Professional Documents
Culture Documents
Multi-Criteria Analysis
This annex presents a tool to allow systematic modelling of decision makers preferences.
The annex has been prepared by ERM.
MULTI-CRITERIA TECHNIQUES
1.1
BACKGROUND
Multi-criteria techniques are tools developed in the field of decision theory to aid
problem-solving. They entail the systematic modelling of a decision makers1
preferences to solve in an explicit manner a choice between options involving a
number of, often conflicting, objectives. Through the aggregation of disparate
information onto a common index of utility or value they aim to provide a
rational basis for classifying choices.
The field of decision theory is wide, and has received contributions from, inter
alia, the disciplines of engineering, mathematics, psychology, management
science and economics. The multi-criteria approach examines how all the
relevant aspects of a problem are assessed and traded off by decision-makers.
Essentially it is a top-down exercise, based on a decision-makers perception of
how a decision can be decomposed into trade-offs between objectives. Elicitation
techniques may be used to reveal such outcome preferences. The multi-criteria
technique employs data on the performance of competing options against the
decision-makers stated objectives and develops a composite utility function for
each option. Detailed source material is widely available, for example in Keeney
(1992), Keeney and Raiffa (1976), von Winterfeldt and Edwards (1986) and
Watson and Buede (1987).
1.2
INTRODUCTION
Multi-criteria techniques encompass a large family of methods of which 40 or
more different approaches are distinguishable in the literature, from the highly
sophisticated through to simple rating systems (Nijkamp, 1986; Nijkamp et al
1990)2.
The common rationale of these methods is to establish a broad framework for
assessing the impact of making a choice, simplifying the decision into its
constituent elements. In most cases the method requires developing a complete
set of alternative solutions to a problem (the options), assessing all relevant
performance information for criteria which judge the value or utility of the
options, and trading-off the relative significance of the criteria to resolve the
problem. Subjective and implicit decision making can be thus made objective and
(1) 1 The singular decision maker should be regarded are interchangeable with plural decision makers throughout this note.
(2) 2 Many techniques have clearly defined methodologies, including multi-attribute value theory (MAVT), multi-attribute
utility theory (MAUT) and the analytical hierarchy process (AHP).
1.3
MEETING OBJECTIVES
In making a choice, a decision maker should consider all the relevant costs and
benefits of the options (in the widest possible sense) to ensure they make a sound
decision that adequately addresses all concerns. The relative preference for
alternative options can be judged by quantifying their performance against a set
of relevant objectives, attributes or dimensions, which in total describe the
options value to the decision maker (Miller, 1985). The preferred option should
be that which, on balance, comes closest to meeting the decision makers
objectives, which may often conflict.
In practice, it is unlikely that any one option will perform best against all
objectives and can be clearly preferred; each will demonstrate different
advantages and disadvantages. Describing the balance between objectives, and
identifying the preferred option is a complex problem.
1.4
MAKING TRADE-OFFS
This complexity may be resolved by making trade-offs between objectives, which
entails determining a trade-off or substitution ratio; how much of one can be
surrendered in order to achieve another. In practice, this is usually done
intuitively when a choice is made, recognising, for example, that option As
superior performance against objective X clearly outweighs option B being
preferred for objective Y. However, comparisons are rarely as simple as this.
Even where the comparison is straightforward, there are considerable advantages
in making trade-offs explicitly, within an explicit decision-aiding framework.
The process is structured and comprehensive, ensuring that all concerns are
identified and addressed. The approach should retain sufficient flexibility for the
robustness of trade-off decisions to be thoroughly explored, and it should be
sufficiently transparent to ensure that the reasons behind a particular choice are
made clear. The advantages of a structured approach are particularly apparent
where there are many alternatives and/or numerous conflicting objectives, where
such a transparent logic can aid communication, debate and the route towards
consensus.
DEFINING ALTERNATIVES
The next step in the analysis is the identification of the set of alternative options.
Note that these should be compiled after the objectives and constraints have been
set. The option set should be comprehensive to ensure that no viable option is
omitted. However, it may be advantageous to start with a limited, but diverse,
set. This avoids analysing what may be a large number of options which are
closely related, and allows poorly performing options to be weeded out quickly.
(4) 1 It is important that those responsible for approving a choice actively participate in the process and therefore own the
decision. It is their beliefs and values which should drive the analysis. Supporting analysts may, of course, develop background
material at each step, but in order for the framework to reflect and assist the decision making process the choices made should be
those of the key players. The decision maker(s) may choose to include external stakeholders in this process if they wish.
PREDICTING PERFORMANCE
In order to estimate the performance of each option the decision maker may need
to consult a number of sources. These will clearly vary with the decision under
analysis, but are likely to include site-specific and/or generic data concerning
each of the decision criteria. Developing a performance matrix may involve
applying data from the literature, bespoke and proprietary models and
canvassing opinion from the public, experts and using Delphi Panel techniques.
Wherever appropriate, measures of risk or uncertainty can be included in the
performance data, using central estimates, ranges and distributions, and can be
used in the evaluation and sensitivity analysis.
Qualitative assessment
Where the performance of an option cannot reasonably be quantified against an
objective, a qualitative assessment can still provide a valuable input into the
analysis. Such an assessment may take the form of a ranking using a Likert Scale,
a subjective evaluation of performance or even simple descriptions or numbering
of impacts. In each case a basis for distinguishing between the options for that
objective is established, and it can be included in the process rather than
overlooked. There is no need to exclude information simply because it is soft or
fuzzy, which may be the only way of describing performance against some
intangible criteria (Nijkamp, 1986).
