard assessment techniques, they find for the
conditional utility functions uC and uA that

u has an additive form if for constants kX and kY

u(X, Y ) = kX uX (X) + kY uY (Y )
u has a multilinear form if for constants kX , kY , and kXY

u(X, Y ) = kX uX (X) + kY uY (Y ) + kXY uX (X) uY (Y )

u has a multiplicative form if for constants kX , kY , cX , and cY

u(X, Y ) = (kX uX (X) + cX ) (kY uY (Y ) + cY )

uC (15) = 1.0 uC (30) = 0.5 uC (50) = 0.2 uC (60) = 0.0

uA (200) = 1.0 uA (300) = 0.8 uA (400) = 0.5 uA (600) = 0.0
The City finds Cost three times as important as Acres lost. The
resulting utilities for a number of Cost-Acres pairs are then:
u(50, 300)
u(30, 400)
u(60, 200)
u(15, 600)
u(15, 200)

The constants are often called weights or scaling constants.

The different functional forms are only valid under certain
assumptions; the values of the scaling constants are thereby
often constrained!
170 / 401

=
=
=
=
=

3 uC (50) + 1 uA (300) = 3 0.2 + 1 0.8 = 1.4

1.5 + 0.5 = 2.0
0.0 + 1.0 = 1.0
3.0 + 0.0 = 3.0
3.0 + 1.0 = 4.0
171 / 401

Pricing out an example

Interpreting weights or scaling constants

Reconsider the City of Utrechts two attribute utility function
u(C, A) = 3 uC (C) + 1 uA (A). Suppose that the City has
explicitly expressed the indifference (50, 600) (60, 350), which
indeed holds given all current functions:

Suppose the following utilities are assessed by the City of

Utrecht for Cost and Acres lost:
uC (15) = 1.0 uC (30) = 0.7 uC (50) = 0.4 uC (60) = 0.0
uA (200) = 1.0 uA (300) = 0.8 uA (400) = 0.5 uA (600) = 0.0

u(50, 600) = 3 uC (50) + uA (600) = 3 0.2 + 0 = 0.6

u(60, 350) = 3 uC (60) + uA (350) = 3 0 + 0.6 = 0.6

The City decides that it is willing to sacrifice 200 acres of land if

that would save e 10 million. This implies that, for example,

Now, however, the City finds out that all alternatives result in at
least a loss of 350 acres of land, and we rescale uC such that
uC (350) = 1. The expressed indifference still holds, implying that
u(60, 350) = kC 0 + kA 1 = u(50, 600) = kC 0.2 + kA 0
From this it follows that kC = 5 kA .
Did C just turn from three times as important as A to five times
as important?

u(40, 200) = u(30, 400)

that is,

a uC (40) + b uA (200) = a uC (30) + b uA (400)

We conclude that
uA (400) uA (200)
0.5 1.0
0.5
a
=
=
=
b
uC (40) uC (30)
0.4 0.7
0.3
And thus find that u(C, A) = 0.5 uC (C) + 0.3 uA (A)
Is this valid?

172 / 401

Assessing scaling constants (I)

173 / 401

Assessing scaling constants (II)

Pricing out:
Assess the marginal rate of substitution, that is, determine the
value of one objective in terms of another.
Let X and Y be two attributes with values x1  . . .  xn , n 2,
and y1  . . .  ym , m 2. Let eX and eY be the units of
measurement for the two attributes, respectively.
Suppose you are willing to sacrifice s units of Y for 1 unit of X.
Then for all xi , yj : u(xi , yj ) = u(xi + eX , yj s eY )
If u(X, Y ) is additive and uX (X), uY (Y ) are linear functions,
then this implies that
a uX (xi ) + b uY (yj ) = a uX (xi + eX ) + b uY (yj s eY )
As a result,
a
uY (yj s eY ) uY (yj )
=
b
uX (xi ) uX (xi + eX )
174 / 401

Swing weighting:
Consider a set of n 2 attributes X1 , . . . , Xn . Swing weighting
assigns weights to attributes based on either a rank-order or a
rating on attributes and one consequence.
Take the (theoretically) worst possible consequence as
benchmark. Then apply the following procedure:
rate(benchmark) 0 ;
rank(benchmark) n + 1 ;
for i = 1 to n
do Z answer to:if you could swing one attribute from worst
to best value, which would you swing?;
Swing(Z);
rank(Z) i.
if i = 1 then rate(Z) 100
else rate(Z) answer h0, 100i.
175 / 401

Swing weighting an example

Swing weighting cntd

Consider a set of n 2 attributes X1 , . . . , Xn . Let rank(Xi ) and
rate(Xi ) denote a ranking and a rating for attribute Xi ,
respectively.
The weight w(Xi ) for attribute Xi can now be determined using
either of the following two approaches:
rate(Xi )
direct rating: w(Xi ) = Pn
j=1 rate(Xj )
rank-sum weighing:
w(Xi ) =

n rank(Xi ) + 1
reversed rank(Xi )
=
number of ranks
n+1

Attribute
to swing
(benchmark)
Cost
Acres
total:

consequence
mill. e acres
60
600
15
600
60
200

direct rating:

rank rate
2+1
0
1 100
2
30
3 130

rank-sum weighting:

w(C) =

100
= 0.77
130

w(C) =

21+1
2
=
3
3

w(A) =

30
= 0.23
130

w(A) =

22+1
1
=
3
3

176 / 401

177 / 401

Lottery weights: an example

Assessing scaling constants (III)
Assume once more that the City of Utrecht decides to model the
utility function u(Cost, Acres) as an additive function:

Lottery weights:
Consider two attributes X and Y (the following extends
straightforwardly to n > 2 attributes).

u(C, A) = kC uC (C) + kA uA (A)

Let (x0 , y 0 ) denote the worst possible consequence, and

(x+ , y + ) the best possible consequence.

The weight kC is assessed from:

1.0

The weight w(X) for attribute X equals p, where p is the

indifference probability that follows from:
p
1.0

(x+ , y 0 )

(15, 600)

kC

(1 kC ) (60, 600)

(x , y )

The weight kA is assessed using:

(1 p)

(15, 200)

1.0

(x0 , y 0 )

(60, 200)

kA

(15, 200)

(1 kA ) (60, 600)

178 / 401

179 / 401

Additive independence the formal definition

Use of an additive utility function is justified given the
assumption of additive independence [AI].

MAUT with n = 2 attributes:

when can we use additive and multilinear forms?

Two attributes X and Y are additive independent if preferences

for lotteries over X Y can be established by comparing the
values one attribute at a time. More formally,
D EFINITION

Two attributes X and Y are additive independent if the paired

preference comparison of any two lotteries, defined by two joint
probability distributions on X Y , depends only on their
marginal distributions.

