You are on page 1of 40

Multi-Attribute Utility Theory

Tuesday 2nd February


Definitions
Not a monetary value- but • Expected utility, in decision
takes into account the theory, the expected value
attitude of the decision of an action to an agent,
maker (ranks the attitude 0 calculated by multiplying
to 1) the value to the agent of
each possible outcome of
the action by the probability
of that outcome occurring
and then summing those
numbers.
Multi-Attribute Utility Theory

 Introduction
 Simplifying assumptions of a multi-attribute utility
concept
 Case Study of a multi-attribute utility problem
 Decision tree representation of multi-attribute utility
problem.
•Weighted Objective Decision Analysis.
•Case Study: The Heathrow Rail Link

Goodwin & Wright pages 142 onwards


PREAMBLE
Expected Monetary Value (EMV) = the sum of the pay-offs, weighted by their respective probabilities. (it’s
all about the £/$/ /Euro/)

Expected Utility Criterion (EUC) = a broader consideration of pay-offs in respect of the decision-maker’s
attitude to risk (not just about the £/$/ /Euro)

Key points
1. Utility Theory describes how it is impossible to derive the value from a good or service,
but it is usually possible to rank the alternatives in order of preference to the consumers.

2. The Expected Utility Value is particularly useful in situations where potential gain or loss
is of significant size to the capacity of the concerned organisation. (See Appendix 1)

3. Multi-attribute utility theory represents a formal approach of countering problems posed


by conflicting objectives amongst a set of finite alternatives.
PREAMBLE
Expected Monetary Value (EMV) = the sum of the pay-offs, weighted by their respective
probabilities.

Expected Utility Criterion (EUC) = a broader consideration of pay-offs in respect of the


decision-maker’s attitude to risk

Key points
1. Utility Theory describes how it is impossible to derive the value from a good or service,
but it is usually possible to rank the alternatives in order of preference to the consumers.

2. The Expected Utility Value is particularly useful in situations where potential gain or loss
is of significant size to the capacity of the concerned organisation.

3. Multi-attribute utility theory represents a formal approach of countering problems posed


by conflicting objectives amongst a set of finite alternatives
.
Multi-attribute (1)
 The outcomes of most decision problems cannot be immediately characterised in terms of a single
value such as cost or profit. Typical construction project multi-attribute objectives include:

Lowest Earliest
Cost Time

Highest
Quality

 Equally, many decision problems have been found to have objectives in which satisfaction of one,
leads to the dissatisfaction of some other(s). Hence a mechanism of objective resolution needs to
be formulated or developed.

Put simply “cannot satisfy all of the people all of the time”
Built Environment.
Multi-attribute (2)
 The outcomes of most decision problems cannot be immediately characterised in terms of a single
value such as cost or profit. Typical construction project multi-attribute objectives include:

Lowest Earliest
Cost Time

Highest Multi-attribute
Quality utility region

 Equally, many decision problems have been found to have objectives in which satisfaction of one,
leads to the dissatisfaction of some other(s). Hence a mechanism of objective resolution needs to
be formulated or developed.

Put simply “trying to satisfy (to some extent) as many people as possible some of the time”
The triple constraint
Cost, time, quality
Three elastic bands joined together; pull one and it changes the others
the Built Environment.
Multi-attribute (key points)
 
 A multi-attribute utility theory represents a formal approach for countering problems posed by
conflicting objectives among a set of finite alternatives.

 A multi-attribute utility function is a quantitative representation of a decision-maker’s preference


structure for such multiple objectives (Moore & Thomas, 1988)
Goodwin & Wright (page 143 onwards)

 The general aim of the theory is to produce an overall utility function which yields an utility index of
measure of worth for a given set of alternatives.

 A suitable utility function should include the preference structure of the decision maker.

 The multi-attribute utility equation is:


and the Built Environment.
Multi-Attribute Utility Concept (simplifying assumptions)
The choice of attributes by which to judge alternative designs, systems, products or projects is one of
the most important tasks in multiple-attribute decision-making.

