Professional Documents
Culture Documents
Multi-Criteria Decision Making (MCDM) Techniques in Planning
Multi-Criteria Decision Making (MCDM) Techniques in Planning
(MCDM) Techniques in
Planning
.
BRYAN
H.
MASSAM
PERGAMON PRESS
OXFORD . NEW YORK . BEIJING . FRANKFURT
SAO PAULO' SYDNEY TOKYO TORONTO
JPP 30:1-A
This monograph
0305-9006/38 $0.00+.51.
Copyright 1983 Pergamon Press pic
Contents
Abstract
Acknowledgements
1.
2.
3.
Overview of Literature
1.1. Introduction
1.2. MCDM and Related Fields
1.3. MADM, MODM, MAUT, PCT
1.4. Toward a Generic Planning Problem
9
9
10
15
18
21
21
22
28
32
36
37
38
39
40
40
40
42
45
45
46
47
48
49
54
60
4.
Progress in Planning
4.1. Introduction
4.2. Site Selection Problems
4.2.1. The location ofhealth centres in Zambia
4.2.2. The fire station location problem in North York. Canada
4.3. Route Alignment Problems
4.3.1. Motorway in France (Bourges-Montlucon)
4.3.2. Interstate highway in Georgia, U.S.A.
4.4. Priority Rating Problems
4.4.1. Paris subway station renovations
4.4.2. Ontario transportation projects
4.5. Conclusions
S.
63
63
65
65
67
69
69
71
72
73
74
7f.
Bibliography
81
Abstract
This monograph provides an introduction, a review and a critique of multi-criteria
techniques as used in the resolution of planning problems. It argues that the
techniques allow information on planning goals and objectives to be converted into
evaluation criteria and to be brought into a framework that incorporates the opinions
of interest groups. A review of the literature suggests that multi-criteria decision making
(MCDM) techniques have been developed mainly in operations research, though social
psychology, regional science and business management also offer useful contributions.
With the continuing interest among planners on evaluation techniques it is appropriate
that the MCDM work be closely scrutinized to inform debates on improvements to
planning processes.
In order to discuss the utility of the techniques for planners a generic planning
problem (GPP) is defined. This involves the evaluation of options, using multiple
criteria and different opinions of interested parties. Alternative objectives are also
considered. The basic features of the major classes of MCDM techniques are discussed
within the framework of the GPP and the fundamental assumptions and data
requirements of the techniques are outlined in non-technical language.
To give a better understanding of the role of the techniques in planning a set of
empirical case studies is included. Three types of problem are considered, first, the site
selection problem, second, the route alignment problem and third the priority rating
problem given a large number of options and a fixed budget. In conclusion the author
argues that recognition of the complexity of the milieu within which planning occurs
means that the search for an ideal formal MCDM technique to solve planning
problems is a chimera, but their role in helping to enlighten and organize choice
among planning options is important. Also included are criteria for judging an
MCDM technique and comments on the thorny issue of determining what is meant by
the quality of planning.
Acknowledgements
Some twenty years ago I worked in La Direction Scientifique of SEMA-METRA in
Paris and it was there that I met Bernard Roy and was introduced to multi-criteria
decision-making techniques. During the intervening years I have followed with great
interest the attempts which have been made to refine and apply the techniques to
practical planning problems. Particularly I have been concerned about the lack of
contact among theoreticians and practitioners, and this has provided the main
incentive for this monograph, namely to try to bridge the gap between these two
solitudes.
lowe a number of debts of gratitude for help in the preparation of this manuscript.
I will not try to enumerate the precise amounts lowe other than to say a sincere thank
you for the hospitality, informal chats, detailed comments, advice, criticisms,
references and encouragement that was offered to me. My thanks extend to colleagues
and friends in Australia, Canada, France, Israel and the United Kingdom and include
in alphabetical order, Colin Adrian, Ian Askew, Wade Cook, Derek Diamond, Barry
Garner, Andrew Karski, Nurit Kliot, Mal Logan, Toni Logan, Lionel Lawrence,
Virginia Maclaren, Jean Marchet, Brian McLoughlin, Jim Micak, Bernard Roy, Arie
Shachar, Jean Siskos, Ian Skelton, Bob Snowdon, Stanley Waterman and Don Webb.
Also, I am particularly grateful to the Planning Group under James Balfour, at M.
M. Dillon (consulting engineers, planners, environmental scientists), Toronto, Canada,
for our discussions regarding the use of MCDM techniques.
I alone bear responsibility for the contents of this monograph, the opinions and the
conclusions.
Partial financial support was provided by the Social Sciences and Humanities
Research Council of Canada, and the Joint Program in Transportation of The
University of Toronto and York University and the Faculty of Arts, York University,
this is gratefully acknowledged.
Florence Davies and Agnes Fraser typed the manuscript and the figures were drawn
by the staff in the Cartographic Laboratory at York University - many thanks.
CHAPTER 1
Overview of Literature
1.1. INTRODUCTION
10
Progress in Planning
MCDM techniques. We will seek to identify a generic planning problem within the
context of a planning process recognizing that most planning processes are conflict
resolution situations of one sort or another. Given that one of the aims of this
monograph is to demonstrate the practical use of the techniques, we wish to avoid
providing a generic problem which is too vague or abstract.
The second chapter will include basic descriptions of the data requirements and the
underlying principles used in MCDM techniques. We will also provide a checklist
which can help judge the worth and utility of a technique. While the application of this
checklist can be changed to suit particular planning problems we envisage that the
items included will cover most cases. In Chapter 3 we will focus on a discussion of a
selection of MCDM techniques. We will avoid esoteric details of algorithms which
may detract the reader's attention from obtaining an appreciation of the basic
principles and assumptions of the techniques. However, some detail must be provided
so that the analyst, planner, or interested decision-maker is not left with just a list of
references without appropriate operational definitions or practical applications. It is in
Chapter 4 that a small set of empirical planning problems will be treated using selected
MCDM techniques. The problems will be selected to demonstrate the variety of actual
planning problems which are covered by the generic planning problem. In the final
chapter we will offer an assessment of the MCDM techniques as they could or should
be used in planning, and given that many share a concern with improving planning we
must tackle the question of judging how successfully a particular planning process
operates. This topic will form part of Chapter 5.
The study of MCDM is closely related to other topics, for example, Multi-Attribute
Decision Making (MADM), Multi-Attribute Utility Theory (MAUT), Multi-Objective
Decision Making (MODM) and Public Choice Theory (PCT). Authors who have
reported on these topics include Keeney and Raiffa (1976); Zeleny (1975); Thiriez and
Zionts (1976); Bell et al. (1977); Nijkamp (1979); Nijkamp and Spronk (1981); Rietveld
(1980); Hwang and Yoon (1981); French et al. (1983); Voogd (1983); Linstone (1984)
and Fandel and Spronk (1985). Nijkamp and Spronk (1981) suggest somewhat
optimistically that, "multicriteria analysis appears to be becoming a new mode of
thinking for decision-making, planning theory, choice analysis and conflict
management," however they add the important caveat; "it occurs too often that those
designing new techniques ... are hardly aware of the varying needs and desires of the
potential users of these techniques" (p. 1.). Bell et al. (1977) admit that "the analytical
tools that were developed to aid decision makers facing complex problems originally
addressed only single-objective problems ... the artificiality and restrictiveness of that
approach for most real world problems has led to the development of various methods
11
for handling multiple objective problems". A note of caution and realism has been
added by Batty in his article in Rationality and Planning edited by Breheny and Hooper
(1985). Batty indicates that formal rationality models may be too restrictive in dealing
with complexity for they tend to impose order where there may be no order. This is a
view shared by Radford (1980). He is critical of the use of formal decision analysis
approaches for tackling plan evaluation problems and he argues that neither of the
two most important approaches to organization decision-making, namely the rational
comprehensive style or the incremental approach is entirely satisfactory. It seems to
Radford that the mixed scanning approach propounded by Etzioni (1967) embodies a
fundamentally new idea, for instead of regarding the conceptualization of a decision
problem as given in terms of alternatives and outcomes, it views this conceptualization
as an active process engaged in by decision-makers. Perhaps this can be stated more
forcefully as the consensus-building approach. This of course recognizes that complex
decision problems of plan generation, evaluation and selection do involve problems of
designing and judging alternatives, but Radford (1986) argues this takes place within
the context of long-term strategies and shorter-term tactics. With this in mind he has
attempted to elaborate on a strategic-tactical model which could be applied to plan
evaluation, selection and implementation exercises.
In the 1970s decision analysis was offered as the rational way to tackle a whole
range of choice problems. Bell et al. (1977) suggested that "the aim of decision analysis
is to decompose a problem into two parts: one to indicate the probabilities of different
possible consequences of each alternative and the other to evaluate the desirability of
those consequences". This seems a perfectly reasonable and sensible way to proceed as
long as all the necessary data can be obtained and there is agreement among all parties
regarding the alternatives and their consequences. Rarely are these demanding
conditions satisfied. At this point we should remind ourselves of some of Hardin's
work (1984) on the Tragedy of the Commons problem. He argues that while technical
solutions to choice problems may always be welcome it is clear that there is a class of
real problems which can be defined as "no technical solution problems". This is often
the situation in planning in which individual or group interests compete with those of
the whole society or collectivity. The dilemma these conflicts pose is taken up by
Hollis et al. (1985) and others and we discuss it later in this chapter.
It seems clear that the distinctions among the four areas, MODM, MADM, MAUT
and PCT, while conceptually significant, are probably less important when we
consider that they are all related to the generic planning problem. This problem can be
characterised initially as follows:
Given a set of alternate plans, each characterised by a set of assessments for selected criteria, and a
set of interest groups whose opinions regarding the selection of criteria and the assessments have to
be considered, provide an appropriate procedure to define the attractiveness of the alternate plans
with a view to identifying the best one.
Later we will look at the two sub-problems which are contained within this large one
and then we will restate the above problem as a single generic planning problem
(GPP). At this stage we will avoid the thorny tasks first, of offering formal definitions
of the terms in the statement, and second, placing the problem into the broader socio-
12
Progress in Planning
political context of a society. Both of these items will be taken up later. We suggest
that many practical planning problems can be accommodated by the GPP and thus it
should be appealing to practitioners who are less concerned with the esoteric
characteristics of each of the four related areas. Comments on each of the four areas
are provided further on in this chapter. Ideally we would like to offer a format for a
planning process which would allow all these approaches to be integrated, however
this is probably an unrealistic goal, therefore it seems more appropriate to restrict
attention to providing suitable descriptions of the principles of MCDM techniques so
that clients can make judicious selections to suit their particular purposes.
At the outset it shouldlbe made clear that we do not intend to argue that a
particular technique exists which will provide the definitive solution to a planning
problem, and that this solution must be implemented. We do however subscribe to the
view that the process of organizing information as an integral part of a technique can
serve as a valuable part of a planning exercise. While a technique might offer insights
regarding the information; the interpretation of the results depends upon the values
and attitudes of those who examine the technical results. Keeney (1981) makes this
case very clearly:
There is no such thing as an objective value free analysis. Furthermore, anyone who purports to
conduct such an analysis is professionally very naive, stretching the truth, or using definitions of
objective and value free which are quite different from those commonly in use.
He goes on to add that a logical systems analytical framework is needed that makes
explicit the necessary professional and value judgments. The element in the planning
process which is lacking is the ability to integrate.
What has been lacking is not information but a framework to integrate and incorporate it with the
values of the decision makers to examine the overall implications of each alternate [plan].
13
1974; Bernstein and Mellon, 1978; Wilson, 1974; Wilson and Kirkby, 1975; Batty and
Hutchinson, 1983). In toto this gave rise to a movement which encouraged the use of
mathematics. However, there seems to be some confusion on this score, for while it is
the case that some of the tools used by mathematicians were cited as being useful,
there have been very few applications of specific axioms, theorems and formal proofs
to planning problems.
In order to place the MCDM techniques into a broader quantitative framework we
can identify three types of work which characterise formal numerical approaches.
(1) Models using inferential or descriptive statistics.
(2) Models based on systems analysis or mathematical programming.
(3) Models based on data base management techniques and information systems.
Elements from all of these could well form part of MCDM techniques and it is
assumed that the student of MCDM will have some knowledge of these three
important areas.
Some have argued that the technocratic approach to plan evaluation (Self, 1975) is
limiting as it tends to reduce choice problems to naive data sets without explicit
appreciation of political values and the role of interest groups. Perhaps more damning
comment is offered by others who believe that the technocratic thrust blinds us to the
real problems of society which can best be understood by examining social structures,
the role of states, politics, legislative instruments and modes of production. Let us not
fall into the trap of building a fence around the MCDM techniques, rather we ask the
question: can we use such techniques to help in the provision of commentary on
matters relating to effectiveness, efficiency and equity as they relate to planning modes
and the outcomes of planning practices.
In a very useful review by Healey (1986) of the recent writings of David Harvey, she
draws attention to comments contained in Harvey's The Urbanisation of Capital (1985)
which show "how planners are drawn into the task of managing change in the built
environment in order to facilitate production, circulation, exchange and consumption.
In doing so, they are caught in the crossfire of the conflicts inherent within and
between these processes. He [Harvey] highlights the significance of the concepts of
social balance, planning for people's welfare, and open decision-making as necessary
supports to interventions which have to be seen as legitimate to diverse and conflicting
interests". Healey goes on to argue that "Harvey demolishes the Aunt Sallies of many
critics of planners as narrowly technocratic, evangelistic bureaucrats, mere lackeys of
capitalism, or naive social engineers". To substantiate this view the following
quotation is used.
. . .the fusion of technical understandings with a necessary ideology produces a complex mix within
the planning fraternity of capacity to understand and to intervene in a realistic and advantageous
way and capacity to repress, co-opt and integrate in a way that appears justifiable and legitimate
(Harvey, 1985, p. 178).
At best we hope that the techniques we discuss will help clarify particular planning
problems by assisting decision-makers, analysts and the public to define feasible
alternate solutions, and to evaluate these as systematically as possible in order to seek
an acceptable solution for implementation. This approach firmly places the techniques
14
Progress in Planning
into the context of a continuous heuristic planning process which involves the
identification of needs and resources as well as political realities of allocation choices.
While it is not our purpose to elaborate on the complex milieu within which planning
occurs, it is clear that one of the pressing problems in the profession has arisen
because of the gap between the work of theoreticians and practitioners. Therefore one
of the purposes of this monograph is to describe basic features of MCDM techniques,
encourage theoretical improvements, and to demonstrate to practitioners that these
techniques may assist in the search for solutions to their specific planning problems.