180 / 401

Interpreting additive independence

181 / 401

Additive independence a practical definition

The ability to establish preferences for lotteries over X Y by

comparing the values one attribute at a time entails the
following:
the decisionmaker should be indifferent between

[p, (x1 , y1 ); (1 p), (x2 , y2 )] and [p, (x1 , y2 ); (1 p), (x2 , y1 )]

since both have the same probability of achieving x1 vs x2 ;

D EFINITION

Consider two attributes X and Y with values x1 , . . . , xn , n 2,

and y1 , . . . , ym , m 2.
X and Y are additive independent iff for an arbitrary pair (xj , yl ),
we have for all pairs (xi , yk ) that
0.5

0.5

(xi , yk )

the decisionmaker should also be indifferent between

(xi , yl )

[p, (x1 , y1 ); (1 p), (x2 , y2 )] and [p, (x2 , y1 ); (1 p), (x1 , y2 )]

0.5

(xj , yl )

0.5

(xj , yk )

since both have the same probability of achieving y1 vs y2 ;

Note that additive independence is a symmetric property.

this can only hold if p = 1 p = 0.5

182 / 401

183 / 401

Additive independence an example

Additive independence the implication for u(X, Y )

Let u(X, Y ) be a two-attribute utility function defined as follows:

u(x0 , y0 ) = 1.0 u(x0 , y1 ) = 0.7 u(x0 , y2 ) = 0.5
u(x1 , y0 ) = 0.9 u(x1 , y1 ) = 0.6 u(x1 , y2 ) = 0.4
u(x2 , y0 ) = 0.5 u(x2 , y1 ) = 0.2 u(x2 , y2 ) = 0.0

Consider two attributes X and Y with values x1 , . . . , xn , n 2,

and y1 , . . . , ym , m 2.
If X and Y are additive independent, then for any pair (xj , yl ),
and all pairs (xi , yk ), we have by definition that

X and Y are additive independent:

0.5 u(xi , yk ) + 0.5 u(xj , yl ) = 0.5 u(xi , yl ) + 0.5 u(xj , yk )

[0.5, (x1 , y1 ); 0.5, (x0 , y0 )] [0.5, (x1 , y0 ); 0.5, (x0 , y1 )]

that is, the change in utility for values of one attribute is

independent of the values of the other attribute:

[0.5, (x2 , y1 ); 0.5, (x0 , y0 )] [0.5, (x2 , y0 ); 0.5, (x0 , y1 )]

u(xi , yk ) u(xj , yk ) = u(xi , yl ) u(xj , yl )

[0.5, (x1 , y2 ); 0.5, (x0 , y0 )] [0.5, (x1 , y0 ); 0.5, (x0 , y2 )]

[0.5, (x2 , y2 ); 0.5, (x0 , y0 )] [0.5, (x2 , y0 ); 0.5, (x0 , y2 )]

This means that all subutility functions u(X, yk ) are the same, up
to translation: u(X, yk ) = u(X, yl ) + ckl for some constant ckl .

...
184 / 401

185 / 401

Exploiting additive independence an example

To establish u(C, A), the City of Utrecht assesses

Exploiting Additive independence

Consider two attributes X and Y with values x1 , . . . , xn , n 2,
and y1 , . . . , ym , m 2.
If X and Y are additive independent, then three points u(xj , yl ),
u(xj , yk ), and u(xi , yl ) serve for establishing a forth: u(xi , yk )
u(X, Y ) can be constructed entirely from two intersecting subutility functions:
1. u(X, yk ) for some value yk of Y
2. u(xi , Y ) for some value xi of X

u(C, 600) for fixed acres:

u(15, 600) = 0.75 u(50, 600) = 0.15

u(30, 600) = 0.35 u(60, 600) = 0.00
and u(60, A) for fixed cost:

u(60, 200) = 0.25 u(60, 400) = 0.15

u(60, 300) = 0.20 u(60, 600) = 0.00
These two functions intersect at u(60, 600) = 0.
Assuming additive independence, we have that for all ci and aj :
u(ci , aj ) + u(60, 600) = u(ci , 600) + u(60, aj )
All other points can now be computed. For example,
u(15, 200) = u(15, 600) + u(60, 200) 0 = 1.0

186 / 401

187 / 401

The additive utility function Ia

T HEOREM

The additive utility function IIa

Let X and Y be two attributes with values x1  . . .  xn , n 2,

and y1  . . .  ym , m 2. Attributes X and Y are additive
independent iff the two-attribute utility function is additive:

L EMMA

Let X and Y be two attributes with values x1  . . .  xn , n 2,

and y1  . . .  ym , m 2.

u(X, Y ) = kX uX (X) + kY uY (Y ),

Attributes X and Y are additive independent iff the two-attribute

utility function is additive:

where
u(X, Y ) is normalised with u(x1 , y1 ) = 0 and u(xn , ym ) = 1;
uX (X) is a conditional utility function on X, normalised by
uX (x1 ) = 0 and uX (xn ) = 1;
uY (Y ) is a conditional utility function on Y , normalised by
uY (y1 ) = 0 and uY (ym ) = 1;
kX = u(xn , y1 ) and kY = u(x1 , ym ) are positive scaling
constants, summing to 1.

u(X, Y ) = u(X, y1 ) + u(x1 , Y ),

where u(x1 , y1 ) = 0.

188 / 401

189 / 401

The additive utility function IIb

Proof of Lemma:
() Let xi and yk be arbitrary values of X resp. Y .
Additive independence implies

The additive utility function Ib

Proof of the Theorem:
() From the previous lemma we have that for arbitrary values
xi and yk of X resp. Y

u(x1 , y1 ) + u(xi , yk ) = u(x1 , yk ) + u(xi , y1 )

Setting u(x1 , y1 ) = 0, we get u(xi , yk ) = u(x1 , yk ) + u(xi , y1 ).
() Suppose u(X, Y ) = u(X, y1 ) + u(x1 , Y ), and let xj and yl be
arbitrary values of X resp. Y .
Then for all values xi and yk of X resp. Y , we have that
u(xi , yk ) + u(xj , yl ) = u(xi , y1 ) + u(x1 , yk ) + u(xj , y1 ) + u(x1 , yl )
= u(xi , y1 ) + u(x1 , yl ) + u(xj , y1 ) + u(x1 , yk )

Normalising u(x1 , Y ) gives us u(x1 , Y ) = kY uY (Y ) with

kY = u(x1 , ym );
similarly, u(X, y1 ) = kX uX (X) with kX = u(xn , y1 ).
() Analogous to proof of lemma.