The following observations regarding the attributes used to discriminate among alternatives can be
made:
1. Each attribute distinguishes at least two alternatives. [In no case should the identical values for an attribute apply to all
alternatives].
2. Each attribute captures a unique dimension of the decision problem (i.e. attributes are independent and non-
redundant).
3. Taking all the attributes together, we assume they are sufficient (enough) for the purposes of choosing the best
alternative.
4. Differences in values assigned to each attribute are assumed to be meaningful in distinguishing among feasible
alternatives.
5. Perform consistency checks to see if the multi-attribute utility function really does represent the decision-maker’s
preferences.
6. In order to aid the functionality of multi-attribute utility functions, assumptions on its nature are formulated or invoked
DEFINITIONS.
*Linearity - basically implies a constant trade-off between one attribute and another
*Additivity - implies the rate of trade-off between two variables may depend upon the values of those two variables but will
not depend upon the values of the other variables.

Attribute – “a quality or feature regarded as a characteristic or integral part of someone or something”


So it could be decisions makers care about : cost, safety record, environmental issues, company profile,
survival, personal ambition, beating a rival, finishing early, the company’s long term sustainability, doing a
good job here so we win more business later on………….
and the Built Environment.
More on the assumptions

The following observations regarding the attributes used to discriminate among alternatives can be
made:
1. Each attribute distinguishes at least two alternatives. [In no case should the identical values
for an attribute apply to all alternatives].
2. Each attribute captures a unique dimension of the decision problem (i.e. attributes are
independent and non-redundant).
3. Taking all the attributes together, we assume they are sufficient (enough) for the purposes of
choosing the best alternative.
4. Differences in values assigned to each attribute are assumed to be meaningful in
distinguishing among feasible alternatives.
5. Perform consistency checks to see if the multi-attribute utility function really does represent
the decision-maker’s preferences.
6. In order to aid the functionality of multi-attribute utility functions, assumptions on its nature
are formulated or invoked:
*Linearity - basically implies a constant trade-off between one attribute and another.
*Additivity - implies the rate of trade-off between two variables may depend upon the values of those two variables
but will not depend upon the values of the other variables.
and the Built
Linearity and additivity
 ADDITIVITY = implies the rate of trade-off between two variables may depend upon the values of
those two variables (but will not depend upon the values of the other variables).

 LINEARITY = the property of a mathematical relationship or function which means that it can be graphically
represented as a straight line. “No matter what the value of my asset position, market share and turnover, I would
always consider a 1 percent unit increase in market share as valuable as a $100,000 increase in assets.'

 Please note – other assumptions exist but the concepts of additivity and linearity are sufficient for
this Unit.

Attribute – “a quality or feature regarded as a characteristic or integral part of someone or something”


So it could be decisions makers care about : cost, safety record, environmental issues, company profile,
survival, personal ambition, beating a rival, finishing early, the company’s long term sustainability, doing a
good job here so we win more business later on………….
The triple constraint
The three attributes = Cost, time, quality
But as we just saw - there are other attributes to consider
and the Built Environment.
More Ideas on the Utility Concept.
 Many behavioural scientists believe that economic decisions are governed mainly by self-
interest and rationality.

 Rationality, in this context suggests that extrinsic incentives are assumed to shape human
behaviour. (1)

 However, recent developments have shown that intrinsic incentives (i.e. other less conscious
factors) help to shape human behaviour (1,4)

 Typically, decision-makers are assumed to form probabilistic beliefs or expectations (e.g. about
the states of nature and the effect of the actions) and to process the information (i.e. come to
their decision) according to statistical principles.

 Expected Utility, is a measure of the individuals implicit value or preference, for each policy in
the risk environment. Utility theory is the formal approach to measuring the decision-maker’s
attitude towards risk (2). However, it has been found that the decision-maker’s preference
sometimes contradict the expected utility hypothesis (3,4)
_____________________________________________________________

(1) Kahneman & Smith, 2002, (2) Raftery, 1994. (3) Allais paradox (4) Goodwin &
Wright
and the Built Environment.
More Ideas on the Utility Concept.
 Many behavioural scientists believe that economic decisions are governed mainly by self-
interest and rationality.