Thus we hope to try to bridge the gap, or at least put in place some connecting links
which can be reinforced later. We argue for complementarity of action without
prejudice regarding the merits of either theoretical or practical work. As the state-ofthe-art of techniques changes, so student and professional planners need to be
informed, and those who try to make improvements need to be aware of the problems
of applying the existing techniques. The pursuit of scholarly goals demands that we
continue to seek new ways of defining and solving problems, and society can
legitimately ask that advances and benefits of new ideas be incorporated into
organizations, such as local governments or planning agencies. Certainly in the private
sector, techniques are scrutinized to assess their contributions to the improvement of
the effectiveness or efficiency of the organization, as well as the competitive edge of
the firm. Obviously a planning company must be concerned with the costs and benefits
of using a new technique and the possible effects on their business. In the public sector
the planning and implementing bodies will look at the techniques with a view to
judging their contributions to enhancing the effectiveness and efficiency of the
planning process, also consideration will possibly be given to questions of equity, and
under certain circumstances to expediency. For these reasons we suggest that it is
timely that students and practising planners enhance their awareness of some recent
advances in these new techniques. It is clear that the techniques while they may
eliminate spurious conclusions, speed up the planning process, and involve opinions of
a variety of interest groups, they do not necessarily avoid conflicts among proponents
and opponents of plans. Neither are the techniques abstractly objective; the task of
making the techniques operational demands that opinions, preferences and choices be
expressed. This exercise of expression may assist in the elucidation of otherwise vague
and nebulous elements which intuitively some may argue should be a part of the
planning process. The implementation of a plan, the adoption by users, the decision to
postpone action involve choices and possibly negotiations, arbitration and the use of
the law. The MCDM techniques have little or no direct role in this phase of the
planning process, though they may serve in any monitoring exercise. Only occasionally
however is on-going monitoring of a plan's performance a major feature of the
planning process. This is perhaps particularly true in the case of the public sector with
respect to urban plans for example. It is probably much less the case when we turn to
corporate planning and this is hardly surprising given the nature of public and private
enterprises. Whereas the former typically involves many interest groups, and a number
of objectives, the latter may have a narrow set of interest groups, possibly just the
management board or the investors and a restricted set of objectives, all focusing upon
profitability for instance.
15
If the planning problem is to evaluate a finite feasible set of known alternatives and
to select the best one, given that each alternative is characterised by scores for a set of
attributes, then this is generally known as the MADM problem. However, if the
problem is to define the set of alternatives given a series of objectives which act as
constraints and/or goals, and then to find the best one, this is usually referred to as
the MODM problem. It is this latter problem which typically is tackled using
mathematical programming. As Zeleny (1984) has recently noted:
Operational sciences (operations research, management science, decision science, systems analysis)
have devoted most of their history to problems characterised by a single, aggregate criterion of
choice. They dealt with problems of measurement and search in connection with simple problems of
limited practical interest. All decisions, public and private, individual and collective, are characterised
by multiple, apparently conflicting criteria .... In very few problems of interest we find a single,
preemptively important criterion of choice. Individuals and their organizations face multiple
objectives, attributes, goals and criteria, .... Decision makers, in public and private institutions as
well as in the government, have to learn to work with multiple and parallel criteria of choice ...
(p. xii).
Both MADM and MODM conventionally begin by assuming a single decisionmaker, or at least a unified set of opinions, regarding the relative importance of the
attributes, objectives and goals. Clearly the process of estimating the scores and
measuring impacts is fraught with difficulties, and perhaps the best we can do is to
assign some probability values. If this approach is adopted then the problem can be
recast as one which seeks to evaluate the expected utilities of the alternatives and to try
to identify the alternative with the highest expected utility value and define this as the
best one. This is usually referred to as the MAUT problem. It has been extensively
described by two of its proponents, Keeney and Raiffa (1976), as an appropriate way
of aiding in a decision process. Vincke (1986) suggests that MAUT is "the privileged
approach of American researchers in multicriteria analysis but is much less used in
Europe, except by the economists for whom utility theory is a classical tool". We
might further remark that the French and Dutch schools have placed considerable
emphasis on MODM and MADM research.
It is clear that the major thrust of most researchers who have tackled these three
classes of problem is to provide an enlightened approach to decision-making. Recently
Midgley and Piachaud (1984) offer the view that:
Although it is recognized that planning techniques [such as MCDM techniques] have many
limitations ... , we believe that they are helpful aids to policy-making which enhance objectivity and
efficiency. To reject their use is to legitimise Machiavellian tendencies and traditionalism in
organisational politics and to deny the need for greater rationality in decision-making (p. 6).
While we need not subscribe to their view that rejection of techniques suggests support
for Machiavellian approaches we do believe that the development of open decisionmaking processes which use information rigorously may help to lead to expeditious
planning and the avoidance of planning disasters. As Hall (1980) has pointed out the
story of planning disasters, while they make fascinating journalism, should also lead
towards improved planning, and it is with this in mind that the latter part of Hall's
16
Progress in Planning
So troubling is this result that Hollis and his colleagues argue "that to explain how
collective actions problems are solved is one of the greatest challenges to social
science". In a recent article in Scientific American, Blair and Pollak (1983) review the
contributions of Arrow and his analytical work on the search for an ideal constitution
to provide the best choice for society while taking into account the views of
individuals. The results are well known so only a brief summary is offered here. Blair
and Pollak note that:
Three widely shared objectives - collective rationality, decisiveness and equality of power - stand
in irreconcilable conflict ... there is little comfort here for those designing ideal procedures for
collective choice. Nevertheless, every society must make collective choices and devise voting
procedures, however imperfect they may be.
We cannot shy away from trying to make evaluations of alternate plans, and while
very few jurisdictions attempt to offer any kind of formal voting procedure to elicit
opinions on particular plans, most jurisdictions have, in place, processes which involve
public participation. This may occur informally by pressure groups and the media for
example, or via the courts or other formal bodies empowered to solicit opinions and
render judgement. Arbitration, negotiations and bargaining as well as the use of
legally enforced sanctions often characterise part of the process of plan evaluation and
implementation. Surprisingly there appear to be very few attempts to incorporate the
costs of implementation into the formal plan evaluation exercise. A naive effort has
been made by Schlager (1968) who suggested that the evaluation should include a
simple probability factor to estimate the chances of a plan being acceptable. Other
more detailed efforts have been made by Wolpert and his students at the University of
Pennsylvania in the early 1970s [see for example Mumphrey et al. (1971)]. They
attempted to develop a procedure to identify the least-cost implementation plan taking
17
into account the fact that a facility package should consider not only physical costs
but also the short-run placation costs and long-run welfare distributional effects. It is
clear that the utility of MCDM techniques can be enhanced if the cost of
implementation of the alternatives is incorporated into the evaluation process.
One of the rare articles which has attempted to link formal plan evaluation
procedures with the broader social choice environment is provided by Sagar (1981).
He looks specifically at Lichfield's Planning Balance Sheet (PBS) method and the
Goals Achievement Matrix (GAM) procedure of Hill within the context of local
participatory planning. Sagar (1981) claims that "the Planning Balance Sheet and the
Goals Achievement Matrix have become established as the foremost challenges to
cost-benefit analysis in the social evaluation of planning alternatives". We can suggest
that PBS and GAM are closely related to MCDM techniques and that their success
augurs well for the future use of a variety of similar techniques. However, Sagar also
notes that: "A disadvantage of the multi-objective decision methods ... in relation to
participatory planning is their mathematical form and rather complicated
algorithms ... [also] ... relatively little attention has so far been paid in their
development to situations involving groups of decision makers" (Sagar, 1981, p. 427).
It is hoped that this monograph will contribute to the efforts of those who wish to see
closer cooperation between theoreticians who develop MCDM approaches and those
who are faced with the task of generating and evaluating plans, offering comments on
mitigation, compensation and the like, and ultimately making executive decisions as to
which plan to implement.
In the early seventies de Neufville and Marks (1974) anticipated " ... a grand
synthesis of the approaches [formal multi objective methods and the like] into a body
of procedures that will enable the planner or designer to analyse the options available,
and to present relevant information about the choices open to the interested groups,
and to suggest areas of possible compromise on a choice" (p. 303). With specific
reference to transportation planning Hutchinson (1974) claimed that:
It seems that if any real progress is to be made in evaluation, then the basis of the method should be
some consistent theory of democratic group decisions (p. 293).
Hill (1977) clearly recognized this as is evident from the following quotation:
The challenge is to find a way to enable interested and affected publics to enter the planning process
at the stages of objective formulation, generation of alternatives, and plan evaluation (p. 202).
Some ten years after these exhortations of Hill and others, Voogd (1983) reported
that: "There are only a few empirical applications known in urban and regional
planning which bear some resemblance to the ideas [formal multi-criteria evaluation]
elaborated in this book" (p. 237). Innovations in planning technology may be slow,
but unless current practises can be systematically compared to alternate modes which
may include MCDM techniques, there is little chance of change in the status quo of
the two solitudes - practitioners and theoreticians.
JPP 30:1-8
18
Progress in Planning
One of the basic features of many urban planning exercises is the generation of
alternate solutions. This is the starting point for our work, for without alternatives we
have no reason for undertaking any evaluation or assessment. However there is of
course the opportunity to provide an assessment of the status quo and possibly the
generation of normative plans using an optimization procedure for example. While
there are a number of different ways of generating alternate plans, fundamentally we
can identify proactive and reactive modes of behaviour. Some may deny that an urban
problem exists and insist that the status quo is the desired alternative, others may
argue for a particular plan and these proponents may stimulate reactions from interest
groups who previously were not motivated to offer any suggestions. The political,
economic and social context within which planning occurs, determines whether greater
or lesser emphasis is placed on soliciting the views of interest groups who may wish to
react to a proponent, also to the encouragement to offer plans by more than one
proponent or by directing a proponent to generate a particular plan which will be
implemented. These various degrees of deliberation and debate focus on the attention
which a community places on the desirability of building a consensus regarding urban
planning ventures. In the interests of encouraging open debate, while designing an
expeditious process, attempts have been made by many jurisdictions to publicise
planning options and allow public hearings prior to issuing a permit for development
and construction. Appeal processes are typically part of the planning environment.
The necessity of decision-making stems from the fact that a conflict exists and a
prerequisite for choice is that two or more alternatives must be judged and compared.
For urban planning this clearly suggests that more than one plan is available and that
evaluation and judgement of the merits is called for in order to move towards an
informed opinion. It may also be the case that a vital part of this judgemental
evaluation exercise is to attempt to build a consensus and seek public approval for a
planrying scheme. The search for such approval may be quite informal and certainly
not necessarily involve a vote or referendum, it may also be in a form which will
disarm or at least neutralise or satisfy those who felt aggrieved by a proponent's
initiative. Compensation practises, mitigation devices and compulsory purchase orders
all form part of the milieu within which much planning occurs. Such is the nature of
urban planning in democratic societies.
Perusal of the vast literature on MCDM clearly indicates that the techniques have
rarely been viewed within the context of this broader decision-making framework as
has been noted earlieL A large number of the articles on MCDM deal with technical
aspects, however, we should recall that one of the reasons for the identification,
codification and organization of the literature on MCDM is to help those who are
faced with making difficult choices. While MCDM techniques fall squarely into the
general area of decision analysis they tend to have been developed in vacuo, and apart
from planning practice.
Let us now turn our attention to two of the more formal and precise statements
which are derived from the initial statement of the generic planning problem which
was posed earlier without forgetting the caveat regarding the need to link the
19
techniques to the broader planning process. We will begin with the plan selection
problem which can be envisaged as one of selecting the best alternate plan from a
small finite set of feasible alternatives given an evaluation of each alternative on a set
of criteria. The set of alternatives can, of course, include the status quo option.
Stewart (1979) has summarized this typical planning problem in the following way:
The problem facing the decision maker is thus to select an alternative such that the collection of
criteria values (jSk, j = 1,2, ... N) is preferable to that of any other alternative.
Where Sk is the utility, attractiveness or score for plan k given a set of N criteria.
A similar problem exists in the social choice literature and it has been stated as
follows by Cook and Seiford (1978):
Consider the problem in which each member of a committee provides an ordinal ranking of a set
of ... projects. Anyone member's ranking is considered equal in importance to the ranking preferred
by any other member. The problem is to determine a compromise or consensus ranking that best
agrees with all the committee's ranking.
It has been shown (Massam, 1984; Rietveld, 1984) that the structure of this problem
is similar to the multi-criteria evaluation problem. These two problems can be
combined to form a multi-criteria social choice problem. This problem is an
elaboration in slightly more formal terms of the one offered earlier, and we can refer
to this as the generic planning problem (GPP).
Given a set of M plans, and for each an evaluation on a set of N criteria, for a set of G interest
groups, classify the M plans in such a way as to identify their relative attractiveness so that agreement
among the interest groups is maximized.
Rietveld (1984) indicates that in the case of the public choice problem nothing is
known about the relative importance of individuals (note that Cook and Seiford
assume equality is the norm), whereas for the evaluation problem some information
on the relative importance of criteria is generally available. In fact it appears to be
very difficult to derive any good formal definitions of satisfactory weights for criteria,
as interest groups clearly have their preferred weighting schemes which reflect their
preferences and priorities only in terms of the final outcome. A discussion on
strategies for handling the weighting problems for criteria is given in Chapter 2.
According to Sagar (1981) it can be shown that ex ante weighting of attributes and
goals has logical weaknesses which follow from assumptions of rational choice under
uncertainty. Who would willingly agree on a set of weights ex ante and risk that their
preferred alternative was replaced by another which they perceived to have
inappropriate consequences? It is probably preferable for the planning process to
include meaningful participation of interest groups in order to attempt to build a
consensus regarding a preferred plan. Conflicts cannot always be avoided, therefore
they must be managed. Amrhein (1985) has noted that "MCDM problems arise when
a decision-making process involves the simultaneous evaluation of multiple conflicting
objectives". Clearly criteria and objectives are closely related. Specifically we would
argue that planning goals (general statements of intent) be converted into objectives
(precisely defined areas of concern), and these in turn be formulated as clearly,
20
Progress in Planning
and Hatry (1972) offers a plea for careful measurement when he claims that "without
adequate measurement, so-called evaluations are likely to be little more than public
relations stories by the sponsors and of minimal practical use".
It is necessary to consider the merits of plans using criteria which clearly relate to
planning goals, and not to base the evaluation only on readily accessible data, but we
should not adopt superior attitudes and assume that numbers present some absolute
truth. At least let us recognize that they are open to different interpretation as regards
their significance and importance for a particular individual or group at a specific
point in time.
With all these general points behind us let us now begin to look specifically at the
elements of MCDM techniques as they can be used to tackle the generic planning
problem outlined at the end of this chapter.
CHAPTER 2
It is clear from an examination of the generic planning problem that was discussed
at the end of Chapter I that there are three major components which must be
considered. These components are: the alternate plans; the criteria which are used to
evaluate the plans; and the interest groups. We will begin this chapter by focusing on
these three components and consider the ways they can be combined as a set of
matrices. This will lead into a discussion on the scores and values in each matrix, and
a review of selected procedures for standardizing these scores. The rationale for such
an exercise will also be given. At this stage we will introduce the notion of benchmark
plans; also we will stress the need to recognize that errors are likely to occur in the
estimation of the scores.
One of the areas of concern for those using MCDM techniques rests with the
problem of deciding how to deal with the criteria. Should they be ordered from most
to least important? Should they be weighted? If so, then what rules should be used to
derive the order or to assign the weights? These and related questions focusing on the
treatment of criteria will be dealt with in a separate section of this chapter. Finally we
will offer a checklist of criteria which can be applied to assess the utility of a particular
technique. It should be noted that the emphasis in this chapter is on the general
principles of the techniques. In Chapter 3 we will turn our attention to consider
alternate strategies for analysing the information in the matrices. These strategies are
the MCDM techniques and we will select a variety of these to show the different ways
which can be used, remembering that the basic purpose of the exercise is to assist in
the process of generating, evaluating and classifying alternate plans so that a preferred
one can be identified. While we take as our starting point a finite set of alternate plans,
there is no reason why this set cannot be re-defined in the course of the planning
process, in fact this is probably a most desirable feature of a process in which
consensus-building and open participation are critical elements. The techniques we
refer to in Chapter 3 are far from an exhaustive list, however we argue that those
which are included are suitably representative. For example there are many MCDM
techniques which use an additive approach to deal with scores in order to derive a
utility value for a particular alternative plan. We will select examples from this body of
work rather than deal with each variation on this general theme. It should be noted
21
22
Progress in Planning
that we specifically exclude those techniques which are based upon formal
mathematical optimization procedures, for example, linear or goal-programming.