= u(xi , yl ) + u(xj , yk )
We conclude that X and Y are additive independent.

u(xi , yk ) = u(x1 , yk ) + u(xi , y1 ) if u(x1 , y1 ) = 0


190 / 401

191 / 401

Assessing the additive utility function an example

Utility independence the formal definition

Suppose the City of Utrecht assesses the following conditional

utilities for Cost and Acres lost:
uC (15) = 1.0 uA (200) = 1.0
uC (30) = 0.5 uA (300) = 0.8
uC (50) = 0.2 uA (400) = 0.5
uC (60) = 0.0 uA (600) = 0.0
For scaling constants kA and kC we know that kA = 1 kC and
that kC = u(15, 600). This latter utility is assessed using the
following lottery:
kC
(15, 200)

1.0

(15, 600)

Use of a multilinear or multiplicative utility function is justified

under (mutual) utility independence [(M)UI].
Attribute X is utility independent of attribute Y if conditional
preferences for lotteries over X given a fixed value for Y do not
depend on the particular value for Y . More formally,
D EFINITION

An attribute X is utility independent of an attribute Y iff for any

lotteries [(X, yk )]1 and [(X, yk )]2 over X Y with Y fixed to value
yk , we have

[(X, yk )]1  [(X, yk )]2 = [(X, yl )]1  [(X, yl )]2 y of Y

(1 kC ) (60, 600)

The indifference probability kC is found to be 0.75. We therefore

conclude that
u(C, A) = 0.75 uC (C) + 0.25 uA (A)

NB: [(X, y)] represents a conditional lottery over X Y

involving consequences over different values of X combined
with a fixed value for Y .

192 / 401

193 / 401

Utility independence a practical definition

Interpreting utility independence

Independence of conditional preferences for lotteries over X of
the value of Y , entails the following:
if [p, (x1 , y1 ); (1 p), (x2 , y1 )]  [p, (x3 , y1 ); (1 p), (x4 , y1 )]

then the decision maker should also prefer

[p, (x1 , y2 ); (1 p), (x2 , y2 )] over [p, (x3 , y2 ); (1 p), (x4 , y2 )]
since there is only a change in sure outcome of attribute Y
(y1 vs y2 ), which should not affect preferences among
lotteries over X;
similarly, if (xC , y1 ) is the certainty equivalent of the first
lottery, then (xC , y2 ) should be the certainty equivalent of the
third lottery;
Note that the above means that also the preference order on
values of X should be independent of the value of Y !
194 / 401

D EFINITION

Consider two attributes X and Y with values x1 , . . . , xn , n 2,

and y1 , . . . , ym , m 2.
X is utility independent of Y iff for an arbitrary triple xi , xj , xk
with xi  xj  xk , there exists a probability p such that for all
values yl of Y , we have that
p

1.0

(xj , yl )

(xi , yl )

(1 p)

(xk , yl )

If X is utility independent of Y and Y is utility independent of X

then X and Y are mutually utility independent.
Note: the above definition, in terms of a probability equivalent,
can also be rephrased in terms of a certainty equivalent for a
50-50 lottery.
195 / 401

An example

Utility independence the implication for u(X, Y )

Let u(X, Y ) be a two-attribute utility function defined as follows:

u(x0 , y0 ) = 1.0 u(x0 , y1 ) = 1.0 u(x0 , y2 ) = 0.7
u(x1 , y0 ) = 0.6 u(x1 , y1 ) = 0.8 u(x1 , y2 ) = 0.5
u(x2 , y0 ) = 0.0 u(x2 , y1 ) = 0.5 u(x2 , y2 ) = 0.2

Consider two attributes X and Y with values x1 , . . . , xn , n 2,

and y1 , . . . , ym , m 2.
First we observe that x1  xj  xn for each value xj of X.
Now, if X is utility independent of Y , then we have by definition
that for any xj there exists a pj such that for all yl ,

Then we have:
X is utility independent of Y :
[1.0, (x1 , y0 )] [0.6, (x0 , y0 ); 0.4, (x2 , y0 )]
[1.0, (x1 , y1 )] [0.6, (x0 , y1 ); 0.4, (x2 , y1 )]
[1.0, (x1 , y2 )] [0.6, (x0 , y2 ); 0.4, (x2 , y2 )]

u(xj , yl ) = pj u(x1 , yl ) + (1 pj ) u(xn , yl )

that is, pj =

Y is not utility independent of X, as in the context of x1 we

have y1 y0 y2 , and in the context of x2 we have
y1 y2 y0 !

u(xj , yl ) u(xn , yl )
is a linear function of u(X, yl ).
u(x1 , yl ) u(xn , yl )

This means that all subutility functions u(X, yk ) are the same, up
to (positive) scaling and translation:
u(X, yk ) = ckl u(X, yl ) + dkl for some constants ckl > 0 and dkl .

196 / 401

197 / 401

Proof of the Proposition (sketch):

() First observe that for all x, x1  x  xn . Utility
independence holds iff for each x a probability p exists such that
(x, Y ) [p, (x1 , Y )]; (1 p), (xn , Y ) and therefore (main theorem)

Utility independence an equivalence

(I)

P ROPOSITION

Consider two attributes X and Y with values x1 , . . . , xn , n 2,

and y1 , . . . , ym , m 2.
X is utility independent of Y , iff for each value yl of Y , there
exist real functions gl > 0 and hl , such that

1
2

u(X, Y ) = p u(x1 , Y ) + (1 p) u(xn , Y )



= p u(x1 , Y ) u(xn , Y ) + u(xn , Y )

solve p from (I) with Y set to yl ;

substitute this result in (I) to get the desired result.

() Let gl > 0 and hl be such that for arbitrary yl :

u(X, Y ) = gl (Y ) u(X, yl ) + hl (Y )

(II)

for all values of X and Y .

2
3
4
198 / 401

u(X, Y ) = gl (Y ) u(X, yl ) + hl (Y )

for arbitrary xj and yj , choose xi and xk s.t.

(xi , yj )  (xj , yj )  (xk , yj );
continuity: p : u(xj , yj ) = p u(xi , yj ) + (1 p) u(xk , yj );
apply (II) to u(xi , yj ), u(xj , yj ) and u(xk , yj ).
rewrite to find the desired result.


199 / 401

Exploiting utility independence

Consider two attributes X and Y with values x1 , . . . , xn , n 2,
and y1 , . . . , ym , m 2.

Three subutility functions

T HEOREM

If X is utility independent of Y , then one subutility function

u(X, yk ) and two points u(xi , yl ), and u(xj , yl ) serve for
establishing a second function: u(X, yl )

Let X and Y be two attributes with values x1  . . .  xn , n 2,

and y1  . . .  ym , m 2.
If X is utility independent of Y then

u(X, Y ) can be constructed entirely from three subutility functions:

1. u(X, yk ) for some value yk of Y
2. u(xi , Y ) for some value xi of X
3. u(xj , Y ) for some value xj of X,
xj 6= xi

u(X, Y ) = u(x1 , Y ) [1 u(X, y1 )] + u(xn , Y ) u(X, y1 )

where u(X, Y ) is normalised by u(x1 , y1 ) = 0 and u(xn , y1 ) = 1.