 Rationality, in this context suggests that extrinsic incentives are assumed to shape human
behaviour. ---- PUT SIMPLY coming or operating from outside

 However, recent developments have shown that intrinsic incentives (i.e. other less conscious
factors) help to shape human behaviour ---- PUT SIMPLY coming or operating from inside

 Typically, decision-makers are assumed to form probabilistic beliefs or expectations (e.g. about
the states of nature and the effect of the actions) and to process the information (i.e. come to
their decision) according to statistical principles.

 Expected Utility, is a measure of the individual’s implicit value or preference, for each policy in
the risk environment. Utility theory is the formal approach to measuring the decision-maker’s
attitude towards risk. However, it has been found that the decision-maker’s preference
sometimes contradict the expected utility hypothesis ----put simply humans are humans
and the Built
Practical steps to deriving a typical multi-attribute utility function
 
Steps to derive a multi-attribute utility function for a construction project include:
 Derive a single-attribute utility function for the attributes (say over-run time and project cost)

 Combine the single-attribute utility functions to obtain a multi-attribute function in order to


compare the alternative courses of action in terms of their performances over both attributes.

, ) = ] + + ][]

Above equation is exactly as before.

Worked examples are coming up ------------

Expected Utility, is a measure of the individual’s implicit value or preference, for each policy in
the risk environment. Utility theory is the formal approach to measuring the decision-maker’s
attitude towards risk. (Raftery, 1994). However, it has been found that the decision-maker’s
preference sometimes contradict the expected utility hypothesis. A vivid illustration of such is
indicated in the findings of Allais paradox ----------------[ see Appendix 3].
and the Built
Case study of a multi-attribute utility problem (1)
 A major sub-sea installation project. What attributes does it have?
 Quality, safety, time, cost, develop future technologies, support employment, support UK security
through less reliance on Middle East oil, minimise environmental risk, re-use of equipment
and the Built
Case study of a multi-attribute utility problem (2)
 Company policy states to simplify the project by having only two major attributes of:
 Cost
 Time

From there

 The company appoint a project manager (the “decision maker”).

 He estimates the sub-sea project will be completed within a cost range of £50 to £140 million

 He estimates the sub-sea project can be completed within 3 to 9 months.

Questions: is the PM qualified to make these estimates? Does he have the experience to make the necessary
decisions on his own? Decisions of this size deserve other people’s input? Decisions of this size and
complexity deserve a sensitivity analysis?
and the Built
Case study of a multi-attribute utility problem (3)

  In order to find the single-attribute value (), a recourse is made to the probability equivalence
method of utility elicitation in which an offer is made to the project manager to make a choice
between the following alternatives:

a) A project where over-run is certain to be at its best level (i.e. 0 months), but where cost is

certain to be at worst level (i.e. £140 million)

b) A lottery which offers a probability of that both cost and over-run will be at best level (i.e. 0

weeks and £50 million) and a () probability that they will be at their worst levels (i.e. 6

months and £140 million).


and the Built
Case study of a multi-attribute utility problem (4)
 STAGE ONE: DERIVATION OF AN UTILITY FUNCTION

We assign “Best Outcome” (say, zero time overrun) a numerical utility value of “1.0”.

We assign “Worst Outcome” (say, 6-month time overrun) a numerical utility value of “0.0”.

Assuming that after being asked a series of questions (i.e. the elicitation session), the project
manager indicates that he is indifferent (i.e. unconcerned) between
(a) A sub-sea project which will certainly over-run by 3 months; and
(b) A gamble offering a 60% chance of a project with zero over-run ; and a 40% chance
of a 6-month over-run.