Such techniques fall into the MODM category and require explicit formulation of
constraints and objectives as equations prior to their application. In Chapter 4 we will
examine a set of case studies to show how the general principles laid out in Chapters 2
and 3 can be made operational.
23
P,
P2
Pi
II
PM
C,
C2
j Si
C
J
cN
I,
12
C,
c2
j SI
cj
cN
I,
12
II
I L
P,
P2
is I
Pi
PM
24
Progress in Planning
limits the incorporation of differing views on the magnitude and importance of the
impacts as well as hindering the debate on implementation of each proposed plan.
Such a debate usually considers an assessment of mitigation and compensation
Pi
PM
C1
C2
II
Cj
1L
is!
j I
CN
C1
PLANS
INTEREST GROUPS
CRITERIA
11 1 2 II I L
P1 P2
P;
C2
Cj
CN
P1
P2
Pi
PM
Cj
PM
25
PM
1 1 12
I(
measures. The single number approach has been referred to as a Grand Index method
by McAllister (1980) and he asserts that:
... the use of a grand index in evaluations should be abandoned. True, it does represent a neat
technical solution to the age-old evaluation dilemma of making trade-offs between the many, diverse
impacts of complex public actions [but] ... our knowledge of what determines the welfare of society
is entirely too weak to reduce it to a mathematical equation, which all grand index methods imply
can be done .... A major reason against the use of grand index schemes is that they short-circuit the
citizen participation process (p. 264).
'
Lootsma et al. (1986) offer a novel way to try to identify clusters of similar opinions
among the interest groups. A summary is given in Fig. 4. Pair-wise comparisons of
criteria by each interest group yield a set of prefe~ence scores which range from +4 to
-4 and these data are used in a cluster analysis to try to identify groupings among the
interest groups. The overall grouping which is indicated by the dotted line in Fig. 4
can be considered as an indication of the general preferences for the full set of
opinions. Formal measures to indicate the consistency of the opinions can be used and
while this may appear to enhance objectivity it is hardly likely that they will be of
great use in resolving conflicts among the interest groups. Lootsma et al. (1986)
conclude that "the main objective of our pilot exercise was to introduce to the
members of the Dutch Energy Research Council the use of multi-criterion decision
analysis as a discussion model that highlights the points of agreement and
disagreement amongst them, so that they can concentrate on the latter in order to
reach a compromise". Seen within this broader framework the method they propose
appears to be useful.
Let us now consider three other types of matrix for describing relationships among
the components. These are shown in Fig. 5 and they can be described as pair-wise
comparison matrices as the cells in each matrix refer to pairs of plans, criteria or
interest groups. We can envisage in the first case a series of such matrices for the
plans, each being offered by a particular interest group. The scores in such a matrix
are conventionally measures of similarity or difference. Saaty (1980) has done much to
promote this approach. These can be expressed as measures of dominance - more of
26
Progress in Planning
INTEREST GROUPS
FIRST CRITERION
1m
I,
SECOND CRITERION
as b
I
yy
DENDOGRAM
linking similar
profi les for
interest groups
I
score for interest group I. for pair-wise
comparison of criterion a with criterion b
The sca Ie for the score ranges from
o when a is similar to b. to
this later. A similar approach can be adopted to handle the second matrix which deals
with the criteria, again a series of matrices could be derived each to represent the views
of a particular interest group. In the case of the third matrix we have the problem of
deciding who should provide the scores? Hardly the interest groups, however if the
planner's client is a government agency and the planning problem is one of tackling a
public facility location problem, for example to find a suitable location for a wastedisposal unit, then the planner might seek to identify the possible coalitions and
conflicting groups of interested parties in such an issue. Possibly the governmental
agency is primarily concerned with building a consensus and therefore requires
information on the strength of opinions and estimates of the similarities and
differences of these opinions regarding siting options. The planner may be able to
identify shared concerns among the interest groups as well as points of sharp contrast
and confrontation. Such information can be of considerable importance to the
C,
c2
Cj
CN
I,
12
II
IL
27
C,
c2
I,
12
28
Progress in Planning
their particular objectives. As well, it is clear that the interest groups will seek to
understand the views of each other and to estimate ways to ensure particular
outcomes. It is the recognition of this that has led Radford (1986) to move away from
formal methods for analysing the matrices of the style presented in Figs 1,2 and 3 and
to seek a more general bargaining-negotiating framework to handle the conflicts. We
do not contest the validity of this, rather we argue that the MCDM techniques we
offer could possibly be incorporated by the interest groups into the strategic-tactical
model framework he has proposed. We do however strongly oppose the view that plan
evaluation and selection can be reduced to a simple matrix and this can be analysed
using a procedure such as PATTERN or a dendogram to yield the preferred result. It
is vital that the MCDM techniques be incorporated into the planning process.
This first stage which involves the organization of information as a set of matrices
provides a useful framework for handling the debate on the identification of options,
criteria and interest groups. Perhaps no further formal analysis of the information in
these matrices is undertaken in the plan evaluation process and the debate on the
relative merits of each option is undertaken within a bargaining framework in which
each of the interest groups seeks to present its case as forcefully as possible, while
avoiding unsatisfactory outcomes and recognizing that delay has a price. If, however,
detailed analysis of the options is seen to be an important part of the planning process
then we must look further at these matrices and consider how the scores in each one
can be derived and used.
29
costs and benefits. So overall, while we might try to standardize scores in order to
make the criteria commensurate, it would be naive to suppose that simple arithmetic
operation on the new numbers will yield the right or even acceptable solution. What
we might hope for perhaps is that the standardization procedure can yield values
which we can then use in a classification exercise and add to this a series of sensitivity
tests. It is this latter operation which is a vital part of using MCDM techniques.
The four measurement scales which can be used for the scores in the matrices are:
(i) ratio,
(ii) interval,
(iii) ordinal,
(iv) nominal.
The characteristics of each of these are discussed in standard texts and the details will
not be repeated here. See for example Voogd (1985).
The ratio scale contains most information, as a basic point of departure, an origin is
given and the magnitude of the measurement units; the nominal scale provides
categorical information, and as such it is hard to use for comparative purposes.
Recently there has developed a more concerted effort to develop a body of literature
on the analysis of categorical data, but as yet it has hardly been linked to the MCDM
techniques. A mistake frequently made in the analysis of a matrix is to assume that an
ordinal scale has the same properties as an interval scale. This point has been
forcefully made by Nowlan (1975) and the case is reviewed in Massam (1980). As a
general rule, whenever numbers are placed in a matrix a declaration of the scale
should be made so that all ambiguities are avoided and precision is stressed. Ordinal
scales are frequently used by planners and care must be taken to ensure that simple
arithmetic operations are not conducted using such scales. The rationale for this rests
on the notion that the ordinal scale does not give details of the magnitude of
differences; ratio and interval scales contain such information. We should also avoid
using geometrical approaches for analysing ordinal data, for example it is not
legitimate to draw a graph and claim that the data on the axes have only ordinal
properties. There have been attempts to introduce the idea of a fuzzy boundary which
defines the edge of a measurement scale (Siskos and Hubert, 1983) and Brans et al.
(1986) have attempted to examine systematically the intensity of preferences as they
relate to the comparison of pairs of plans for a particular criterion. This work is in
many ways a restatement of the assertion that a particular score may have errors
associated with it and it should be better characterised by upper and lower limits. This
is an important idea from a practical standpoint and one which should and can be
incorporated into MCDM techniques within the framework of sensitivity or
robustness tests.
The assignment of an impact score to a particular cell may lead us to believe that we
are objective and that our personal biases are removed. To some extent this is true,
however we should acknowledge that while the magnitude of an impact may be
assessed with some objectivity, depending upon the instrument, the importance of this
impact is directly related to the opinions, perceptions or the expectations of the
interest groups. A further complication is added for some impacts because the physical
intensity can be perceived differently from the psychological intensity, yet it may be
30
Progress in Planning
the former that produces damage to our health. For example, plan A may generate 10
decibels more noise on a sound meter than plan B, the physical energy is ten times as
great yet subjectively the difference only appears to be twice the amount. A 20 decibel
increase produces one hundred times more energy but only a four-fold increase in
perceived loudness. As ear damage is related to the physical intensity levels we should
be wary of using perceived intensities, yet it is these which cause psychological
annoyance. Great care is needed to examine the nature of the impacts; simple
recording of numbers without supplementary explanations may lead to spurious
conclusions. Yet again we stress the need to deal carefully with magnitudes and
importance as separate, but related issues. A plan to destroy 50 oak trees may be seen
as most unacceptable if these are the only trees in the region, on the other hand
removal of 50 such trees from a large forest may be of little consequence. A related
aspect concerns the marginal effects of increasing the level of the impacts. This point
underlies the rationale for using a function-form to transform basic scores, that is the
raw data, into standard scores. The term function-form is used by Brown and Valenti
(1983) and Hammond et al. (1980). Keeney and Nair (1977, Chap. 14) use the term
single-attribute function and Gardiner and Edwards (1975) refer to this as a value
curve. We will discuss this in Section 2.4 under the heading of a transformation
function.
In summary we. can envisage that the jS; values from matrix 1 on Fig. 1 are
presented as raw data, for example financial costs (long-term, short-term), numbers of
jobs etc. and ideally we would hope that interval data are available for each criterion.
With respect to the jSI and iSI scores perhaps the best we can hope for is a set of data
in the form of ranks. Conceptually it is possible to derive ratio values for jSI and is! by
asking each interest group to place each criterion or plan onto a scale ranging from 0
to 10, for example, or by the division of 100 points among the competing plans or
criteria. This is often referred to as Metfessel allocation (Solomon and Haynes, 1984;
Rowe and Pierce, 1982b).
In the case of the matrices in Fig. 5 the score in each cell refers to the degree of
similarity. For example we could use a ratio scale from 1 to 9 following the ideas of
Saaty (1980). A value of 1 indicates that the comparison of the two plans, criteria or
interest groups are identical, whereas a value of 9 indicates that one of the pair is
"absolutely more important", and following Saaty a score of 3 indicates "weakly more
important", 5 suggests "strongly more important" and 7 "very strongly more
important". Saaty has proposed that the similarity scores can be summarised in the
form of a positive reciprocal matrix of the style shown in Fig. 6. In this example the
plans are compared. Obviously the values on the principal diagonal are 1'so With
respect to the comparison between PI and P2 , a value of 9 indicates that PI is
"absolutely more important" than P2, the reciprocal value of 1/9 is placed in the cell
(P 2 , PI); plan i is "weakly more important" than PI and the reciprocal value of 1/3 is
placed in the cell (PI' Pi)' Plans 1 and M are identical, hence the values of 1 in cells
(PM, PI) and PI' PM)' If the information regarding the pair-wise comparisons can be
summarised in a positive reciprocal matrix then Saaty argued that an ordering of the
plans (referring to Fig. 6) could be derived. In technical language the principal
eigenvector is used to indicate the relative importance of the plans. Cook (1986a) has
P2
9
Pi
31
PM
1;3
Pi
suggested that there is room for improvement in this method and refers to the
technical contributions on the use of such eigenvector approaches by Cogger and Yu
(1985) and Johnson et al. (1979). From the planning perspective, while it seems
appropriate to seek values for the cells it is not clear about the level of confidence that
can be placed in the eigenvector results as a procedure for determining the relative
importance of the plans and hence an ordering. Research continues on this problem as
does other work by Cook et al. (1985) on binary matrices under the general topic of
tournaments. This work assumes that a 1 or a 0 in a cell indicates a clear preference
for a plan in a pair-wise comparison.
Another approach for describing similarities for the matrices shown in Fig. 5 is to
assign values to each cell which represent proximity measures, without any indication
of dominance among the pair-wise comparisons. For example using a matrix for plans
we might try to assign interval values ranging from 0 to 1 as indicators of similarity,
with a value of 0 indicating that the distance or difference between the two partners in
a pair is 0, hence the partners are identical. A value of 1 suggests maximum difference.
If correlation coefficients between the pairs of plans are used then a value of 1
indicates the pair is identical and as the value approaches 0 so the difference gets
larger. If similarities are described using this type of data then a symmetrical distance
matrix of the type shown in Fig. 7 is produced. This matrix refers to pair-wise
comparisons of plans and suggests that plans 1 and 2 are identical whereas i and 1 are
very different and the difference between M and 1 is less pronounced. If the data are
presented in this form then the matrix can be analysed using a multi-dimension scaling
method. A good survey of this literature is given by Golledge and Rushton (1972). The
end product of the analysis is a map of the plans showing their relative positions.
Examples of typical one- and two-dimension maps for five plans are shown in Fig. 8
and empirical examples in Massam (1980, Chap. 6).
In order to interpret such a map it is necessary to assign labels to the axes or the
ends of the scales, or to incorporate plans with known characteristics into the analysis,
such plans are referred to as benchmarks. The strategy of using such benchmarks is a
useful one whenever a classification exercise is undertaken. Details of benchmark
plans are given in a later section of this chapter and an example of a two-dimension
map is presented in Section 3.6 of Chapter 3.
32
Progress in Planning
P,
P2
P,
P2
Pi
PM
0.5
Pi
PM 0.5
ONE DIMENSION
P4
TWO DIMENSION
The basic reason for standardizing data rests with the argument that the
standardized values will give an indication of the underlying structure in the
information. For example we might choose to calculate Z scores given a set of
observations for a criterion for a set of plans with the view that we want to use some
underlying statistical properties in our analysis. This approach has been used. by Smith
(1977) for classifying census tracts using socio-economic variables with a view to
determining the social well-being of residents of each tract. However, such an
approach lacks validity unless we are confident that the basic distribution of values for
the particular criterion is normally distributed. Also, from a practical perspective if we
have only a few plans, for example five, then the exercise of calculating Z scores lacks
33
credibility. If we wish to combine values for a set of criteria then it 'is necessary that
the scales be commensurate and thus frequently raw data are standardized to achieve
this end. We should recognize however that there are many ways of standardizing raw
scores and in this section we will summarise the five basic approaches. Karski (1985)
has referred to the fourteen methods proposed by Langley (1974) in his work for the
Department of the Environment (U.K.) on evaluation techniques which could be used
in the preparation of structure plans. It appears that this large number is generated
from using slight variations on the maximum and minimum values used in some of the
calculations. The first four can be presented as simple formulae whereas the fifth is
somewhat different because the method of converting the raw scores depends upon the
shape of the transformation function. This will be discussed separately.
Let us consider a set of five plans (A, B, C, D, E) and a single criterion - jobs
created. The results of the four standardization procedures are given in Table 1.
TABLE 1. Results of four standardization procedures
Raw Data
PlanA
B
C
D
E
0.12
0.21
0.14
0.03
0.50
18
30
20
4
74
0.24
0.40
0.27
0.05
1.00
0.20
0.35
0.22
0.00
1.00
0.21
0.35
0.24
0.04
0.87
The procedures in Table 1 are summarised below. While we note that the relative
positions of the plans remain unchanged the differences between plans depend upon
the standardization procedure. This is shown in Fig. 9. Hence if the classification
procedure relies on this information then potentially the final results, namely the
ordering of the alternate plans, is subject to the particular standardization routine
which is adopted. Such an example is provided by the discordance index which is part
of ELECTRE, an overranking method developed in France. Comments on this are
provided in Section 3.6 of Chapter 3.