201 / 401

200 / 401

An Example
Suppose the City of Utrecht assesses the following utilities for
Cost and Acres lost, given A fixed at 600, and C fixed at 15 and
60, respectively:

Three subutility functions

Proof of the Theorem (sketch):

u(15, 600) = 0.75

u(30, 600) = 0.50
u(50, 600) = 0.10
u(60, 600) = 0.00

Utility independence implies the existence of functions g > 0, h

such that
(I)

u(X, Y ) = gl (Y ) u(X, yl ) + hl (Y ) yl

u(15, 200) = 1.00

u(15, 300) = 0.90
u(15, 400) = 0.80
u(15, 600) = 0.75

u(60, 200) = 0.20

u(60, 300) = 0.15
u(60, 400) = 0.10
u(60, 600) = 0.00

Cost is utility independent of Acres lost. We normalise u(C, A)

such that u(60, 600) = 0 and u(15, 600) = 1:
1
2
3
4

set yl to y1 ;
solve (I) for x1 to get h1 (Y ), using u(x1 , y1 ) = 0;
solve (I) for xn to get g1 (Y ), using u(xn , y1 ) = 1;
substitute these results in (I) to get the desired result.

u(15, 600) = 1.00

u(30, 600) = 0.67
u(50, 600) = 0.13
u(60, 600) = 0.00

u(15, 200) = 1.33

u(15, 300) = 1.20
u(15, 400) = 1.07
u(15, 600) = 1.00

u(60, 200) = 0.27

u(60, 300) = 0.20
u(60, 400) = 0.13
u(60, 600) = 0.00

Now,
u(C, A) = u(60, A) [1 u(C, 600)] + u(15, A) u(C, 600)

202 / 401

203 / 401

The additive utility function IIIa

The additive utility function IIIb

C OROLLARY

Proof of the Corollary:

Let X and Y be two attributes with values x1  . . .  xn , n 2,

and y1  . . .  ym , m 2.

Utility independence implies that

u(X, Y ) = u(x1 , Y ) [1 u(X, y1 )] + u(xn , Y ) u(X, y1 )

If X is utility independent of Y , then

with u(x1 , y1 ) = 0 and u(xn , y1 ) = 1.

the two-attribute utility function is additive iff

The lottery equivalence translates into

[0.5, (x1 , y1 ); 0.5, (xn , Y )] [0.5, (x1 , Y ); 0.5, (xn , y1 )]

0 + u(xn , Y ) = 1 + u(x1 , Y )
Under the above conditions, we have that

Substitute u(xn , Y ) with 1 + u(x1 , Y ) in the above results in

u(X, Y ) = u(x1 , Y ) + u(X, y1 )

u(X, Y ) = u(x1 , Y ) + u(X, y1 )

with u(x1 , y1 ) = 0 and u(xn , y1 ) = 1

where u(x1 , y1 ) = 0 and u(xn , y1 ) = 1.

204 / 401


205 / 401

The multilinear utility function Ia

The multilinear utility function Ia

C OROLLARY

Let X and Y be two attributes with values x1  . . .  xn , n 2,

and y1  . . .  ym , m 2.

Proof of the Corollary:

If X is utility independent of Y , then

Utility independence implies that

u(X, Y ) = u(x1 , Y ) [1 u(X, y1 )] + u(xn , Y ) u(X, y1 )

the two-attribute utility function is multilinear iff

with u(x1 , y1 ) = 0 and u(xn , y1 ) = 1.

u(xn , Y ) = 1 + b u(x1 , Y )

Substitute u(xn , Y ) with 1 + b u(x1 , Y ) in the above results in

for some constant b > 0.

u(X, Y ) = u(x1 , Y ) + u(X, y1 ) + (b 1) u(x1 , Y ) u(X, y1 )

Under the above conditions, we have that
u(X, Y ) = u(x1 , Y ) + u(X, y1 ) + (b 1) u(x1 , Y ) u(X, y1 )

with u(x1 , y1 ) = 0 and u(xn , y1 ) = 1.

where u(x1 , y1 ) = 0 and u(xn , y1 ) = 1.

206 / 401

207 / 401

Exploiting mutual utility independence

Exploiting mutual utility independence example

Consider two attributes X and Y with values x1 , . . . , xn , n 2,

and y1 , . . . , ym , m 2.

Suppose the utilities, given fixed A = 60 resp. C = 600,

assessed by the City of Utrecht are:

If X and Y are mutually utility independent, then u(X, Y ) can be

constructed from two subutility functions and one point:
1. u(X, yk ) for some value yk of Y
2. u(xi , Y ) for some value xi of X
3. u(xj , yl ) for some values xj 6= xi of X, and yl 6= yk of Y

u(15, 600) = 0.75 u(30, 600) = 0.35 u(50, 600) = 0.15 u(60, 600)
u(60, 200) = 0.25 u(60, 300) = 0.20 u(60, 400) = 0.15 = 0.00

if X is utility independent of
Y , but Y is utility dependent
of X:

if X and Y are mutually utility

independent:

In addition, the utility u(30, 300) = 0.30 is assessed.

Mutual utility independence implies strategical equivalence of all
horizontal functions with u(C, 600) and all vertical functions
with u(60, A):
600

we first determine the function

u(C, 300);
we then use u(C, 300) to derive all vertical functions from
u(60, A).

300

60

30

X
208 / 401

209 / 401

Example continued (II)

Example continued (I)

The utilities assessed by the City of Utrecht:

The utilities assessed by the City of Utrecht:

u(15, 600) = 0.75 u(30, 600) = 0.35 u(50, 600) = 0.15 u(60, 600)
u(60, 200) = 0.25 u(60, 300) = 0.20 u(60, 400) = 0.15 = 0.00

u(15, 600) = 0.75 u(30, 600) = 0.35 u(50, 600) = 0.15 u(60, 600)
u(60, 200) = 0.25 u(60, 300) = 0.20 u(60, 400) = 0.15 = 0.00

In addition, the utility u(30, 300) = 0.30 is assessed.