2.4
VALUING PERFORMANCE
Even in a simple matrix, it is unlikely that one option will out-perform the others
against all the objectives identified by the decision maker. Making a choice will
be difficult, but can be simplified by making trade-offs between the objectives or
criteria. The decision maker must consider how much of one objective it is
prepared to surrender in order to achieve another. This can be achieved by first
establishing how valued, or desirable, are the performances of the alternative
strategies with respect to the objectives, and then weighting these derived
functions and combining them into an overall measure of performance.
To do this, performance measures should be converted to value by normalising
scores to a common range, say 0-100, where the most favoured scores 100, and
the least 0. However, the scale between best and worst performance may not
always be linear. The value associated with a unit on the objective/criteria scale
may change according to the point on the scale at which that unit lies. These
changes in preference along a scale, or value functions, are more likely to be
important where there are dramatic ranges of performance, or performance
thresholds, and can be elicited by skilled facilitators or through the use of
appropriate decision support software1.
Where uncertainty or risk associated with performance is evaluated, the function
can take account of preference changes with respect to the probability of various
outcomes and is generally called a utility function.
(5) 1 There are many software tools on the market developed to service the growing use of decision tools in fields as diverse as
mathematics, psychology, economics and management science. Many of these originate in the USA and can be traced through
software listings under decision support or multi-attribute analysis. Tools developed in the UK include Hiview and Equity
developed by Enterprise LSE (Houghton Street, London, WC2A 2AE, tel: 0171 955 7128, fax: 0171 955 7427).
2.6
WEIGHTING OBJECTIVES
Introduction
In order to establish a composite measure of performance across all the objectives
selected, or a combined value or utility function, thus providing a basis for
identifying preferred options, the objectives/criteria must be weighted according
to how important each is regarded in relation to the others.
Require a decision-maker to articulate how much one criterion willing to forego
in meeting higher levels of performance against another criterion. (Miller, 1985).
The trade-off or substitution ratio expressed allows preference for different
criteria to be expressed on the same scale (Vinke, 1992).
Weighting Techniques
A range of techniques can be used for estimating and modelling preferences,
depending on the time available, the difficulty of the task and the required
precision of the outcome (Nijkamp et al, 1990). These are described in detail in
the literature, with many authors having their own particular preferences.
Techniques for developing weights include ex post analysis, analysis of
documentation where weights are implicit, use of hypothetical priorities and
interactive methods based on interviews, questionnaires and elicitation
techniques. Some approaches have developed prescriptive weighting sets to
ensure consistent analysis with regard to a specific problem (Environment
Agency, 1999).
Interactive methods include ranking techniques, verbal statements on weights,
distribution of points, scenario formulation and pairwise comparison or swing
weighting (Nijkamp, 1986, Reid and Christensen, 1994; Keeney, 1992). These
offer the opportunity to engage decision makers further in structuring and
resolving the problem, and for exploring the consequences of their expressed
preferences.
Weights represent a particular value and preference set, and clearly they will
change with the views of the decision maker, with corresponding alteration to the
preferred outcomes. The effect may be explored with different weighting sets for
different stakeholder groups, for example, business groups, policy makers, green
groups, professional groups and academics (Macdonald, 1997; Sobral et al 1981;
Chung and Poon, 1996) or by using a weighted sum of group specific weights
(Miller, 1985).
Decision Conferencing
Where an interactive method is used to determine weights with a group of
decision makers or stakeholders, the process is best conducted in a facilitated
decision conference where the decision makers have been prepared with suitable
background material on the evaluation of performance and associated
uncertainties. This material should place the performance of each option firmly
within an appropriate context to assist the statement of preferences. The meeting
should discuss, and where necessary, revise, the framework for the analysis and
the background information supplied, and should debate preferences for the
ranges of performance1 offered by the alternatives against all criteria, establishing
a consensus on the weighting factors to be used.
2.7
RESULTS
Once an appropriate set of weights has been derived these can be applied to the
normalised performance scores and preferred options identified. The results of
the analysis can be surprising, and should not be taken as inviolate, the approach
is flexible and open -ended, not deterministic. A thorough examination of the
sensitivity of the overall conclusion to the assumptions made in the analysis, to
uncertainties, and to weighting factors stemming from plurality of opinion is
necessary to explore the decision envelope around the preferred options, and
examine the robustness of the indications (Stirling, 1996).
The weighted, normalised data should be examined to discover the factors that
are most significant in determining the overall ranking. The decision maker(s)
should discuss whether it they are happy with these, their implications and the
related assumptions, or whether they indicate an error or misunderstanding in
performance evaluation or the weighting process. In many cases iteration will be
necessary to refine the alternatives, carry out more precise modelling or debate
further the weights which should be used.
2.8
(6) 1 The requirement is to weight the range of performance against objective X against the range of performance against
objective Y, rather than weight the inherent importance of the objectives out of context. Thus the group should derive a trade-off
between objectives ,i.e. how much of the range of performance against objective X are they prepared to substitute for a change in
performance of best option to worst option for objective Y.
2.9
Table 2.1
Socio-political impact
Social equity
Ease of administration /
implementation
Compatibility with the public
administration principles
Environmental impact
Land used
Material recovered
Waste coverage
Waste elimination
Net energy recovered
Local air pollution
Transportation
Global air pollution
Potential for waster pollution
Land contamination and
future restriction
Disamenity
Other health risk
Noise