Mathematically,

U (3 month) = [0.6 x 1] + [0.4 x 0] = 0.6

Note ; “say”, “assume”, “indicates”, “series of questions”


See Slide Conclusions (2)
and the Built Environment.
Case study of a multi-attribute utility problem (5)

In a similar process, the project manager indicates indifference between


(a) A sub-sea project which will certainly over-run by 1
month; and
(b) A gamble offering a 90% chance of a project with zero over-
run and a 10% chance of a 6-month over-run.

U (1 month) = [0.9 x 1] + [0.1 x 0] = 0.9


 The same elicitation process is used for all the “possible costs of the subsea storage project”

Project over-run Utility Cost of the Project (£ Utility


targets(months) million)
0 1.0 50 1.00
1 0.9 60 0.96
3 0.6 80 0.90
6 0.0 120 0.55
140 0.00
• Not indifference to the project-

Indifference
• lack of concern about,
• unconcern about,
• apathy about/towards,
• nonchalance about,
• lack of interest in,
-but the point where the decision
• disregard for, maker’s no longer has a preference.
• obliviousness to,
• un-involvement in/with

For example:
• reached the expected profit value,
so other attributes become more
important.
Elicitation techniques
• Brainstorming.
• Document Analysis.
• Focus Groups.
• Interface Analysis.
• Interviews.
• Observation.
• Prototyping.
• Requirements Workshop

 Numbers are the goal as this


removes ambiguity, lack of
consistency between different
people etc
and the Built Environment.
Case study of a multi-attribute Utility Problem (6)

  STAGE TWO Definition and estimation of the values of parameters


An important step that follows after the derivation of the utility function is the definition and estimation
of the values of parameters in the equation:

 , ) = The multi-attribute utility if, attribute 1, has a level , and attribute 2 has a level

 ] = The single-attribute utility if attribute 1, has a level ,

 = The single-attribute utility if attribute 2, has a level ,

 = Used to weigh the single-attribute values


and the Built Environment.
Case study of a multi-attribute utility problem (7)
 STAGE THREE OBTAINING MULTI-ATTRIBUTE UTILITY FUNCTION
We combine these two utility functions to obtain the multi-attribute utility function.

Utility 1.0 - Utility 1.0 -

0.9 -

0.7 -

0.5 - 0.5 -

0.3 -

0.1 -
0 1 2 3 4 5 6 50 140
Over-run time (months) Project cost ( £ million)

 
 If the two functions are mutually utility independent it can be shown that:

, ) = ] + + ][]
and the Built
Case study of a multi-attribute utility problem (8)
 Graphically;
CHOICE 1 CHOICE 2
Best time = zero
Best cost = £50m

Over-run Cost 1
𝑘
 
Best = zero Worst = £140m
 1 −
𝑘
1 Worst time = 6 months
Worst cost = £140m

CHOICE 2 CHOICE 2
Best time = zero
Best cost = £50m

Over-run Cost 2
𝑘
 
Worst = 6 months Best = £50 m
 1 −
𝑘
2 Worst time = zero
Worst cost = £140m
and the Built Environment.
Case study of a multi-attribute utility problem (9)

  Assuming the decision-maker for the sub-sea project indicates that:

= 0.8

And = 0.6

Therefore in a multi-attribute situation where mutual utility independence holds, it is valid that:

= 1,

So for the above scenario, = 1 -

Hence, = 1 – 0.8 – 0.6 = -0.4

Writing down the multi-attribute utility function gives:


, ) = ] + + ][]

, ) = 0.8 ] + 0.6 - 0.4 ][]


.
Case study of a multi-attribute utility problem (10)
 
 STAGE FOUR: CHECKING MULTI-ATTRIBUTE UTILITY FUNCTION FOR CONSISTENCY
Considering the oversights and biases that could attend the decision-making process, the decision-
maker is often encouraged to check for consistency. This is done to ensure a fair representation of a
decision problem. Checking for consistency can be done through:

“Tracking back” through the analysis and explaining why one option has performed well
and another has not. .