1.
Raw score
! all scores
2. Raw score
Max score
Raw score
-J !
JPP JO:I-C
(raw score)2
34
Progress in Planning
RAW DATA
Number of jobs
I
0
CA
1~
..
CA
..
.I
CA
t-I-------------------------t
D
CA
B
- - - - - - - - - - - - - - - - - +
- - - - - i4
+-4
CA
Number
74
of jobs
1.0
(a)
0"----------
1.0
0"------------
INCREASING
1.0
(c)
(b)
INCREASING
1.0
Cd)
oL---------.:::::::..2
INCREASING
35
36
Progress in Planning
annual population growth and average student-teacher ratio. Obviously each planning
exercise dictates the precise criteria which must be considered. Brown and Valenti
(1983) have argued that there are basically two ways to determine the shape of these
curves. First, by direct specification and second, as a result of questioning a
representative of an interest group in order to determine points of indifference on the
curves and hence the shape of the function. While theoretically appealing it is likely
that for practical purposes it is difficult to obtain meaningful results to hypothetical
trade-off questions. Keeney and Nair (1977) elaborate on the way of obtaining the
form of these curves for a set of criteria for a practical planning problem.
The matrices given in Fig. 1 and Fig. 2 which focus attention on the plans, begins by
identifying M feasible alternatives. This is the set {PI' P2 , Pi, PM}' The generic planning
problem seeks to find a classification of these plans in order to identify their relative
attractiveness and one way of doing this is to compare the alternatives to a given
standard. It is this standard which we can refer to as a benchmark plan. There are a
number of different ways that we can approach the problem by defining this
benchmark plan and the four most common ones are:
(i) status quo option plan,
(ii) ideal best plan,
(iii) hypothetical worst plan,
(iv) plan of minimum satisfaction.
To illustrate these let us consider an impact matrix of the style shown in Table 2
using two criteria and a set of six feasible alternatives. The characteristics of the four
benchmark plans are included to give a (10 X 2) matrix as presented in Table 2.
TABLE 2. Sample impact matrix with benchmarks
Costs 106 $
No. jobs created
C1
C2
PI
P2
PJ
p.
P,
P6
30
14
26
18
42
10
16
20
19
17
13
11
0
0
ii
iii
iv
16
20
42
10
35
15
The plan iv has been defined by a budget constraint of $35 million and the need to create at least 15 jobs.
The matrix in Table 2 suggests that while the status quo option involves no financial
outlay in the form of construction costs we also have no generation of employment.
The characteristics of the ideal best and hypothetical worst options are derived from
the existing data in the matrix, and the values for the minimum satisfaction plan are
defined by considering the available budget, the goals and objectives of the planning
exercise and taking into account norms and standards. Hence constraints and
expectations can be used to characterise this plan. We suggest that the exercise of
producing these benchmark plans or norms may help in the interpretation of the
values in the matrix. Certainly if we resort to the use of multi-dimension scaling, for
37
ii
In Hall's (1980) overview of some recent great planning disasters he pays particular
attention to the fact that underestimation of costs and overestimation of benefits are
the major factors which have contributed enormously to the disasters. Explicit
recognition of possible errors associated with the values in the matrices may help to
reduce the incidence of planning mistakes. Having said this we must try to face up to
the problem of determining the errors, and perhaps the most satisfactory approach is
by using a series of sensitivity tests each assuming a different amount of error. For
example we might produce a classification of the plans using information in a matrix
of the style shown in Fig. 1 under the assumption that the estimates for the impact
scores 08;) were perfectly accurate. Next we could modify these estimates by a given
amount and examine the new classification. Estimates for the errors could be offered
by technical analysis or examination of past experiences, another approach is to make
arbitrary adjustments in order to identify the maximum amount of error which would
be needed in order to change a classification radically. This type of approach looks at
the robustness of the classification. It might be the case that a small change in the
score for a particular criterion could alter the relative attractiveness of the plans quite
dramatically. In other cases it could be that considerable variations in the scores
consistently produce the same classification. This point is clearly illustrated with
respect to the analysis of an impact matrix which involved three interest groups, three
plans and five criteria (Massam, 1986) for a specific set of data regarding the
38
Progress in Planning
evaluation of alternate marina facilities for a lake near Ottawa, Canada. A variety of
different scores produced the same ordering of the plans. The message is clear, namely
whenever an MCDM technique is used to analyse a set of data it is necessary to
conduct a series of analyses to take account of possible error in the scores and values.
In the last section we made a case for sensitivity tests to deal with possible errors
associated with scores and impact values, and in this section we can continue the
argument by suggesting that we should also undertake sensitivity tests for the criteria.
At the outset of a planning problem we may be able to define a long list of criteria
which will encompass all the possible goals and objectives of the various interest
groups. However before we can apply a particular MCDM technique it may be
necessary to organize the criteria in a particular way. There are three basic strategies
for handling criteria. First, by assuming they are all of equal importance, hence the
length of the list determines those criteria that will be considered in the analysis.
Second, by ordering the criteria from most to least important, and third, by assigning
weights to the criteria to indicate their relative importance. A variation on the second
method is to divide the complete set of criteria into groups and possibly order these
groups according to their importance.
One method for determining the weights for criteria is to use Saaty's (1980) positive
reciprocal matrix approach from the set of pair-wise comparisons. Another method is
offered by Keeney and Nair (1977). They start by deriving an order for the criteria
using the opinions of an interest group and then they seek to identify the trade-offs
among hypothetical values for pairs of criteria to identify indifference levels. While
such an approach is logical and can yield values which allow weights to be assigned to
the criteria we should remind ourselves that we are asking interest groups or their
representatives to make a priori statements about hypothetical levels for criteria and as
such we might feel that the results of such questioning lack credibility. Roy and
Bouyssou (1986) have critically examined this approach and specifically they have reexamined the data used by Keeney and Nair using an overranking method (ELECTRE
III). While both approaches give the same best plan, the classifications of the set of the
nine alternatives using six criteria vary quite dramatically. Both procedures concur on
the worst two plans. Sagar's point about ex ante weights for criteria that was reported
in Chapter 1 is worth bearing in mind, for who would willingly participate in the
search for weights without knowing the outcome of this participation. In summary we
argue that alternate ways of treating criteria using the three basic strategies within a
general sensitivity framework provides probably the most balanced approach. We can
take this a step further and suggest that a series of scenarios be described each
characterised by a particular set of criteria or ordering of criteria. For each scenario
the planning options could be assessed and comparisons among the results made. It
may occur that different interest groups have unique orderings of criteria and different
combinations of criteria, yet in the final analysis some of the planning options appear
consistently as among the most preferred. Sensitivity tests may identify such patterns.
39
2.8. SUMMARY
CHAPTER 3
41
It should be noted that a unique solution is not guaranteed. In its basic form this
procedure does not allow any trade-off among criteria, in technical terms this is
sometimes referred to as a non-compensatory approach. If a single plan scores highly
and above all others on the most important criterion, then it would be identified as the
best and no matter how well other plans performed for the other criteria or how badly
this first plan performed for the other criteria, there could be no modification. The
method seeks to identify a single best plan. At first glance we might believe that a noncompensatory approach is generally an unacceptable one in planning as compromises
and trade-offs appear to characterise many planning choice problems. However, for
some planning problems precise explicit constraints do exist and options which do not
satisfy these have to be excluded. It might be the case that such an approach
determines that no feasible planning solution exists given a particular constraint and
this means that the search exercise either stops, or the constraint has to be relaxed.
Perhaps the classic example of this in current planning applies to the search for sites
for the storage of radioactive waste material. Possibly there is consensus that such sites
must be confined to geological areas with at least a certain thickness of a particular
type of rock, and yet be at least a minimum distance from densely populated areas
while not being further than a certain maximum distance from the sources which
generate the waste. Clearly other criteria are involved in this problem, but if we restrict
attention to just these three the feasible sites are likely to depend upon the precise
definitions of rock thickness, distance to populated areas and distance to source. Those
involved in the search for appropriate sites must decide if there are clear definitions of
these criteria which must be satisfied and if these criteria can be ordered
lexicographically, or if trade-offs are possible.
The cartographic sieve-mapping and overlay approaches of McHarg (1969), often
involve constraint mapping and as such are a type of LaM. Regions or districts may
be excluded from consideration as possible locations for a facility if they contain
archaeological remains for example, no matter how attractive they are with respect to
other criteria. Other similar approaches are provided by the conjunctive and
disjunctive models. While the former uses the principle that an alternative is rejected if
the score for a particular criterion does not reach a minimum standard, the latter
selects an alternative on the achievement of the highest score for a selected criterion.
Details of these models are given in Massam (1980), Hwang and Yoon (1981), and
Solomon and Haynes (1984). In general, in order to pose a planning problem in
lexicographic terms it is vital to have a clear strong consensus on the importance of a
small number of criteria which must be satisfied, prior to the consideration of other
criteria which may be traded one against another. Such consensus may be built around
health concerns, especially with respect to pollution levels, or around cultural values
and the protection of national heritage. The exigencies of economic development and
job creation may however militate against the formation of rigid consensus and lead to
a relaxation of the firm initial conditions attached to the criteria.
An alternative way of posing the lexicographic problem may be to try to order the
interest groups rather than the criteria, and to see if particular plans emerge which
satisfy the highly ranked target groups. For many planning problems target groups
can be defined as those which specifically should be helped by a plan, these groups
42
Progress in Planning
could head the list. In summary a lexicographic framework may assist in focusing a
debate on the ordering of the evaluation criteria and establishing the firmness of
opinions regarding critical levels for the criteria. However unfortunately, lexicographic
ordering does not produce a classification of all the feasible plans and not all the
information for the full set of criteria is necessarily used. In the process of comparing
plans for a particular criterion it is possible to take account of measurement errors
associated with the estimates for the impacts. For example, a threshold value can be
defined which must be exceeded before one plan is rated more attractive than another.
By altering the threshold values for each criterion we can judge if the same plan
consistently appears as the most attractive. This exercise of examining the stability of
the solution is a sensitivity test. A further type of sensitivity test can be undertaken by
changing the ordering of the criteria. An example of this is given in Massam and
Askew (1982) for a problem involving the evaluation of twenty-four policies using a
set of five criteria.
c,
P,
\
" \
oP4
\
\
0-
BEST
-t
P5 - - - -
---
- - oP3
BEST
FIG. 13. Graphical presentation of five plans using two criteria (raw scores).
43
is given in many economics and policy texts, for example, Stokey and Zeckhauser
(1978). Theoretically elegant and potentially able to handle many criteria within an
analytical framework this approach has considerable intellectual appeal, however
problems relating to the determination of the shape of indifference curves for practical
choice problems militate against its widespread adoption in plan evaluation and
selection. Also there are difficulties of incorporating several interest groups and
uncertainties regarding the selection of criteria and the determination of the predicted
impacts.
In Fig. 13 the benchmark is given by minimizing the values for the two criteria,
however, potentially three other combinations may be feasible for an ideal benchmark;
the full set of four benchmarks is listed below:
Benchmark
C1
C2
1
2
3
4
min
min
max
max
min
max
min
max
The selection of the efficient plans obviously depends upon the definition of the
benchmark and its position on the graph. The particular problem will dictate the
appropriate definition, we should also note the other types of benchmarks that could
be used as discussed in Chapter: 2. One further point should be stressed, namely that
prior to constructing a graph it is mandatory to ,have interval or ratio data.
At first glance at Fig. 13 we might believe that P 5 is the preferred plan as it appears
closest to the best benchmark. We cannot support this unless we know that the criteria
are measured on commensurate scales: If we wish to adopt this approach then we
could standardize the criteria scores and the benchmark scores, then measure the
distance between each plan and a benchmark and use these distance measures to
classify the alternatives. This approach has been called Technique for Ordered
Preference by Similarity to Ideal Solution (TOPSIS). It is discussed in some detail in
Hwang and Yoon (1981). The distance between a particular plan and a benchmark is
given by the general Minkowski distance measure:
where Pi is plan i, and Pb is the benchmark plan, and each plan is characterised by N
criteria. The value for criterion} for plan i is Cij and for the benchmark it is Cbj. The
scores obviously must be commensurate and hence the raw data have to be
standardized. It is also assumed that all the N criteria are equally important. Clearly
this list of criteria could be modified in the course of a set of sensitivity tests. If we set
p = 1 then the distance is sometimes called the rectangular distance or manhattan
metric, if p = 2 the straight-line distance is calculated. Modifications to this approach
44
Progress in Planning
could be made to reflect the fact that different interest groups may have different
definitions of the benchmark plan. Also alternate constrained plans could be defined
as benchmarks, following the ideas presented in Chapter 2. We should note that the
distance which we calculate is a dimensionless number and as such it is difficult to
interpret. However, if under a variety of alternate schemes, using different benchmarks
and different sets of criteria, some plans consistently head the list we can be fairly
confident they are among the best ones. Equally, if the same group of plans
consistently appears at the bottom of the list then these plans are clearly the least
attractive. Of course different scenarios may give rise to different groups of plans
appearing at the top of the list. To be aware of these conflicts may help in the
negotiation process which needs to consider the implementation of a plan. Potentially
a benchmark plan which is the easiest to implement could be defined and the
alternatives compared to this one. In summary, while it is most unlikely that TOPSIS
or one of the variations on this theme could be used to solve a planning problem, it is
possible that the general strategy of comparing alternatives to a benchmark may help
in the debate and search for a preferred plan. Szidarovszky and Duckstein (1986) draw
attention to the fact that "many of the multi objective decision making techniques
which have been developed make use of a geometrical definition of the 'best' solution.
By this phrase, it is meant that the decision maker would like either to minimize the
distance from some 'ideal' or 'goal' point or to maximize the distance from some
'pessimistic' or 'status quo' alternative". They develop a technique called RHOD and
a dynamic variation entitled DYRHOD and apply it within the framework of MODM
to a specific planning problem in the Bakony Transdanubian region of Hungary. The
problem concerns the selection of water management policies to allow bauxite mining
to proceed over a ten-year period. Three conflicting objectives involving mining, water
supply and recharge for the thermal spas of Budapest are identified. The approach
they offer complements the static one offered by TOPSIS.
The Minkowski distance measure defined earlier can be used to compare pairs of
planning options. However, because the measure is a dimensionless number it is hard
to know how to interpret its significance. One approach which can be used is to
consider that errors are associated with each criterion and then to calculate the
distance between pairs of plans in the presence and absence of such errors. It may
appear in the latter case that a particular pair of plans generate a distance measure
which becomes very small when errors are included. This clearly raises the necessity to
acknowledge that errors will almost certainly exist when values or scores are assigned
to impacts for the criteria.
In Chapter 2 brief comments were offered on the use of multi-dimension scaling and
its application to produce 'maps' of alternate plans. It was stressed that the
interpretation of such maps can be greatly assisted if benchmark plans are included
and more specifically the distance between the benchmark and each plan can be
calculated as in the TOPSIS approach. The attractiveness of the plans decreases as the
distance to the best benchmark increases. If the distance is calculated from a multidimension map, then the same general Minkowski distance formula as noted above
can be used, except that N refers to the number of dimensions of the map. For a twodimension map of the style shown in Fig. 12, N = 2, and if the straight-line distance is
45
Almost two hundred years ago Borda (1781) and Condorcet (1786) began the formal
search for a procedure to aggregate individual preferences for a set of alternatives into
a group consensus. While their concern was to consider a set of alternate candidates in
an electoral process the generality of their problem is immediately apparent for it lies
at the heart of Public Choice Theory, as was pointed out in Chapter 1. There have
been a number of useful reviews of this literature, for example, Fishburn (1971, 1974),
Richelson (1978) and Cook and Kress (1984). One of the most prolific contemporary
workers who is tackling the consensus problem using formal approaches is Cook and
we will draw on his results. In order to provide some organization to this literature we
will arrange the material under four headings which will allow us to move from the
earliest simplistic approaches to more recent analytical work which is based on an
axiomatic approach. We will also examine one of the basic algorithms and indicate
some problem areas. The four headings are:
(1) Borda-Kendall method.