In addition, the utility u(30, 300) = 0.30 is assessed.

we first determine the function u(C, 300): since C is UI of A,

there exist functions g and h such that e.g.:

we then use u(C, 300) to derive all vertical functions from

u(60, A): since A is UI of C, there exist functions f and k such
that e.g.:

u(C, 300) = g600 (300) u(C, 600) + h600 (300)

u(C, A) = f60 (C) u(60, A) + k60 (C)

We require two equations to find g600 (300) and h600 (300):

We require two equations to find f60 (C) and k60 (C):

u(30, 300) = 0.30 = g600 (300) u(30, 600) + h600 (300)

u(C, 300) = f60 (C) u(60, 300) + k60 (C)

u(C, 600) = f60 (C) u(60, 600) + k60 (C)

= g600 (300) 0.35 + h600 (300)

u(60, 300) = 0.20 = g600 (300) 0.00 + h600 (300)

Given that u(C, 300) and u(C, 600) are known functions, we
can compute an f60 (ci ) and k60 (ci ) for each ci .

resulting in u(C, 300) = 0.29 u(C, 600) + 0.20

210 / 401

211 / 401

The multilinear utility function IIa

The multilinear utility function IIIa

T HEOREM

Let X and Y be two attributes with values x1  . . .  xn , n 2,

and y1  . . .  ym , m 2. If X and Y are mutually utility
independent then the two-attribute utility function is multilinear:
u(X, Y ) = kX uX (X) + kY uY (Y ) + kXY uX (X) uY (Y )
where
u(X, Y ) is normalised by u(x1 , y1 ) = 0 and u(xn , ym ) = 1;
uX (X) is a conditional utility function on X, normalised by
uX (x1 ) = 0 and uX (xn ) = 1;
uY (Y ) is a conditional utility function on Y , normalised by
uY (y1 ) = 0 and uY (ym ) = 1;
kX = u(xn , y1 ) > 0, kY = u(x1 , ym ) > 0, kXY = 1 kX kY are
scaling constants.

L EMMA

Let X and Y be two attributes with values x1  . . .  xn , n 2,

and y1  . . .  ym , m 2.
If X and Y are mutually utility independent then the two-attribute
utility function is multilinear:
u(X, Y ) = u(X, y1 ) + u(x1 , Y ) + k u(X, y1 ) u(x1 , Y )
where
u(x1 , y1 ) = 0, u(xn , y1 ) > 0 and u(x1 , ym ) > 0;
u(xn , ym ) u(xn , y1 ) u(x1 , ym )
k=
is a scaling constant.
u(xn , y1 ) u(x1 , ym )

212 / 401

213 / 401

The multilinear utility function IIIb

The multilinear utility function IIb
Proof of Lemma: (sketch):
Mutual utility independence implies the existence of functions
f > 0, g > 0, h, and k such that for arbitrary xl and yl
(I) u(X, Y ) = gl (Y ) u(X, yl ) + hl (Y ) and

u(xi , yj ) = u(xi , y1 ) + u(x1 , yj ) + k u(xi , y1 ) u(x1 , yj )

(II) u(X, Y ) = fl (X) u(xl , Y ) + kl (X)

1
2
3
4
5

Proof of the Theorem:

From the previous lemma we have that for arbitrary values xi
and yj of X resp. Y , we have

set yl to y1 and xl to x1
solve (I) for x1 to get h1 (Y ) and for xn to get g1 (Y )
solve (II) for y1 to get k1 (X) and for ym to get f1 (X)
substitute these results in (I) and (II) (I*) and (II*)
now solve (II*) for xn and fill in this result in (I*) to get the
desired result.

214 / 401

if u(x1 , y1 ) = 0, u(xn , y1 ) > 0 and u(x1 , ym ) > 0;

Normalising u(x1 , Y ) gives us u(x1 , Y ) = kY uY (Y ) with
kY = u(x1 , ym ).
Similarly, u(X, y1 ) = kX uX (X) with kX = u(xn , y1 ).
Finally, define kXY = k kX kY .

215 / 401

Assessing a multilinear utility function: example

Interpreting scaling constants (I)

Suppose the City of Utrecht assesses the following conditional

utilities for Cost and Acres lost:
uC (15) = 1.0 uC (30) = 0.5 uC (50) = 0.2 uC (60) = 0.0
uA (200) = 1.0 uA (300) = 0.8 uA (400) = 0.5 uA (600) = 0.0

1.0

(15, 600)

Suppose that u(0, 10) = u(100, 0).

100 u = 0.75 = k
Y
Y

Constants kC = u(15, 600) and kA = u(60, 200) are assessed:

kC

Let X and Y be two attributes such that Y ranges from 0 to 100

and u(X, Y ) = 0.25 uX (X) + 0.75 uY (Y ).

(15, 200)

(60, 200)

Then rescaling results in:

u = 0.25
u=0

10

(1 kC ) (60, 600)
kA
(15, 200)
1.0

Suppose we decide it is sufficient for Y to range from 0 to 10.

u=1

u = 0.25 = kX
100

0
X

u = 1

10 u = a = k

(1 kA ) (60, 600)

The indifference probabilities found: kC = 0.75 and kA = 0.45.

So, u(C, A) = 0.75 uC (C) + 0.45 uA (A) 0.2 uC (C) uA (A)

u = 0

kX
= U (100, 0) = U (0, 10) = kY

Did Y just turn from three times

as important as X to equally important?

u = a = k

100
X

216 / 401

217 / 401

Interpreting scaling constants (II)

Consider a multilinear utility function u(X, Y ) over attributes X
and Y with values x1  . . .  xn , n 2, resp. y1  . . .  ym , m 2.
kXY : consider two lotteries over values xi  xj and yk  yl :
0.5

0.5

(xi , yk )

A:

(xi , yl )

B:
0.5

AB
AB
AB

(xj , yl )

0.5

(xj , yk )

The multiplicative utility function

T HEOREM

Let X and Y be two attributes with values x1  . . .  xn , n 2,

and y1  . . .  ym , m 2. If X and Y are truly mutually utility
independent then the two-attribute utility function is
multiplicative:
u(X, Y ) = (k u(X, y1 ) + 1) (k u(x1 , Y ) + 1)

kXY = 0
kXY > 0 (X, Y are complements)
kXY < 0 (X, Y are substitutes)

kX , kY : if you would rather swing x1 to xn than y1 to ym , then

kX > kY , and vice-versa;
Even a mighty important attribute will have a small scaling
constant if its range is relatively small!
218 / 401

where
u(x1 , y1 ) = 1, (xn , y1 ) > 1 and u(x1 , ym ) > 1;
|u(xn , ym ) u(xn , y1 ) u(x1 , ym )|
k=
> 0 is a scaling
u(xn , y1 ) u(x1 , ym )
constant.
219 / 401

The additive utility function IVa

The multiplicative utility function

Proof of the Theorem:
Mutual utility independence implies that u(X, Y ) is multilinear:

C OROLLARY

Let X and Y be two attributes with values x1 , . . . , xn , n 2, and

y1 , . . . , ym , m 2.

u(X, Y ) = u(X, y1 ) + u(x1 , Y ) + k u(X, y1 ) u(x1 , Y )