Offering the decision-maker(s) new sets of alternatives to rank.

Sense checking, logic etc

Making sure no mistakes were made

• Reviewing previous similar examples

Sensitivity analysis should be carried out on the probabilities and utilities to examine the
effect of changes in values of and
In reality: Multi-Attribute Utility Theory in the oil industry

The oil company would have to balance


• Time
• Safety
• Pollution
• Restoration of the environment
• Oil platform dismantling
• Profit
• Government requirements
• Health & Safety
• Which part of the world they are operating in?
• Reputation

• Think of examples of indifference?


Other examples: where Multi-Attribute Utility Theory could be used: Transport Company

• Profit
• Passenger numbers
• Performance standards (e.g. % on time, quality of service)
• Customer satisfaction
• Customer choice (bus v train v tram v walk v cyle)
• Contracted responsibilities (e.g. provide unprofitable services, reduce
air pollution, provision of accessible buses)

Think of examples of indifference?


Construction - Multi-Attribute Utility Theory in reality:

• Profit
• Time
• Quality
• Reputation
• Safety
• Long term performance of the asset (e.g. sports centre)
and the Built
Conclusions (1)
 Multi-attribute utility concept provides a formal mechanism to counter conflicting objectives of a project in
order to achieve an optimal risk analysis function.

 Approaches to decision-making under complete uncertainty include Maximin, Maximax, Laplace Rule, Hurwicz
criterion of realism and Minimax Regret Criterion

 The Expected Monetary Value (EMV) and the Expected Utility Value (EUV) are two most widely-used criteria for
decision problems involving RISK - where probabilities of occurrence of each option can be estimated

 Allais paradox (Appendix 3) is an illustrative situation where a decision-maker’s preference can contradict the
expected utility hypothesis. Others like the decision to insure a house –emphasise that it is humans who make
decisions. Hence, the principles of the EUV and EMV needs to be properly understood and cautiously approached
before applying to decision-analysis problems.
and the Built
Conclusion (2)

 Further studies have provided a way to handle multi-attribute utility problems which have multiple number
of attributes and where mutual utility independence does not exist. [We will not cover these in any detail) However,
although the complexities of these problems have provided models of little practical value. it is often very
necessary to balance the feasibility of the analysis and the need to model reality and complexity as far as
possible

 Crucial set-backs of the multi-attribute utility concept are that it [1] requires major commitment of time and
efforts (e.g. the elicitation exercise) and [2] the available mechanism(s) lack a thorough explicitness.

“stated clearly and in detail, leaving no room for confusion or doubt” : think back a few slides Note ; “say”,
“assume”, “indicates”, “series of questions”

 Sensitivity analysis should always be employed to judge the robustness of an option even in a multi-
attribute utility situation. This is in order to obtain guidance on the aspects of the project that require most
attention.- see previous lectures
and the Built Enent.

Appendix 1: case study: The Heathrow Rail Link


and the Built Enent.

The Heathrow Rail Link (1)


The aim of this study, carried out in the early seventies, was to produce recommendations
on the construction of a rail link between Central London and Heathrow airport. Four
possible schemes were identified:
 LT: An extension of the Piccadilly Line (on the London Underground system) to Heathrow
with the present air-line coach services continuing.

 BR1: A British Rail (overground railway) link between Victoria main line station and Heathrow,
with a check-in option at Victoria.

 BR2: A similar link to BR1, but with the coach services continuing.

 BR3: A similar link to BR1, but without a check-in at Victoria and with coach services
continuing.

A twenty-five year planning horizon (1970 – 1994) was used and all costs and benefits were
considered with reference to continuation of the present ‘coaches only’ system.
and the Built
The Heathrow Rail Link (2)
The costs and benefits were identified and, with the help of appropriate models, e.g. for the
forecasts of air traffic, five items were identified for each year in the planning period.