(2) Toward an axiomatic approach.
(3) Cook and Sieford distance method.
(4) Some problem areas.
46
Progress in Planning
axioms of social choice theory as enunciated by Arrow (1951). One of Arrow's axioms
deals with pair-wise determination and this requires that society's ranking of any pair
of alternate plans depend only on the individuals' rankings of these two alternatives.
The wayan individual rates another alternative should not bias the comparison of the
two under consideration. This axiom is sometimes referred to as the irrelevant
alternative axiom. Unfortunately this axiom is not satisfied by the simple ranking
procedure that sums scores and determines the attractiveness of a plan on the basis of
the scores. This is illustrated below for three interest groups II> 12 and 13 and three .
plans PI' P2 and P 3 Let us examine the following preference structure:
1st
2nd
3rd
(priority order)
and the assignment of 3 points for 1st, 2 for 2nd and 1 for 3rd positions yields final
scores of 6 for PI> P2 and P3 ; suggesting that society's view is that all the plans are
equally attractive. However if we consider that for II the preference order is changed
to P 1>P 3>P 2 , from P I>P 2>P 3 , then the final scores are: PI = 6; P2 = 5; P 3 = 7. Hence
by only changing the positions of P2 and P3 for one individual we have effectively
altered the way in which s~ciety views PI and P2 , for this time PI is not equal to P2
Cook and Sieford (1982) note that the B-K method is least satisfactory if tied ranks
occur in the original set of preferences and they offer a modification called the
minimum variance (MV) approach which they argue is appropriate for handling ties.
In essence the MV approach uses the principle from statistics that the sample mean
minimizes the square of the deviations. This work represents an extension of the ideas
offered by Kemeny and Snell (1962) who proposed that the preference orders be
considered as a set of distance measures. Explicit in this approach is the notion that
the scale and the rankings have interval properties, and specifically the distance
between a pair of adjacent ranks is constant. Cook and Sieford (1978) note that:
The problem is to determine a compromise or consensus ranking that best agrees with all
the ... rankings [of the interest groups]. Implicit in this statement ... is the existence of a measure of
agreement or disagreement between rankings.
47
are given in a paper by Cook and Sieford (1978), though we should note that they
found that the axioms were consistent and they identified a distance function which is
unique. This is a most satisfactory theoretical result and we would hope that it
provides a firm basis for the solution of some practical problems, however the axioms
relating to the distanct measure are not exactly the same kinds of social choice axioms
which Arrow proposed, so we are left with only a partial solution to the consensus
problem.
Let us now consider the consensus problem more explicitly by examining the Cook
and Sieford distance measure and the sample data they used. They examine a matrix
of the style shown in Table 3.
TABLE 3. Ordered preference matrix: 10 interest groups: 5 plans [after Cook
and Sieford (1978, 1984)]
PI
P2
P,
p.
Ps
I,
12
I,
I.
Is
16
17
18
I.
110
I
4
3
5
2
4
3
1
5
2
3
2
5
4
1
1
4
2
5
3
2
5
4
1
3
1
4
3
2
5
4
1
5
3
2
1
3
4
2
5
3
1
5
4
2
3
4
2
1
5
The data in Table 3 can be converted into a square (5X5) distance matrix in the
following way. The distance, d ll , in the matrix is given by:
+ 13-11 = 13
and
d21 =14-11+13-11+12-11+
+14-11= 21
and
13
21
24
22
20
11
15
16
16
12
11
11
17
11
12
12
14
14
16
12
27
19
16
18
20
48
Progress in Planning
Table 4 can be used in an assignment algorithm to find the total distance associated
with any ordering of the five plans, and hence we can try to find the ordering which
gives the smallest distance. The distance matrix can be viewed as an assignment matrix
as shown in Table 5.
2
Plans
I
2
3
4
Priorities
3
*(13)
*(11)
*(12)
*(18)
*(12)
The order of the plans shown in Table 5 from most to least preferred is P" P s, P 2,
P3 , P4 and the total distance for this order is 66. This is the minimum value; however
this is not a unique order. The following order P" Ps, P 3 , P2 , P4 is also given by Cook
and Sieford as having a distance value of 66. The order P" Ps, P2 , P4 , P 3 is given using
concordance analysis (this approach is discussed in Section 3.6) and this too has a
value of 66. Any other order has a larger value therefore Cook and Sieford conclude
that the consensus or compromise ordering is given by the minimum distance
assignment. This is conventionally referred to as the median ranking.
In this section we will focus on just two problems. First we will consider the merits
of the median ranking and suggest that while it gives an optimum solution to the
distance-minimizing type of consensus problem it is less satisfactory if we examine the
results in terms of the opinions of the interest groups. This is an important practical
matter if we are concerned with finding the plan which is acceptable to a majority of
the interest groups. Second we will note that the results using a distance measure are
likely to be treated sceptically as the consensus measure is a dimensionless number and
the distance techniques appear to be at arm's length to a planning process which
typically involves estimation of preferences as merely starting points in a framework of
bargaining and negotiation.
The orders PI' Ps, P3 , P2 , P4 and PI' Ps, P2 , P4 , P 3 identified earlier contain the
following pair-wise comparisons as shown in Table 6.
49
PI> Ps
PI> P3
PI> P,
PI> p.
Ps > P3
Ps > P,
Ps > p.
P3 > P,
P3 > p.
P, > p.
6/10
8/10
6/10
7/10
SilO
SIlO
6/10
6/10
3/10
6/10
58
100
PI>P s
PI >P,
PI >p.
P I >P 3
Ps>P,
Ps>p.
PS >P3
P,>P.
P,>P 3
p.> P3
6/10
6/10
7/10
8/10
SilO
6/10
60
100
SIlO
6/10
4/10
7/10
To examine the data in Table 6 to see the extent to which they agree with the original
information in the matrix shown in Table 3, we need to consider the number of
interest groups who accept the proposed ordering given by the median ranking
method. These are shown as proportions in Table 6. It is clear that the order PI' P 5 , P 2 ,
P4 , P 3 agrees more closely with the original set of opinions. In fact out of the 100 pairwise comparisons 60% agree with this order while 58% agree with the alternate order
even though both orders have the same distance value of 66. The dilemma that this
poses for the practitioner is to decide if the distance approach should be used and
which of the two solutions should be recommended. If our interest is to identify the
top one, or two alternatives, the distance approach may be satisfactory, however, for a
complete ordering of the alternatives it seems to be less satisfactory than an approach
using concordance analysis.
The second point is to stress that the preference orders, as stated in Table 3, do not
reflect variations in strength of preferences by interest groups or firmness of opinions
or possible errors which interest groups are prepared to acknowledge. Before we could
use a distance-type approach to tackle a practical problem it would be necessary to
add these dimensions to the original data, as we must seek to build a dynamic plan
evaluation methodology and one which allows preferences to evolve as new
information becomes available, in the form of mitigation measures or incentives, for
example. The static rigid nature of the consensus problem formation is likely to
militate against its acceptance by those who are tackling planning problems, not least
of all because the assumption that all interest groups have the same importance is
rarely found. Finally we can note that for the data used by Cook and Sieford,
concordance analysis identifies the ordering of the five plans which not only has the
lowest distance value, but also has the highest level of agreement with the opinions of
the ten interest groups.
Perhaps the largest single collection of MCDM techniques are subsumed under this
heading. In general the additive models (often referred to as compensatory models),
seek to reduce the plan evaluation and selection problem to one in which each of the
JPP 30:1-D
50
Progress in Planning
alternate plans is eIassified using a single score which represents the attractiveness or
utility of a plan. The selection of a preferred plan is based upon these scores. Hwang
and Yoon (1981, p. 99) suggest that the "simple additive weighting (SAW) method is
probably the best known and very widely used method of MADM". While we might
contest the legitimacy of using a simple additive function for combining impacts in
order to obtain a single value for each alternate plan it has been argued by Hwang and
Yoon (1981, p. 103) that "theory, simulation computations, and experience all suggest
that the SAW method yields extremely close approximations to very much more
complicated non-linear forms, while remaining far easier to use and understand".
An interesting review of the errors which can be associated with the use of SAW
models is given by Rowe and Pierce (1982b). They use hypothetical data and introduce
errors of known types and magnitudes in an attempt to determine "in a general way
the sensitivity of the weighting summation decision model to some of the classes of
error to which it is subject". For those who rely on the use of SAW to tackle a plan
selection problem it is most important that clear recognition of the potential errors be
incorporated into the study, and specifically that sensitivity tests be run as part of the
analysis. Solomon and Haynes (1984) also conclude that while there are a variety of
models for accumulating impacts into a final score, "the use of the simple weighting
summation model is probably justified." It is this use of a formal single dimensionless
number ~hich causes considerable problems for anyone who wishes to use the results
and participate in a debate which involves compromises, trade-offs and a discussion of
impacts in terms of jobs, houses, costs, environmental damage and the like. The
aggregation of impacts across a wide range of variables is worrying to practising
planners. Some argue that the worth of a particular plan is not a simple additive
function of the worth of the various components or even a readily identifiable
multiplicative function, for that matter. This is stressed in a paper by Roy and
Bouyssou (1986). In philosophical terms we are reminded of the debate propounded
by Moore in Principia Ethica when he argued that "the worth of what he termed an
organic whole bears no regular proportion to the sum of the values of its parts". In
particular, "the value of a whole must not be assumed to be the same as the sum of the
values of its parts" (Rosenbaum, 1975, p. 127). Yet this concern does not seem to
hinder the development of the additive models. However, we can observe that there
appears to be limited dialogue between theoreticians and practitioners in this area.
One or two notable examples of cooperation should be mentioned, for example, the
work of Keeney (1972) and the examples provided by Edwards (1971), and in Chapter
1 we have already referred to Jaakson's (1984) work using PATTERN, one of the
simplest of the additive models.
We should further note that if we wish to defend the use of a simple additive model
the following two conditions should be satisfied. First, that preferences for, or the
trade-off for pairs of criteria, for example Cr and Cs should be preferentially
independent of fixed levels for any other criteria, for example Ct. Second, that Cr, for
example, should be utility independent of the other criteria. Intuitively these conditions
are appealing however very rarely are they verified prior to the application of an
additive model. One case of verification is given by Keeney and Nair (1977) in their
study on the use of a set of six criteria for evaluating nine alternate sites for a nuclear
51
power plant in the Pacific northwest of the U.S.A. They use the MAUT approach. The
verification of the two conditions depends upon the confidence we can attach to
opinions regarding the trade-offs among criteria at hypothetical levels. These types of
questions are obviously not easy to pose in manageable practical ways to the various
interest groups.
A caveat could be included here if we draw attention to multiplicative models which
explicitly seek to accumulate the scores for individual criteria in such a way as to take
account of cumulative effects which are greater than the sum of the individual criteria
values. An example of this work is given by Keeney and Nair (1977). They develop a
formal methodology to establish if a multiplicative model is appropriate, then to
establish the parameters for the model in a particular situation. The intellectual beauty
of this approach contrasts with its practicality, first in seeking the sympathy of
planners for its adoption, and second to elicit the appropriate answers to hypothetical
choice questions to determine the values for the parameters. Rowe and Pierce (1982a)
draw attention to the fact that those who are asked to describe indifference levels or
trade-offs may not have great confidence in their ability to make the necessary
judgements. They draw this conclusion after interviewing knowledgeable panel
members with respect to a specific power plant site selection project on Long Island,
New York. Let us recall Sagar's comments referred to earlier, namely that individuals
are unlikely to attach importance to a particular set of weights for individual criteria
without an appreciation of the significance of their opinions on the final outcome. It is
the outcome which is of prime importance to an interest group, not the parameter
values of a multiplicative or additive model or criteria weights per se. Roy and
Bouyssou (1986) have analysed the data of Keeney and Nair (1977) using a
concordance method (ELECTRE III). Of the nine available sites for the facility all
agree on the best one and the two least favoured, however the relative attractiveness of
the full set of alternatives depends upon the method used. Roy and Bouyssou (1986)
draw attention to the different underlying principles of MAUT and ELECTRE III,
specifically the nature of the compensatory features of the impacts on the set of six
criteria. The ELECTRE III method is only partially compensatory in contrast to
MAUT. It is concluded from this particular analysis that:
... these inevitable disagreements do not imply that decision-aid [MCDM techniques] is useless but
simply that a single problem may have several valid responses. Given that two different decision-aid
models cannot be implemented in the same decision process, the decision-maker must be conscious of
the qualitative choices implied by the different models - often conveying the analysts' own ethical
choices - before coming to personal conclusions on the choice to be made. In this domain, the many
different approaches reflect in our view the complexity of the researcher's task much more than a
scientific weakness (Roy and Bouyssou, 1986, p. 214).
The expected utility models developed in the field of MAUT fall into the additive
model category. The use of an expected value criterion (EVC) for judging the worth of
a particular plan can be criticised on the grounds that the alternate plans are presented
just once for evaluation. Radford (1980) puts the point succinctly as follows:
[EVC is] Generally agreed to be appropriate in decision problems that are repeated many times in
exactly the same form. Choice of the course of action that gives the greatest average benefit is
intuitively accepted under these circumstances. However, the use of EVC is less readily acceptable in
a single occurrence of a decision situation involving uncertainty.
52
Progress in Planning
53
:0Ql
.t:!
.s
UJ
I
C
<...>
en
>-
....
::;
I
D
f=
::::>
MINIMUM
(not a scale)
MEAN SCORES
FIG. 14. Typical examples of results of PROLIVAN: four alternate plans (A,B,C,D).
The basic procedure for most types of additive model can be summarised as a set of
six steps as follows:
STEP 1.
Define the set of plans PI ... PM
Define the set of interest groups II ... II
Define the set of criteria C I ... CN
STEP 2.
Obtain impact values for the criteria, for each plan and convert the raw
data into standardized values using function forms or other
normalisation procedure. The opinions of the interest groups can be
used to obtain the function forms.
STEP 3.
Determine the precise way in which the standardized values are
aggregated to give a final utility value for each plan. The opinions of the
interest groups are used to obtain the relative weights for the criteria.
STEP 4.
Determine the utility value for each plan.
STEP 5.
Undertake a series of sensitivity tests to examine the effects on the utility
values of altering in a systematic way:
(i) the alternate plans to be considered,
(ii) the set of criteria to be considered,
(iii) the accuracy of the raw data,
(iv) the function forms for each criterion,
(v) the weights for the criteria, and the function which is used to
accumulate the scores for the criteria into a final score for each
plan.
54
Progress in Planning
STEP 6.
Justify the sensitivity tests in the context of the planning exercise, write
reports and summarise results in terms of the distribution of costs and
benefits and present the perspectives of each of the interest groups.
Avoid using dimensionless numbers. Incorporate the results into the
larger planning process and repeat steps if necessary.