If X and Y are mutually utility independent then

where
u(x1 , y1 ) = 0, u(xn , y1 ) > 0 and u(x1 , ym ) > 0;
u(xn , ym ) u(xn , y1 ) u(x1 , ym )
k=
is a scaling constant.
u(xn , y1 ) u(x1 , ym )

the two-attribute utility function is additive if

[0.5, (x1 , yk ); 0.5, (xi , y1 )] [0.5, (x1 , y1 ); 0.5, (xi , yk )]

Now, u(X, Y ) u (X, Y ) = k u(X, Y ) + 1, where k = |k| > 0.

u (X, Y ) is multiplicative:

for some values xi , yk for which

(x1 , y1 ) 6 (x1 , yk ) and (x1 , y1 ) 6 (xi , y1 )

u (X, Y ) = k (u(X, y1 ) + u(x1 , Y ) + k u(X, y1 ) u(x1 , Y )) + 1

= (k u(X, y1 ) + 1) (k u(x1 , Y ) + 1)

Under the above conditions, we have that

= u (X, y1 ) u (x1 , Y )

u(X, Y ) = u(x1 , Y ) + u(X, y1 )

where u (x1 , y1 ) = 1, u (xn , y1 ) > 1 and u (x1 , ym ) > 1.

where u(x1 , y1 ) = 0, u(xn , y1 ) > 0 and u(x1 , ym ) > 0.

220 / 401

221 / 401

The additive utility function IVb

An example

Proof of the Corollary (sketch):

Let X and Y be two attributes with values x0 , x1 , x2 and y0 , y1 , y2 .

Let u(X, Y ) be a two-attribute utility function defined as follows:

Given mutual utility independence, we have that

u(X, Y ) = u(X, y1 ) + u(x1 , Y ) + k u(X, y1 ) u(x1 , Y )

u(x0 , y0 ) = 1.0 u(x0 , y1 ) = 0.9 u(x0 , y2 ) = 0.5

u(x1 , y0 ) = 0.8 u(x1 , y1 ) = 0.7 u(x1 , y2 ) = 0.3
u(x2 , y0 ) = 0.5 u(x2 , y1 ) = 0.4 u(x2 , y2 ) = 0.0

where u(x1 , y1 ) = 0, u(xn , y1 ) > 0, u(x1 , ym ) > 0.

Let xi , yk be values such that

X and Y are mutually utility independent and for x0 and y0 we

have

u(x1 , yk ) + u(xi , y1 ) = u(x1 , y1 ) + u(xi , yk )

that is, u(x1 , yk ) + u(xi , y1 ) = u(xi , yk ).

u(x0 , y0 ) + u(x2 , y2 ) = u(x0 , y2 ) + u(x2 , y0 )

Under this constraint, we find from the multilinear form that

k u(xi , y1 ) u(x1 , yk ) = u(x1 , yk ) + u(xi , y1 ) u(xi , y1 ) u(x1 , yk )
= 0
Since u(xi , y1 ) 6= u(x1 , y1 ) and u(x1 , yk ) 6= u(x1 , y1 ), k must be
zero.

222 / 401

and
u(x0 , y0 ) 6=

u(x0 , y2 )
u(x2 , y0 )

u(X, Y ) is therefore additive.

223 / 401

Preferential independence the formal definition

Attribute X is preferentially independent of attribute Y [PI] if
conditional preferences for values of X given a fixed value for Y
do not depend on the particular value for Y . More formally,
D EFINITION

An attribute X is preferentially independent of an attribute Y iff

for any consequences (xi , yk ) and (xj , yk ) over X Y with Y
fixed to yk , we have

Verifying independences

(xi , yk )  (xj , yk ) = (xi , yl )  (xj , yl ) yl of Y

If X is preferentially independent of Y and Y is preferentially
independent of X then X and Y are mutually preferential
independent [(M)PI].
224 / 401

Preferential independence example

Validating preferential independence

Let X and Y have values x0 , x1 , x2 and y0 , y1 , y2 . Assume the

two-attribute utility function is defined as:

Let X and Y be two attributes with values x1 , . . . , xn , n 2, and

y1 , . . . , ym , m 2.

u(x0 , y0 ) = 1.0 u(x0 , y1 ) = 1.0 u(x0 , y2 ) = 0.7

u(x1 , y0 ) = 0.6 u(x1 , y1 ) = 0.8 u(x1 , y2 ) = 0.5
u(x2 , y0 ) = 0.0 u(x2 , y1 ) = 0.5 u(x2 , y2 ) = 0.2
Then we have:
X is preferentially independent of Y :
x0  x1  x2 for each value of Y ;
Y is not preferentially independent of X:
y1 y0 y2 for X = x1
y1 y2 y0 for X = x2 !

225 / 401


226 / 401

Use the following procedure to verify whether X is preferentially

independent of Y :
1 choose some consequence (xj , y1 );
2 ask the decision maker for a value xi for which
(xi , y1 ) (xj , y1 ) for some value xj of X and y1 of Y ;
3 for a number of values y 6= y1 of Y , ask the decision maker
whether (xi , y) (xj , y) still holds;
4 repeat steps 1, 2 and 3 for different values of X.
5 check orientation (no preference reversal?!)

227 / 401

UI implies PI

PI implies UI ?

P ROPOSITION

Consider two attributes X and Y with values x1 , . . . , xn , n 2,

and y1 , . . . , ym , m 2.

If X is preferentially independent of Y then X not necessarily

utility independent of Y .

If X is utility independent of Y then X is preferentially

independent of Y .

Counter example:
Consider the following utility function:
u(x0 , y0 ) = 1.0 u(x1 , y0 ) = 0.8 u(x2 , y0 ) = 0.3
u(x0 , y1 ) = 0.5 u(x1 , y1 ) = 0.1 u(x2 , y1 ) = 0.1

Proof: Consider xi , xj and yk such that (xi , yk )  u(xj , yk ),

that is, u(xi , yk ) u(xj , yk ) 0.
UI now implies the existence of functions g > 0 and h such that
for arbitrary yl :

X PI Y since yi : x0  x1  x2 ;
we have no X UI Y :

u(xi , yk ) = gl (yk ) u(xi , yl ) + hl (yk ) and

u(xj , yk ) = gl (yk ) u(xj , yl ) + hl (yk )

(x1 , y) [p, (x0 , y); (1 p), (x2 , y)]

From gl (yk ) > 0, we now have u(xi , yl ) u(xj , yl ) 0 for

arbitrary yl .

holds for p 0.71 if y y0 , and for p = 0, if y y1 .

228 / 401

229 / 401

Validating utility independence

AI implies (M)UI

If X PI Y , then X is also utility independent of Y if in the

following procedure xC is equivalent for all values of Y :

P ROPOSITION

choose a lottery [0.5, P ; 0.5, Q] where P = (x1 , y) and

Q = (xn , y) for some value y of Y and values x1 and xn of X;

ask the decision maker whether or not he prefers the lottery

[0.5, P ; 0.5, Q] to a consequence (xi , y) for some xi ,
x1  xi  xn ;

repeat step 2 until you

converge to the certainty
equivalent (xC , y) of the
lottery;
repeat steps 1 3 for
different values of Y .