1. Capital costs (mostly incurred in the first four to five years);

2. Annual operating costs;

3. ‘Time saved’ benefits for various sections of the community;

4. Congestion benefits (due to a reduction in coaches and private cares travelling to and from
Heathrow);

5. Private and public resources savings (from fares and private vehicle running costs in London).
.
The Heathrow Rail Link (3)
Using a 10% discount rate for items (1) and (2) for future years and established techniques
of monetization such as Travel Cost Method, items (3), (4) and (5) were converted into
monetary values. The present value figures obtained (in £millions) are shown in the table
below:
LT BR1 BR2 BR3
Capital costs +11.9 +22.6 +21.7 +15.9
Annual Operating costs -6.6 -5.0 +4.2 -3.7
Time saved -27.4 -19.2 -19.2 -32.3
Congestion benefits -1.6 -0.2 -0.2 -3.1
Resource savings -7.8 -0.9 -0.6 -0.6

Total cost (NPV) -31.5 -2.7 +5.9 -24.1

The ranking obtained based on increasing levels of desirability is LT, BR3, BR1 and BR2.
This study is an example of a public-policy decisions where, although there are several attributes
(such as cost, comfort, time-saved, etc.), the outcomes are to a close approximation deterministic.
the Built Environment.
Appendix 2: Decision Tree representation of multi-attribute utility problem

A Project Manager has to make a choice between “working normally” and “hiring extra labour” on a major infrastructure
contract. There is a 10% chance that working normally will have no time over-run (0-week over-run) and cost £50,000;
a 60% chance that it will have 3-week overrun and cost £60,000; and a 30% chance that it will have a 6-week overrun
and cost £80,000.

On the other hand, the Project manager can “hire extra labour”, of which there is a 80% chance of no time overrun and
cost £120,000 and a 20% chance of a 1-week overrun and cost £140,000. The utility values for the individual options
on the major infrastructure work are as follows:
Project cost () Utility
Time over-runs () Utility
£50,000 1.00
0-week 1.00
£60,000 0.96
1-week 0.90
£80,000 0.90
3-week
3-week 0.60
0.60
£120,000 0.55
6-week
6-week 0.00
0.00
£140,000 0.00

Given that the time-overruns and project costs are mutually-utility independent and the mutual-attribute utility function
can be written as U (Over-run, Project Cost) = 0.8U(Over-run) + 0.6U(Project Cost) – 0.4U(Over-run)(Project Cost).
Compute the expected utilities of the options and state the optimal course of action.
and the Built Environment
Decision Tree representation of multi-attribute utility problem
  Recall that our multi-attribute utility equation is:

, ) = 0.8 ] + 0.6 - 0.4 ][]


 Hence, U(3 weeks over-run, £60,000 cost) =

0.8 u(3 weeks over-run) + 0.6 u(£60,000 cost) – 0.4 u (3 weeks over-run) u (£60,000 cost)

= 0.8 (0.6) + 0.6 (0.96) - 0.4 (0.6) (0.96) = 0.8256

 The table below shows the derivation for other possibilities from the multi-attribute utility function

Possible Probabilities Over-run Project Utility


choices time Cost
Work 0.1 0 weeks £50, 000 1.0
Normally
0.6 3 weeks £60, 000 0.8256
0.3 6 weeks £80,000 0.54
Hire extra 0.8 0 weeks £120,000 0.91
labour, etc.
0.2 1 week £140, 000 0.72
and the Built Environment.
Decision Tree representation of multi-attribute utility problem
 The decision tree representation of the multi-attribute utility problem below is thus:

Over-run Project Utility


time cost
0 weeks £50,000 1.0
0.1

Expected utility 3 weeks £60,000 0.8256


lly = 0.757 0.6
o rma
n
rk
Wo 6 weeks £80,000 0.54
Possible 0.3
Hir
ee
choices xtra
lab 0 weeks £120,000 0.91
our 0.8
, et
c.
Expected utility
= 0.872
Key
6 weeks £140,000 0.72
 0.2
Utility values
 Probability values
 Pay-off values.

You might also like