It should be remembered that this sequence is cyclical and while it appears rigid in
its formal presentation, in fact practical applications demand that it be rather loosely
defined so that the exercise might in fact begin at STEP 2 when a particular interest
group, acting as a proponent, brings forward a specific plan. While all the steps are
important it is STEP 6 which needs great care in order to establish the legitimacy and
credibility of the whole exercise. Without this step the earlier ones tend to operate in
vacuo.
In July 1966, Roy et al. presented a paper in Rome at a study session on Methods of
Calculation in the Social Sciences. At the time they were associated with SEMAMETRA in Paris, France and they were seeking to use their formal training in
mathematics to develop tools that could be used to tackle practical choice decisionmaking problems. Their work was based upon a systematic analysis of the relationship
between all pairs of a finite set of options using information about the impact scores
for each option on a set of criteria. Data on the relative importance of the criteria were
needed in the formal procedure they developed. The procedure sought to measure the
dominance of one option over another with a view to identifying the option which
outranked the others. The terms 'outranking' or 'concordance methods' are now
generally used to refer to many of the techniques which have grown from this seminal
work. The method they offered was christened ELECTRE and Roy et al. (1966)
referred to it as Elimination and Choice Translating Reality. It was based upon the
derivation and manipulation of two indices, first, a concordance index and second, a
discordance index. Details of these will be given later in this section. A formal
statement of the first version of ELECTRE I was given by Roy (1968) and one of the
most recent versions, referred to as ELECTRE III, is discussed in Roy (1978) and Roy
and Bouyssou (1986).
This technique enjoys considerable popularity in France, especially by those who
have been associated with SEMA-METRA, and also by an important group of
researchers in Holland, notably Nijkamp, Rietveld, Vincke and Voogd. Vincke (1986)
draws attention to "the large variety of outranking methods, depending on the type of
information they use and provide: criteria with or without indifference and/or
preference thresholds, ordinal or cardinal criteria, numerical weights or qualitative
indicators for the relative importance of the criteria, deterministic or fuzzy outranking
relations, choice of the best action or ranking ... " (p. 164).
The diffusion of ELECTRE to other parts of the world has been slow. Hwang and
Yoon (1981) consider "the ELECTRE method to be one of the best methods because
55
of its simple logic, full utilization of information contained in the decision matrix, and
refined computational procedure". This optimistic view is tempered by the opinions of
Cook et al. (1986) who conclude that "while concordance methods do provide a
valuable framework within which to examine multicriteria problems, the subjective
nature of project ratings, weights and thresholds represent serious concerns from a
reliability standpoint".
Roy and Bouyssou (1986) admit that ELECTRE "has no axiomatic basis, and
consequently it is often difficult to interpret certain parameters used in it". They add
that "only considerations based on common sense allow the decision-maker and the
analyst to give them a numerical value". The role of intuition and professional
judgement is clearly recognized. Vincke (1986) notes that "there is no doubt that there
is a lack of basic theory for this approach and, in our opinion, this is one of the
reasons for the reservations expressed by some theoreticians (in United States and
Europe) ... ".
Let us summarise the basic features of the ELECTRE method and the concordance
and discordance indices which are the fundamental elements of the procedure.
The method begins with a matrix of alternate plans, a set of criteria and a set of
weights for the criteria. Generally it is assumed that the impact scores in such a matrix
are ratio data, hence the magnitude of differences for scores for pairs of plans, with
respect to a particular criterion, can be used in the calculations. If the raw data are
standardized and these new values used, then the discordance index can depend on the
standardization procedure which is adopted. It was shown in Section 2.4 that different
scales result from the choice of standardization procedure. The concordance index for
a pair of plans i and i' was initially defined in the following way:
Cii'=
Sum of the weights for those criteria when plan i is rated equal to
or better than plan i'
Sum of all the weights for the full set of criteria
56
Progress in Planning
say that plan e is the best one, however for practical examples we find that a plan with
these ideal concordance and discordance scores does not exist. Therefore it is
necessary to introduce threshold values for the two indicators in order to find one of
the plans which is defined as the best. The selection of threshold values is an arbitrary
business as has been noted earlier and not based upon any theoretical or practical
grounds. Guigou (1971) suggests that, "By successive trials, and by diminishing the
strictness threshold from p= 1 [the optimum concordance value] and q=O [the optimum
discordance value], one finally eliminates and selects ... [a best alternative]." This
approach does not give an overall rating or ranking of all the alternate plans which
would allow the relative attractiveness of each to be judged.
A wide variety of modifications to the formula for calculating the two indicators
have been made and useful summaries are given in Rietveld (1980), Vincke (1986),
Brans et al. (1986) and Voogd (1983), however the arbitrary nature of the final
selection procedure remains if we have to rely on threshold values.
Massam (1980) has attempted to focus attention on the concordance indicator and
to redefine this so that:
Cit
In ELECTRE I the numerator was defined by including the weights for those plans in
which i was equal to and better than i'. This new index gives a set of values for Cit
which range from 1.0 when plan i is preferred to plan i', for all criteria, to 0 when plan
i is never preferred to i'. Also in this case Cit + Cn = 1.0. A summary of the pair-wise
comparisons for the complete set of alternate plans can be presented in the form of a
concordance matrix as shown in Table 7.
PI
P2
P3
P4
Ps
P6
PI
P2
P3
P4
Ps
P6
Row sum
Rank
X
0.75
0.50
0.75
0.75
0.875
0.25
X
0.50
0.00
0.25
0.75
0.50
0.50
X
0.25
0.75
0.875
0.25
1.00
0.75
X
1.00
1.00
0.25
0.75
0.25
0.00
X
1.00
0.125
0.25
0.125
0.00
0.00
X
1.375
3.25
2.125
1.00
2.75
4.50
5th
2nd
4th
6th
3rd
BEST
The values in the matrix in Table 7 were derived from the hypothetical plan evaluation
matrix shown in Table 8.
57
TABLE 8. Hypothetical plan evaluation matrix: Six plans and four criteria
Ct
C2
C3
C.
P3
p.
Ps
P6
7
6
15
10
8
5
24
12
9
3
26
14
PI
P2
3
7
26
18
14
19
11
Wt.
(max)
(min)
(max)
(max)
~
0.25
0.25
0.25
0.25
= 1.00
Two of the calculations for concordance values in Table 7 are given below for
illustrative purposes.
Cp , P, =
> Pz for C ?
(B) : is PI > Pz for C z?
(C) : is PI > Pz for C)
(D) : is PI > Pz for C4?
(A) : is PI
Cp , p.
> P6 for C I?
(B) : is PI > P for Cz?
(C) : is PI > P6 for C 3?
(A) : is PI
NO
__ 0+0+0.125+0
1.0
= 0.125
The values in the concordance matrix in Table 8 assume that the four criteria are
equally important, hence if the sum of all the weights is unity, the weight for each
criterion is 0.25. We could consider that this set of weights represents a view of a
particular interest group, and alternate weighting schemes could be introduced to
reflect different preference structures. Further we could consider that among the set of
six plans, PI is the status quo option and P6 is a benchmark option which is
characterised by ideal scores for each criterion. We can extend the calculation of the
C;;, scores further by introducing a just noticeable difference (J.N.D.) value, that is a
threshold difference between ratings for two alternatives, for each criterion, which
must be exceeded before one of the options is declared superior to another. This
modification allows account to be taken of possible errors in predicting the impact
scores.
In 1971 Guigou summarized the material in Roy's paper of 1968, but the name was
changed to ELECTRA. Ali et al. (1986) and Cook et al. (1986) extend the model and
refer to the version entitled SELECTRA which has been developed at the Ministry of
Transportation and Communications, Ontario, Canada, to tackle project ranking
58
Progress in Planning
problems of the style addressed by Roy and Bouyssou (1986). The specific planning
problem addressed in Ontario is to produce an ordering of a long list of possible
projects taking into account the opinions of a group of officials from different
departments. They use the concordance matrix to produce a binary matrix for the
alternatives of the style shown in Table 9.
TABLE 9. Sample binary matrix derived from a
concordance matrix
PI
P2
Pi
PN
PI
P2
Pi
1
X
PN
X
X
A value of 1 in cell P IP2 in Table 9 indicates that PI 'beats' P2, a zero is recorded in
the P2PI cell. While the concordance matrix gives an indication of the strength of the
dominance of one plan over another, the conversion to l's and O's loses this
information. Cook and his colleagues analyse the binary matrix using theoretical work
on tournaments, the matrix representing the outcomes of a tournament. The final
result is an ordering of the options. Perhaps the greatest criticism of this approach
rests with the reliance placed upon the binary information when in fact the strength of
the 'defeat' in a tournament is given by the full concordance matrix. Why give up this
useful information when it is readily available? Also we could adjust the way in which
pairs of plans are compared. For example, Brans et al. (1986) examine six different
relationships between pairs of plans for a particular criterion and they offer a set of
Preference Ranking Organization for Enrichment Evaluations methods which they
refer to as PROMETHEE. Basically they provide a series of different functions to
describe the intensity with which a decision maker compares pairs of options using
information for a particular criterion, specifically the functions define the area of
indifference between two options for a criterion. Using this approach the concordance
index is re-defined to include a set of parameters, one for each criterion, which
indicate the preferences. A comparison of ELECTRE III and PROMETHEE for a
location problem involving six criteria and six plans is described by Brans et al. (1986),
and they conclude that their modifications give more stable final results than those
offered by ELECTRE III. The significance of this assertion as it relates to the
estimation of the impact scores and the criteria weights remains to be examined, as
does the assessment of PROMETHEE by those who are actually involved in judging
the alternate plans and seeking one for implementation. Much practical work remains
to be done on this topic.
Nijkamp (1982) has developed an outranking procedure which can accommodate
ordinal impact scores. In this procedure, which he called a REGIME method, if Pi is
superior or equivalent to Pi' for a particular criterion then a discordance value of + 1 is
assigned. If Pi is inferior to Pi' for this criterion a value of -1 is given. The set of all
59
1.375
P3
3.25
2.125
P4
1.0
P s 2.75
P6 4.5
5th
2nd
4th
6th
3rd
BEST (hypothetical ideal: benchmark)
This yields the following ordering of the plans, and as an ideal option P6 is included,
the relative merits of the others can be judged in terms of their achievement of ideal
scores. We should however note that the row sum values are dimensionless numbers
and hence the characteristics of the plans in the precise terms of the initial criteria have
been disguised. As part of a meaningful planning process it is important to focus on
the actual criteria values and not depend upon the presentation of dimensionless
numbers in an effort to seek approval or support for a preferred option.
Another way of analysing a concordance matrix is to use a multi-dimension scaling
approach. If this strategy is used then it is important to include at least one benchmark
plan in the analysis in order to interpret the results. This point has been stressed in
Chapter 2. An example of this approach for a practical route selection problem using
seven criteria and eight options, as well as different weighting schemes for the criteria,
is given in Massam (1980).
The two-dimension map of the eight options is shown in Fig. 15 as well as two
benchmark plans, 9 (best) and 10 (worst). Using these two plans an attractiveness scale
of the options was derived by calculating the ratio diw/ dib , where
diw
is the straight-line distance between plan i and the worst (10)
dib
is the straight-line distance between plan i and the best (9)
This is similar to the TOPSIS approach discussed earlier. The ordering is shown in
Table 10.
60
Progress in Planning
-6
1 -4
3-
-s
9
- BEST
-2
-5
7-
diw/dib
8
6
2
4
I
3
5
7
1.65
1.33
0,78
0,78
0,73
0.67
0.59
0.41
For this example it was assumed that all criteria were equally important and that the
J.N.D. threshold was zero, thus indicating that no measurement errors were made. A
fuller discussion on the use of multi-dimension scaling and concordance analysis is
given in Massam (1980; Chap. 6).
3.7. CONCLUSIONS
Two general classes of models can be identified for dealing with MCDM problems,
they are usually referred to as compensatory and non-compensatory. The former rely
on the principle that trade-offs are legitimate so that the method seeks to identify
these, and then determines a utility score for each planning option. The classification
of the plans is based upon these scores. MAUT and the additive models fall into this
61
category and comments on these are included in this chapter. Among the noncompensatory models we can include lexicographic approaches, conjunctive and
disjunctive methods as well as concordance methods.
In a series of recent papers concordance analyses and ELECTRE III have been
compared with a wide variety of multi-criteria techniques using empirical and
hypothetical data sets. A summary of a selection of this work is given in Table 11, and
overall we suggest that concordance approaches go a long way towards satisfying the
questions we listed at the end of Chapter 2 for judging the utility of a particular
MCDM technique.
TABLE 11. Comparison of concordance analysis and other MCDM techniques: selected examples
Author/date
Problem/data
MCDM techniques
Rivett (1977)
Massam and Askew (1981)
I. Structural mapping - a
multi-dimension scaling
approach
2. Utility scores - additive
approach using standardized
scores
3. Lexicographic ordering
4. Factor analysis leading to a
graphical presentation
I. Additive utility approaches
Jaakson (1984)
Massam (1986)
I. PATTERN: an additive
I. MAUT
2. ELECTRE III
I. PROMETHEE
2. ELECTRE III
I. ELECTRE III
62
Progress in Planning
The five approaches to MCDM problems that have been presented in this chapter
provide a broad review of a growing field. Obviously each particular planning problem
will dictate the nature of the alternate plans, the ease with which interest groups can be
identified and the feasibility of using particular criteria for evaluation purposes.
Attention must be given to the justification of the impact scores and potential errors.
Without a clear understanding of these elements there is little justification in pursuing
a particular MCDM technique. However, a judicious use of one or more of the types
of techniques outlined in this chapter may help to expedite the planning process and
assist in clarification of issues relating to effectiveness, efficiency and the equity of
planning options within the context of a dynamic open planning process.
CHAPTER 4
64
Progress in Planning
make a plea for a series of such studies as a major thrust of research activities, for
example, in institutes of higher education which have an interest in the study of
planning processes. The third approach, and the one adopted in this chapter is to refer
to the Generic Planning Problem that we offered in Chapter 1 and identify some major
types of practical planning problems which are encompassed by this. Three are
identified and we will comment briefly on each of these under the following headings:
Site selection problems.
Route alignment problems.
Priority rating problems.
We suggest that these problems cover a full range of planning issues which are
currently being addressed in urban and rural areas of both the developed and
developing world. Further, the problems ~re not restricted to states or other
jurisdictions with particular types of political organization. This is not to say that
differences among jurisdictions regarding the nature of the debate on options are not
well pronounced. A centralised strong political system is unlikely to engage in a
protracted public debate on the merits of a number of different options in order to
provide an environment for consensus. This point has been made very strongly by
Ambrose (1986) in his appraisal of the current situation in the United Kingdom which
he argues is highly centralised and has moved some way from a consensual system.
Another jurisdiction may place much emphasis on the role of public hearings,
participation by interest groups, as well as the encouragement of a full range of
technical studies in order to move towards a consensual solution to a controversial
planning problem. For example, in Ontario, Canada, the Ontario Waste Management
Corporation was created in 1981 with "a primary responsibility to design, construct
and operate a province-wide system for the treatment and disposal of liquid industrial
and hazardous waste, and to develop a long-term program to assist in the reduction
and recycling of such wastes" (Annual Report, OWMC, Ontario, p. 1, 1984-85).
While the Act creating the OWMC does not explicitly mention consensus,
observation of the planning process engaged in by this Crown Corporation, especially
with respect to its search for an appropriate site or sites for a major waste treatment
facility, would suggest that participation of interest groups and incorporation of their
views into the decision-making process is a key element of the process.