Repeat the procedure for different pairs of values for X.

ym

y1

xC

xC

xC

x1

Consider two attributes X and Y with values x1 , . . . , xn , n 2,

and y1 , . . . , ym , m 2.
If X and Y are additive independent then X and Y are mutually
utility independent
Proof (sketch):

We prove X UI Y ; Y UI X is analogous.

AI implies u(X, yk ) = u(X, y1 ) + u(x1 , yk ) for arbitrary yk , where

u(x1 , yk ) is constant w.r.t the value of X.
Continuity implies, for arbitrary xi , xj and xk with
(xi , y1 )  (xj , y1 )  (xk , y1 ), that p such that
(I) u(xj , y1 ) = p u(xi , y1 ) + (1 p) u(xk , y1 ).
Substitution of each u(X, y1 ) in (I) with u(X, yk ) u(x1 , yk ) gives

u(xj , yk ) = p u(xi , yk ) + (1 p) u(xk , yk ) for arbitrary yk .

xn
X
230 / 401

231 / 401

Validating additive independence example

Suppose the City of Utrecht assesses the following conditional
utilities for Cost and Acres lost:
uC (15) = 1.0 uC (30) = 0.5 uC (50) = 0.2 uC (60) = 0.0
uA (200) = 1.0 uA (300) = 0.8 uA (400) = 0.5 uA (600) = 0.0

MUI implies AI ?
If X and Y are mutually utility independent then X and Y are
not necessarily additive independent.

Suppose kC is assessed using the following lottery:

kC

Counter argument:
We have seen that an additional assumption is necessary to
conclude additive independence given mutual utility
independence (see Additive utility function IVa).

1.0

The mentioned corollary can be used to validate AI given MUI;

another option is to assume that AI holds. . .

(15, 600)

(15, 200)

(1 kC ) (60, 600)

resulting in kC = 0.75. As a consistency check, kA is assessed

as well:
kA
(15, 200)

1.0

(60, 200)

(1 kA ) (60, 600)

resulting in kA = 0.45. We now have that kC + kA = 1.2 6= 1.0 !

232 / 401

233 / 401

Summary
The different functional forms for u(X, Y ) are valid under the
following conditions:
additive:

iff AI
(I, II)
if X UI Y and a lottery assumption for Y
(III)
if MUI and a lottery assumption for (xi , yk ) (IV)

multi-lin.:

if X UI Y and a specific s.e. assumption

if MUI
if AI (k = 0)

multiplic.:

if multi-linear and k 6= 0

(I)
(II, III)
(III)

234 / 401

MAUT with n = 2 attributes: isopreference curves

235 / 401

Isopreference curves definition

Utility functions with one utility independent attribute
Let X and Y be two attributes. Suppose that X is utility
independent of Y , but the reverse does not hold.

An isopreference curve describes a set of consequences that

are equally desirable to the decision maker.
D EFINITION

Consider two (sets of) attributes X and Y and a partial function

i : X Y such that for arbitrary values xi 6= xk of X and yj , yl of
Y , we have that

Recall that the two-attribute utility function can be specified

using three subutility functions:

i(xi ) = yj and i(xk ) = yl

u(X, Y ) = u(x1 , Y ) [1 u(X, y1 )] + u(xn , Y ) u(X, y1 )

(xi , yj ) (xk , yl )

An isopreference curve is an interpolant ~ Y (X) of i(X).

where u(x1 , y1 ) = 0 and u(xn , y1 ) = 1.

Note that for any two points (xi , yj ) and (xk , yl ) on the
isopreference curve, we thus have that

Other options are:

replacing one or two subutility functions with one or two,
respectively, isopreference curves.

u(xi , yj ) = u(xk , yl ) = c for some constant c.

236 / 401

As a result, u(X,~ Y (X)) = c is a constant subutility function,

defined on all values of X.

237 / 401

Isopreference curves details (II)

Isopreference curves details (I)

ym

ym

(xi ,~2Y (X))

~2Y (X)

~2Y (X)
yj

(xi ,~2Y (xi ))

yj
~1Y (xj )

yi

yi
y1

y1
x1

~1Y (X)

~1Y (xj )

~1Y (X)

xj

xn

x1

xi

xj

xn

u(xi ,~2Y (xi )) is the utility of a (and any) point on the

Here we have for all values x of X that

isopreference curve ~2Y (X);

(xi ,~2Y (X)) is a line in an X Y -subspace, intersecting
~2Y (X);
u(xi ,~2Y (X)) is a subsubutility function, defined on a single
value of X and a subset of Y .

(x,~1Y (x)) (x1 , yi ), that is: u(x,~1Y (x)) = u(x1 , yi )

(x,~2Y (x)) (x1 , yj ), that is: u(x,~2Y (x)) = u(x1 , yj )
238 / 401

239 / 401

One isopreference curve for one subutility function

An example
Jenny wants to treat her friends to some cookies and candy. Let
xi denote i cookies and yj indicate j pieces of candy.
Jenny is indifferent between (x2 , y10 ), (x6 , y6 ), and (x8 , y4 ). Also,
Jenny is indifferent between (x1 , y9 ), (x3 , y5 ), and (x7 , y3 ).
~1Y (X)

Assessment of one utility

for a single point on the
~1Y (X) curve gives the utility of all points (x,~1Y (x)).

~2Y (X)
y2
x1

Recall that if X is utility independent of Y then

u(X, Y ) = u(x1 , Y ) [1 u(X, y1 )] + u(xn , Y ) u(X, y1 )
Replace subutility function u(xn , Y ) over all
yi by subutility function
u(~ X (Y ), Y ) over all yi :

Y
y10

Let X and Y be two attributes with values x1  . . .  xn , n 2,

and y1  . . .  ym , m 2.

ym

Replace subutility function u(X, y1 ) over all xi

by subutility function
u(X,~ Y (X)) over all xi :
ym

yk

x8

We have for example that ~1Y (x2 ) = y10 and ~2Y (x3 ) = y5 , but also
~1Y (x4 ) = y8 and ~2Y (x4 ) = y4 .

y1x x
k
1

240 / 401

xn

y1x x
k
1

xn
241 / 401

Substitution of u(~ X (Y ), Y ) for u(xn , Y ) (b)

Substitution of u(~ X (Y ), Y ) for u(xn , Y ) (a)

Proof (sketch):
C OROLLARY

Let X and Y be two attributes with values x1  . . .  xn , n 2,

and y1  . . .  ym , m 2.
If X is utility independent of Y then


u(xk , y1 ) u(x1 , Y )
u(X, Y ) = u(x1 , Y ) +
u(X, y1 )
u(~ X (Y ), y1 )

Utility independence implies the existence of functions g > 0, h

such that
(I)

1
2

where
u(x1 , y1 ) = 0
~ X (Y ) is defined such that (~ X (Y ), Y ) (xk , y1 ) for an
arbitrary xk 6= x1 .

242 / 401

u(X, Y ) = gl (Y ) u(X, yl ) + hl (Y ) yl

set yl to y1 ;
solve (I) for x1 to get hl (Y ), using u(x1 , y1 ) = 0;
let xk be the point where ~ X (Y ) intersects the line (X, y1 ),
then u(~ X (Y ), Y ) = u(xk , y1 ); use this in solving (I) for ~ X (Y )
to get gl (Y );
substitute these results in (I) to get the desired result.

243 / 401

An example
Suppose the City of Utrecht assesses the following utilities for
Cost and Acres lost:
u(15, 600) = 0.75 u(60, 100) = 0.25
u(30, 600) = 0.50 u(60, 200) = 0.20
u(45, 600) = 0.20 u(60, 300) = 0.15
u(50, 600) = 0.10 u(60, 400) = 0.10
u(60, 600) = 0.00 u(60, 600) = 0.00
In addition, the City indicates the indifferences
(50, 200) (45, 300) (40, 400)
resulting in an isopreference curve ~ C (A) with u(~ C (A), A) = 0.40
and e.g. ~ C (600) = 35 (does not follow from above).
If Cost is utility independent of Acres lost, then


u(35, 600) u(60, A)
u(C, 600)
u(C, A) = u(60, A) +
u(~ C (A), 600)

Substitution of u(X,~ Y (X)) for u(X, y1 ) (a)

C OROLLARY

Let X and Y be two attributes with values x1  . . .  xn , n 1,

and y1  . . .  ym , m 1.
If X is utility independent of Y then
u(X, Y ) =
u(x1 , Y ) u(xn ,~ Y (X)) u(xn , Y ) u(x1 ,~ Y (X))
u(xn ,~ Y (X)) u(x1 ,~ Y (X))
where
~ Y (X) is defined such that (X,~ Y (X)) (x1 , yk ) for an yk with:
u(x1 , yk ) = 0, (x1 6= xn )

244 / 401

245 / 401

An example

Substitution of u(X,~ Y (X)) for u(X, y1 ) (b)

Proof (sketch):
Utility independence implies the existence of functions g > 0, h
such that
(I)

u(X, Y ) = gl (Y ) u(X, yl ) + hl (Y ) yl

Now let yk be the point where ~ Y (X) intersects the line (x1 , Y ),
then u(X,~ Y (X)) = u(x1 , yk ).
1 set yl to yk ;
2 solve (I) for x1 to get hk (Y ), using u(x1 , yk ) = 0
3 solve (I) for xn to get gk (Y );
Y (X) to get u(X, yk ).
4 solve (I) for ~

5 substitute these results in (I) to get the desired result.
246 / 401

Suppose the City of Utrecht assesses the following utilities for

Cost and Acres lost:
u(15, 200) = 1.00 u(60, 200) = 0.20
u(15, 300) = 0.90 u(60, 300) = 0.15
u(15, 400) = 0.80 u(60, 400) = 0.10
u(15, 600) = 0.75 u(60, 600) = 0.00
u(15, 900) = 0.50 u(60, 900) = 0.15
In addition, the City indicates the indifferences
(15, 1500) (50, 700) (60, 600)
resulting in an isopreference curve ~ A (C) with
u(C,~ A (C)) = 0.00 and e.g. ~ A (30) = 900.
If Cost is utility independent of Acres lost, then
u(C, A) =

u(60, A) u(15,~ A (C)) u(15, A) u(60,~ A (C))

u(15,~ A (C)) u(60,~ A (C))
247 / 401

Use of two isopreference curves

Substitution of u(~1X (Y ), Y ) and u(~2X (Y ), Y )

for u(x1 , Y ) and u(xn , Y ) (a)

Recall that if X UI Y then

u(X, Y ) = u(x1 , Y ) [1 u(X, y1 )] + u(xn , Y ) u(X, y1 )
Replacing u(x1 , Y ) by u(~1X (Y ), Y ), and u(xn , Y ) by u(~2X (Y ), Y ):
ym

C OROLLARY

Let X and Y be two attributes with values x1  . . .  xn , n 2,

and y1  . . .  ym , m 2.
If X is utility independent of Y then

u(X, Y ) =

y1

x1 xk

u(X, y1 ) u(~1X (Y ), y1 )
u(~2X (Y ), y1 ) u(~1X (Y ), y1 )

where
u(X, Y ) is normalised by u(xk , y1 ) = 0 and u(xl , y1 ) = 1;
~1X (Y ) is defined such that (~1X (Y ), Y ) (xk , y1 );
~2X (Y ) is defined such that (~2X (Y ), Y ) (xl , y1 )

xn

X
Is there another possibility with two isopreference curves?
248 / 401

An example

Substitution of u(~1X (Y ), Y ) and u(~2X (Y ), Y )

for u(x1 , Y ) and u(xn , Y ) (b)

Suppose the City of Utrecht assesses the following utilities for

Cost and Acres lost:
u(5, 600) = 1.00 u(50, 600) = 0.10
u(12, 600) = 0.70 u(60, 600) = 0.00
u(15, 600) = 0.75 u(75, 600) = 0.10
u(30, 600) = 0.50

Proof (sketch):
Utility independence implies g > 0, h such that
(I) u(X, Y ) = g(Y ) u(X, yl ) + h(Y ) yl
Let xk , xl be the intersection points for ~1X (Y ), ~2X (Y ), resp., with
(X, y1 ).
1
2

set yl to y1 ;
solve (I) for xk and ~1X (Y ) to get h(Y ) and g(Y ), using
u(~1X (Y ), Y ) = u(xk , y1 ) = 0;
solve (I) for ~2X (Y ) to get u(xk , Y ), using
u(~2X (Y ), Y ) = u(xl , y1 ) = 1;
substitute these results in (I).

249 / 401

In addition, the City indicates the indifferences

(80, 200) (75, 300) (70, 400) (60, 600)
resulting in isopreference curve ~1C (A) with u(~1C (A), A) = 0.00,
and the indifferences
(15, 200) (12, 300) (10, 400) (5, 600)
resulting in isopreference curve ~2C (A) with u(~2C (A), A) = 1.00.
If Cost is utility independent of Acres lost, then


250 / 401

u(C, A) =

u(C, 600) u(~1C (A), 600)

u(~2C (A), 600) u(~1C (A), 600)

251 / 401