Let us now turn to the three problems mentioned earlier. The first two are usually
fairly well-defined and explicit, for example, the task may be to find a new location for
a public facility such as a fire station, a clinic, a school or a waste disposal plant, or
within the context of the private sector a new location for a retail store, a warehouse, a
manufacturing facility, a research laboratory or a residential development. The route
alignment problem can cover a number of transportation and communication
examples, including highways, rail routes, pipelines and power-line corridors. Seen
within a narrow context each of these problems can potentially be defined as an
evaluation exercise of a set of feasible alternatives compared to the status quo.
However, within a broader context we must include the generation of the options, the
actions of interest groups over time, the influence of the bureaucracy or agency that is
handling the planning exercise and the social, political and economic climate of the
jurisdiction within which the problem is being addressed. Hence the need for detailed
65
case studies as we argued earlier. However, at this time we suggest that MCDM
techniques can be usefully applied to these problems as long as they are incorporated
into the planning process. They should not be seen as technical devices to give the
right answer, rather as tools to assist in the organization of a meaningful debate
among interest groups regarding the definition of options, the assessment and
evaluation of alternate actions and the selection, implementation and monitoring of a
preferred option.
The third type of problem seeks to extend the debate on plan evaluation to include
the situation in which the purpose of the exercise is to define a set of different projects
and to generate a priority listing. This listing could be used to guide expenditure
decisions over a given period of time. It is this general problem which will be
addressed in Section 4.4. Two examples for each type of problem will be offered. Also
for each example we will provide a statement of the planning problem in such a way
that an MCDM technique can be applied. A summary of the results of the application
of the technique will be presented, and we will offer a series of suggestions in order
that the formal analysis, based on the MCDM technique, can be incorporated into the
broader planning and decision-making environment. If the suggestions offered are not
considered then we argue that the credibility and legitimacy of using the MCDM
technique may not be clearly established.
In this section we will consider first the problem of selecting locations for primary
health care centres in a region in Zambia, and second, the search for suitable locations
for new fire stations in the City of North York, Ontario, Canada. This work draws on
the studies by Massam (1981) and Massam et al. (1986).
The emphasis in this problem is on the identification of rural health centres which
could be upgraded to serve as centres for the provision of more sophisticated medical
services. The evaluation exercise concentrates on measuring the accessibility of the
existing centres to the dispersed rural population and seeking to identify the centre
which has maximum accessibility as the one which should be upgraded. Data for the
Mpika District in the Northern Province and Seskeke District in the Western Province
were used. For each District the problem was stated in the following terms.
Given a set of existing centres, classify their attractiveness using a set of accessibility indicators and
find the most appropriate centre for upgrading.
66
Progress in Planning
(4)
The population within 12 kms of a centre.
(5)
The population within 30 kms of a centre.
(6)
The distance to the next nearest centre.
The distances were calculated using a base map which allocated the total population
of each District (Mpika 44,500; Seskeke 47,500) to 79 and 84 points respectively. It
was argued that the first three indicators offer measures of the effectiveness and the
equity of each centre. The next two indicators refer to the catchment areas, and
generally the larger the better. The last indicator attempts to assess the overall
distribution of the centres throughout the District, and it can be argued that a
dispersed pattern probably enhances the equity of the distribution, given that the
population is distributed fairly evenly. Impact matrices were prepared for each District
to show the alternate sites and the indicators. Benchmark options were defined as
follows:
Indicator
Benchmark
minimum
mInImUm
minimum
maximum
maximum
maximum
2
3
4
5
6
1
2
3
4
5
6
7
8
Indicators included
Weighting scheme
All
Exclude
Exclude
Exclude
All
Exclude
All
Exclude
No.6
No.4
No.6 and No.4
No.6
No.6
The data were analysed using concordance analysis and the attractiveness of each
centre was determined using the row sum of the concordance matrix. The results for
the experiments indicated considerable stability for the attractiveness scales for the
centre which had the highest scores, and on the basis of this analysis it is very clear
which is the most appropriate centre to upgrade in each District. Now let us turn to
the matter of incorporating these results into the broader health-care planning
framework in Zambia and to do this we will list a series of critical elements which the
analysts, planners and decision-makers will probably have to address. These elements
will not be elaborated as the purpose at this stage is to identify the sorts of issues
67
which have to be considered. We suggest that the results of the multi-criteria analysis
provide a point of reference for discussions on these critical elements. The analysis
points to specific centres which could be upgraded on the basis of an assessment of
accessibility, if other centres are to be selected then a trade-off will have been made.
Hence when the elements are incorporated into the planning process a decision will
have to be made regarding their impact on the locational decision. Some elements may
have a direct impact, for example, proximity to a bus route, main highway or
development zone while other elements may have little locational consequence. This
may be the case if external funding is to be provided by an agency and their mandate
is to offer untied aid. Tied aid may make it mandatory that investments be
concentrated in a particular part of the District. A list of the most obvious elements is
given below, they have not been ordered by importance.
(I) Who will provide the capital and the operating funds?
(2) How important is it to take advantage of economies of scale and concentrate
investments in large places?
(3) How important is it to ensure an equitable pattern?
(4) What are the likely effects of a choice of centre on utilization patterns,
supplies of goods and materials and availability of personnel?
(5) Are there season variations in migration, agriculture etc. which may influence
utilization patterns?
(6) What are the long and middle-term plans for the development of an
hierarchical referral health-care system?
(7) What plans exist for rural development, marketing, and the general provision
of services to rural dwellers?
(8) How important is it to involve local groups in the selection process and the
implementation of a programme of upgrading?
(9) In what ways are the use of the centres influenced by the practises involving
traditional medicines, and in what ways can the centres complement existing
values regarding sickness and health-care?
A consideration of these questions together with the multi-criteria analysis will
surely help in the search for effective policies to use scarce resources and to improve
the health of citizens in Zambia. Obviously this list could be extended to broaden the
debate into a consideration of the relative merits of investing in the health-care sector
as opposed to any other sector such as rural development, transportation or defence.
Such issues would be on the agenda of a central political authority.
There is an impressive body of literature which deals with the location of urban fire
stations and much of this work is based on the view that the ideal site is the one that
minimized the average response time, (i) to all potential demand points. Cast within
this framework we might argue that the location problem is based on a single criterion
and this has given rise to the extensive use of linear programming models to search for
the location that minimizes f. Other closely-related models seek to minimize the
68
Progress in Planning
maximum response time (lmax), arguing that in the interests of improved equity,
attempts should be made to keep this value below a given threshold and those who live
furthest from a station deserve to have priority for any improvements. The specific
problem that was posed in North York, a city of over 550,000 people covering an area
approximately rectangular in shape, of about 13 miles by 6 miles on the northern part
of Metropolitan Toronto, Ontario, was to examine locational patterns of stations. This
examination involved an assessment of the evolution of the patterns as the City has
grown over the last quarter century, an evaluation of the average response times, and
maximum values, and an appraisal of the existing stations and the search for sites for
the next two or three. The Fire Chief has offered specific suggestions for new sites and
relocation of existing ones, and these were evaluated. A set of five possible locations
was identified as potential sites for new fire stations, these sites were selected using the
results of linear programming models, a set of experiments using the intuitive
judgement of students versed in urban planning issues, and the recommendations of
the Fire Chief. Given that the City was particularly interested to identify just three
locations the multi-criteria problem was posed in the following way.
Given a set of five possible locations, evaluate all the ten combinations of three locations, using
measures of accessibility to different types of potential fire hazard, based on land use categories, to
find the set of three locations which have maximum accessibility.
The land use types that were used for the study were industrial, residential and
commercial.
Two accessibility measures from the linear programming model were used to
evaluate the ten alternate locations with respect to residential land use. Also two
distance measures were used for each of the other two land use categories. The first
measure was to the nearest parcel of land, and the second to that parcel which was
further from a fire station. These data were summarised as a 10 by 6 impact matrix. It
was decided to conduct a series of sensitivity experiments to take into account
different weighting schemes for the criteria as well as experiments which focused upon
possible measurement errors for the criteria.
Concordance analysis was used to classify the ten alternate planning options. The
one that appeared to be the best was only slightly different from the one proposed by
the Fire Chief and substance has been added to his recommendation.
Let us view the problem in its broader context and if we do this then a number of
important points must be made. First, the marginal improvement in Tor (max, by
adding three more stations is very small, and we have to address the issue of tradeoffs. Specifically, could fire losses be reduced by adding new equipment, improved
training, better public awareness programmes, or the use of smoke-detectors or trafficflow systems? Why should the focus be on more fire stations? Also, for the City
Council to approve the capital and operating budget they must be convinced of
improvements to the service and especially to perceived improvements particularly for
low probability, high danger fires. This is especially true for one or two institutions
which could possibly suffer considerable loss of life if a fire occurred.
One part of the sensitivity analysis focuses on an evaluation of some randomly
selected and peripheral sites for the new stations. This work showed that while the
69
aggregate statistic Tis hardly sensitive to location, t max varies more markedly. Overall
we suggest that as with the location problem in Zambia, the organization of the site
selection exercise in the context of a set of impact matrices and analysis. using
concordance indices for a set of experiments, provides a useful structure within which
a full debate on trade-offs could take place.
Two approaches to the general problem of finding the best route for a highway will
be offered in this section. The first uses data described by Marchet (1980) for a
motorway location problem in France. The data have been analysed by Marchet and
Siskos (1979) and Massam (1982). The second problem is an evaluation of eight
alignments for an interstate highway in Georgia, U.S.A. Details of the study are given
in Odum et al. (1976) and Massam and Skelton (1986). Whereas the second study
begins with a small set of feasible routes, part of the first study is involved with the
generation and then the evaluation of alternatives.
The problem that the analysts faced was to find a suitable route for a new motorway
between two towns. The suitability of alternate alignments was to be assessed in terms
of the impact on a set of environmental criteria. For the purposes of the exercise the
study area through which the route was to pass was divided into 58 zones, and each
zone was assumed to be homogeneous with respect to the impact on a set of four
criteria. For each zone an interval impact score ranging from 1 to 6 was assigned for
each criterion. A value of 1 indicates the minimum attractiveness for a route. Two
strategies were identified for tackling this problem. The first, discussed by Marchet
and Siskos (1979) attempts to evaluate eight alternate alignments that were provided
by a group of decision-makers. The second approach, that was developed by Massam
(1982), seeks to generate alternate routes given the basic impact matrix of 58 zones and
4 criteria. Further, this approach seeks to compare the eight routes with those selected
from all possible routes joining the two towns.
In terms of the GPP the first problem can be stated as follows:
Given eight alignments and the 58 by 4 impact matrix, evaluate each alignment in terms of the total
impact and find the one which has the least damaging total impact.
.
Marchet and Siskos (1979) used function forms derived from interviews to standardize
the impact scores, then a weighted additive utility model was employed to calculate a
dimensionless score for each of the eight routes. The weights that were used were
derived from an interactive process between the analysts and the decision-makers.
Sensitivity tests were conducted to assess the effects on the final scores that occurred
when slightly different weights were used. The overall process involved close
interaction between decision-makers and formal analysts.
70
Progress in Planning
The other approach that was developed to generate alternate alignments set the
problem in the following terms.
Given an impact matrix of 58 zones and 4 criteria classify the zones in terms of their attractiveness
for a route, and find the combination of zones such that the total attractiveness is maximized.
In order to tackle this problem an ideal zone was defined as having maximum
attractiveness for each criterion. The new impact matrix 58 plus benchmark and 4
criteria was analysed using concordance analysis under the assumption that each
criterion had the same weight. The concordance matrix that resulted from this work
was further analysed using a multi-dimension scaling algorithm to give a twodimension map of the zones. Using an approach similar to TOPSIS the distance
between each zone and the ideal benchmark zone was measured. Using these distances
the next task was to evaluate all combinations of distances for a set of zones that gave
a continuous path between the two towns. The combination which generated the
lowest total distance is the one that is comprised of zones that are most similar to the
benchmark zone, and thus it can be argued that this represents the most attractive
alignment.
Both of these approaches rely on sophisticated analytical procedures and as such
cause problems when the results are being discussed with the public and with decisionmakers who are unlikely to be closely familiar with the techniques. However, the
second approach produced an alignment which had not previously been considered by
the decision-makers. Further, this alignment, when compared with the eight
alternatives under a variety of weighting schemes for the four criteria, consistently
provided the lowest total impact score. We can suggest from this result that in the
process of evaluating a set of given alignments it may be useful to broaden the search
and consider other routes not originally envisaged.
In order to incorporate these MCDM approaches into a decision-making
environment it is vital that good relations be established among the interest groups.
Whereas the use of cartographic overlay methods such as the one offered by McHarg
have widespread appeal and require limited technical ability to understand, the
techniques used by Marchet, Siskos and Massam do require considerable technical
ability. Their credibility will be judged very critically by decision-makers and others,
especially if great emphasis is placed upon the use of just four criteria, a simple sixpoint scale of dimensionless numbers and a single value to describe the attractiveness
of an alignment. For these reasons we suggest that the results of the technical analyses
of the sort described here be presented as maps and clear charts and diagrams
indicating the estimated impacts in terms of actual loss of agricultural and other land,
estimated construction and maintenance costs, as well as unambiguous statements
regarding the benefits, especially their distribution among social groups and over time.
While the MCDM techniques may help generate and evaluate alternate alignments
using dimensionless numbers, in the final analysis it is the conversion of these to 'real'
impacts that are needed to improve the quality of the debate among the interest
groups.
The second example we consider in this section puts considerable emphasis on the
71
consideration of a wide variety of criteria, as well as long- and short-term impacts and
the preferences and close involvement of interest groups.
In 1971 the Institute of Ecology at the University of Georgia was requested by the
Georgia Department of Transportation to "make a summary evaluation of all the
reports already prepared on alternate routes for the uncompleted section of 1-75 north
of Atlanta". Objections by environmentalists had persuaded the Federal Highway
Administration to reject a proposed route that had been offered by the Georgia
Department of Transportation. Eight alternate alignments were generated from
lengthy discussions with a number of interest groups and the evaluation problem was
stated in the following terms.
Make a summary evaluation of each proposed route in terms of a single impact index, compounded
by quantifying, weighting and scaling all component values for which data or expert opinions were
available (Odum et al., p. 59, 1976).
An ad hoc interdisciplinary panel was set up to conduct the evaluation with an even
balance between "those accustomed to dealing with environmental matters and those
skilled in the application of economic and human considerations".
The MCDM technique that was selected involved a weighted additive model in
which the criteria scores had been normalised. Fifty-six criteria were used, and for
each a long-term and a short-term weight was assigned by the panel. The long-term
effects were considered to be ten times as important as the short-term ones. All the
weights were normalised. For each alignment twenty simulations were run using a set
of errors generated from the means and standard deviations of the scores for the
impacts for the eight routes. Hence a mean impact score and the 95% confidence limits
were calculated for each alignment. The results from the MCDM analysis gave a series
of dimensionless numbers which indicated that the set of eight routes could be clearly
divided into two sets. The inferior set contained the original route that had previously
been rejected. The next stage involved further close cooperation between the panel,
analysts and the Georgia Department of Transportation in order to undertake a wide
variety of sensitivity tests and to examine the results in terms of the actual impacts
rather than using dimensionless numbers. Alternate weighting schemes were used to
assess changed priorities. On the basis of this work the Georgia Department of
Transportation accepted a recommendation for a new route, public hearings and
additional engineering studies were conducted. At this stage the Federal Highway
Commission became interested in this MCDM technique and the programmes were rerun using scenarios they wished to have examined. All the analysis clearly pointed to
the separation of the routes into two distinct groups, and at the time when Odum et a/.
produced their report a consensus had been reached to gain the approval of the
Federal agency. An analysis of the data set using MATS and concordance analysis is
reported in Massam and Skelton (1986), and again the clear separation of the two sets
of routes is in evidence.
72
Progress in Planning
We might conclude that the use of an MCDM technique to tackle this problem
served to focus discussions among interest groups in such a way that a consensus was
reached and an alternative was chosen for implementation. It is clear that the
technique per se provided the vehicle to test alternate scenarios systematically, and as
such it appears to have considerable merit.
Odum et al. (1976) conclude:
We would strongly recommend that special training courses in computerized impact analysis be set
up at academic centers designed especially for personnel of state, federal, and private consultant
agencies who will be increasingly called on to make decisions that are in the best long-term public
interest (p. 64).
The MCDM technique can take the analysis only so far, the critical element
remains, namely, do those who are involved in the decision-making process believe in
the utility of the technique? This point has already been made at the end of Chapter 2
and we will come back to it again when we deal with the priority rating problems
addressed in the next section.
The two examples to be considered in this section are drawn from France and
Canada. The first deals with the renovation of the metro stations in Paris and involves
an assessment of the needs with respect to a set of 224 stations that are under the
jurisdiction of the Regie Autonome des Transport Parisiens (RATP). In anyone year
all the renovations cannot be completed therefore the task is to prioritize the
renovation needs so that in a particular year the most pressing problems are dealt
with. Clearly the renovation programme is closely linked to the annual budget, and
often this is determined late in the year. The details of this study are reported in Roy et
al. (1986). The second example refers to attempts which are currently underway in the
Ontario Ministry of Transportation and Communications (MTC). This agency has
developed a model called SELECTRA, after the ELECTRE technique, and is
examining its application to prioritize the variety of Provincial transportation and
communication projects under consideration for funding in a given budgetary period.
Whereas the first example focuses on a specific type of planning problem, and a
small discrete set of fairly homogeneous alternate options, the second example
broadens the perspective to consider a range of projects which come from different
parts of the agency, and are very varied in type. For example they could include,
airport runway repairs, highway upgrading and intersection repairs. It is this variety
which makes the application of an MCDM technique very difficult, as there are
numerous interest groups (proponents), representing different departmental interests,
and the impacts, and hence the needs or costs of delayed investments, are very difficult
to estimate and to make comparable. However, it should be noted that both the RATP
and the MTC have chosen to use MCDM techniques to assist in the organization and
analysis of the priority rating planning problems they face. These techniques are being
used to complement professional judgement and the results will be viewed by the
73
agencies in larger political terms. Overall we might suggest that the MCDM techniques
provide benchmark results which can focus debate on trade-offs.
At the outset we should acknowledge that there are very close professional links
between RATP and the Laboratoire d'Analyse et Modelisation de Systemes pour Aide
ala Dedsion (LAMSADE) of the Universite de Paris Dauphine. Professor Roy is the
Director of LAMSADE and he works closely with a number of colleagues who
continue to refine the outranking MCDM ELECTRE model. It is important to
recognize the linkages because credibility for the use of an MCDM technique has been
built up over time and the decision-makers at RATP probably find they can
communicate comfortably with the formal analysts from LAMSADE. This
establishment of a close mutually supportive network among practitioners and
theoreticians appears to be a necessary condition for the application of an MCDM
technique to a practical planning problem. The Paris subway renovation problem was
tackled using the ELECTRE III model.
In 1982 RATP sought to develop a selection procedure to decide which stations
should be renovated. Roy et al. (1986) note that "this procedure had to be designed so
as to be easily reproduced each year; it had to use the most recent data; and it had also
to take into account the fact that the data were often incomplete or inaccurate". The
problem was re-defined in the following way:
Given estimates for a set of seven criteria describing the status of each station classify the stations in
order to produce a renovation schedule.
The criteria that were used included perspectives of passengers and the goals and
constraints of RATP. Further it was decided to divide the set of stations into two
groups, those that required more expensive complex renovation tasks and a group
comprised of stations which could be improved by more modest expenditures on
smaller specific renovations.
An overranking technique (ELECTRE III) was used as it was argued that this
technique allowed the imprecision of the data to be incorporated via sensitivity tests
especially on the indifference thresholds; also tests of robustness could be readily
conducted to examine the effects on the ratings which resulted from changing the
values of the thresholds for the discordance and concordance indices. A new
procedure was devised to search for a general set of values with respect to
combinations of impacts for criteria which were judged important by the different
interest groups. In summary it was argued that the purpose of the particular MCDM
technique used here is to:
-
JPP JO:I-P
74
Progress in Planning
75
considerable progress has been made in the sense that the agency has devoted
resources to develop and test the MCDM technique we should note that several issues
serve to complicate the process of applying a particular model in this case. These
issues are not unique to this particular agency, but stem in large part from the
complexity of the planning environment. While we can identify a particular problem
as a specific example of the generic problem, we must recognize that the analysis
depends upon clarification of a number of items. While the rating problem for the
Paris stations was tackled quite usefully using ELECTRE III, the application of this
type of procedure to rating transportation projects in Ontario is much more
complicated and this point will be discussed later.
The rating problem that the Ontario Ministry of Transportation and
Communications is addressing can be stated in the following terms.
Given all major capital projects (approx. 200) which range in type from municipal and provincial
highway expansion projects to airport modifications and construction, assess each on a set of 10
criteria and derive priorities for undertaking the projects.
76
Progress in Planning
4.5. CONCLUSIONS
CHAPTER 5
More telling is the comment that "indeed, few conflicts involving social and political
values are ever fully removed". Such conflicts are managed, so in order to judge the
utility of MCDM techniques we must rephrase the question to ask if these techniques
can help in the management of conflicts which are indeed explicit in planning
problems. The second approach which fo.llows closely focuses specifically on public
participation and the search for prescriptive suggestions for involving more interest
77
78
Progress in Planning
groups in planning especially with respect to large projects which potentially have
enormous danger associated with them. The classiC examples are the siting of nuclear
waste sites and nuclear reactors. Within this growing literature we find many case
studies and not a few catchy titles like Not In My Back Yard (NIMBY) and Locally
Unwanted Land Uses (LULUS). The recent work at Rutgers University reported by
Lake (1987) seeks to improve public participation especially as it relates to the siting of
hazardous waste facilities. Perhaps the single most important aspect which appears in
all these studies relates to data concerning estimated costs and benefits, and especially
their distribution and accuracy. This is of special concern for those impacts which are
expressed as probabilities and whose amounts are miniscule, yet which potentially may
have catastrophic health and environmental effects. The problems of estimating such
impacts and judging their social significance will not be made easier by any plan
evaluation technique per se. However, if a technique can focus attention on the major
components of a problem, specifically the three we identify in this monograph, namely
the alternatives, the criteria and the interest groups, then a systematic discussion on the
impacts may result.
It is the search for a framework for conducting such discussion that has given rise to
the third approach and this is the one that either makes general statements about
negotiations and conflict resolution following empirical observations or seeks to offer
a normative or prescriptive position. An excellent example of the former is provided
by Gulliver's (1979) work Disputes and Negotiations, and an optimistic set of views on
an ideal negotiating framework is offered by Fisher and Ury (1983). Gulliver (1979)
deliberately adopts a cross-cultural perspective to show that patterns of interactive
behaviour in conflict resolution, dispute settlement and negotiations are essentially
similar despite marked differences in interests, ideas, values, rules, and assumptions
among negotiations in different societies. He notes that:
The study of negotiations has attracted the attention of a diverse variety of social scientists. The
result is that there has been a gradual accumulation both of materials - conceptual, experimental,
and empirical - and of general or partial explanations and theory ... so far, however, the theoretical
dimensions and implications have been poorly developed.
Most planners may turn away from general statements as their concern is probably
grounded in a pressing need for an immediate approach to deal with a unique
planning problem. To this end they may resort to the method of principled negotiation
(PN) which has been developed as part of the Harvard Negotiation Project and
reported by Fisher and Ury (1983) under the title: Getting to Yes. They argue that PN
seeks "to decide issues on their merits rather than through a haggling process focussed
on what each side says it will and won't do: It suggests that you look for mutual gains
whenever possible, and that where your interests conflict, you should insist that the
result be based on some fair standards independent of the will of either side". We
could of course criticize the naivete of such an implicit view of negotiations but if we
take PN at face value it is clear that it may help resolve differences, and seen within
the context of planning we therefore need a clear exposition of the alternatives under
review, the evaluation criteria and the interested parties, as well of course as the
impact values.
79
80
Progress in Planning
condition. For when all is said and done we are planning for ourselves as individuals
and for the survival of our species. It is in all our interests to improve planning
practises and to continue the debate on the quality of planning.
Bibliography
ALI, I., COOK, W. D. and KRESS, M. (1986) On the minimum violations ranking of a tournament,
Management Science, 32, (6) 660-672.
AMBROSE, P. (1986) Whatever Happened to Planning? Methuen, London.
AMRHEIN, C. G. (1985) An interactive multi-criteria decision model for spatial problems, East Lakes
Geographer, 20, 20-28.
ARROW, K. J. (1951) Social Choice and Individual Values, John Wiley, New York.
ASKEW, I. D. and MASSAM, B. H. (1981) Trouble in store, Area, Institute ofBritish Geographers, 13, (4)
272-275.
BARZILAI, J., COOK, W. D. and GOLANY, B. (n.d.) Axiomatic Foundations for the Analytical
Hierarchy Process, Working Paper, No. 44, School of Business Administration, Dalhousie
University, Canada.
BATTY, M. and HUTCHINSON, B. (1983) Systems Analysis in Urban Policy-Making and Planning, Plenum
Press, New York.
BELL, D. E., KEENEY, R. L. and RAIFFA, H. (Eds) (1977) Conflicting Objectives in Decisions, John
Wiley, New York.
BERNSTEIN, S. J. and MELLON, W. G. (Eds) (1978) Selected Readings in Quantitative Urban Analysis,
Pergamon, Oxford.
BLAIR, D. H. and POLLAK, R. A. (1983) Rational collective choice, Scientific American, 249, (2) 88-95.
BOGART, K. P. (1973) Preference structures I: distances between transitive preference relations, Journal
Mathematical Society, 3,49-67.
BOGART, K. P. (1975) Preference structures II: distances between asymmetric relations, SIAM, Journal of
Applied Mathematics, 29,254-262.
BRACKEN, I. (1981) Urban Planning Methods: Research and Policy Analysis, Methuen, London.
BRANS, J. P., VINCKE, P. and MARESCHAL, B. (1986) How to select and how to rank projects: the
PROMETHEE method, European Journal of Operational Research, 24, 228-238.
BREHENY, M. and HOOPER, A. (Eds) (1985) Rationality in Planning, Pion, London.
BROWN, C. A. and VALENTI, T. (1983) Multi-attribute trade-off system: user's and programmer's manual,
Bureau of Reclamation, U.S. Department of the Interior, Denver, Colorado.
COCHRANE, J. L. and ZELENY, M. (Eds) (1973) Multiple Criteria Decision Making, University of South
Carolina Press, Columbia.
COGGER, K. and YU, P. L. (1985) Eigenweight vectors and least-distance approximation for revealed
preference in pairwise weight ratios, Journal of Optimization, Theory and Applications, 46, 4.
COOK, W. D. (1986a) Ordinal Ranking and Preference strength, Mathematical Social Sciences, 11,295-306.
COOK, W. D. (l986b) A Multicriteria Methodology for Transportation Capital Investment Priority Setting: a
critical analysis of the SELECTRA model, mimeo, 73, Aberfeldy Crescent, Thornhill, Ontario,
Canada, L3T 4C3.
COOK, W. D. and KRESS, M. (1984) Relationships between I' metrics on linear ranking spaces, SIAM,
Journal ofApplied Mathematics, 44, (I) 209-220.
COOK, W. D. and SEIFORD, L. M. (1978) Priority ranking and consensus formation, Management
Science, 24, (16) 1721-1732.
COOK, W. D. and SEIFORD, L. M. (1982) On the Borda-Kendall consensus method for priority ranking
problems, Management Science, 28, (6) 621-637.
COOK, W. D., GOLAN, I. and KRESS, M. (1985) Heuristics for Ranking Players in a Round Robin
Tournament, I.B.M. Israel Scientific Center, Technical Report 88.175.
COOK, W. D., GOLAN, I., KAZAKOV, A. and KRESS, M. (1986) A multicriteria model for ranking
81
82
Progress in Planning
transportation projects: a case study, mimeo, Faculty of Administrative Studies, York University,
Canada,p.21.
DE NEUFVILLE, R. and MARKS, D. H. (1974) Systems Planning and Design, case studies in modelling,
optimization and evaluation, Prentice Hall, New Jersey.
EDWARDS, W. (1971) Social Utilities, Engineering Economist, Summer Symposium Series, 6.
EDWARDS, W. (1977) Use of multiattribute utility measurement for social decision making, In: Bell, D. E.,
Keeney, D. 1. and Raiffa, H. (1977) Conflicting Objectives in Decisions, pp. 247-276.
ETZIONI, A. (1967) Mixed scanning: a third approach to decision making, Public Administration Review,
27, 385-392.
FANDEL, G. and SPRONK, J. (Eds) (1985) Multiple Criteria Decision Methods and Applications, Springer
Verlag, New York.
FISHBURN, P. C. (1971) A comparative analysis of group decision methods, Behavioral Science, 16,
538-544.
FISHBURN, P. C. (1974) Simple voting systems and majority rule, Behavioral Science, 19, 166-176.
FISHER, R. and URY, W. (1983) Getting to Yes, Penguin, Harmondsworth.
FRENCH, S., HARTLEY, R., THOMAS, 1. C. and WHITE, D. J. (Eds) (1983) Multi-objective Decision
Making, Academic Press, London.
FRIEND, J. K. and JESSOP, W. N. (1969) Local Government and Strategic Choice, Tavistock, London.
GARDINER, P. C. and EDWARDS, W. (1975) Public values: multiattribute utility measurement for social
decision making, In: Kaplan, M. F. and Schwartz, S. (Eds) (1975) Human Judgement and Decision
Processes, Academic Press, New York, pp. 1-37.
GOLLEDGE, R. G. and RUSHTON, G. (1972) Multidimensional scaling: review and geographical
applications, Technical Paper No. 10, Association of American Geographers, Washington, D.C.
GUIGOU, J.-L. (1971) On French location models for production units, Regional and Urban Economics, 1,
(2) 107-138.
GULLIVER, P. H. (1979) Disputes and Negotiations: a cross-cultural perspective, Academic Press, London.
GUY, C. M. (1980) Policies for the location of large new stores - a case study, Area, Institute ofBritish
Geographers. 12, (4) 279-284.
HALL, P. (1980) Great Planning Disasters, Weidenfeld and Nicolson, London.
HARDIN, G. (1984) The tragedy of the common, Chapter 8. In: Korten, D. C. and Klauss, R. (Eds) (1984),
People Centered Development, Kumarian Press, W. Hartford, Connecticut.
HAMMOND, K. R., McCLELLAND, G. H. and MUMPOWER, J. (1980) Human Judgement and Decision
Making, Praeger, New York.
HARVEY, D. (1985) The Urbanisation of Capital, B. Blackwell, Oxford.
HATRY, H. D. (1972) Issues in productivity measurement for local governments, Public Administration
Review, 32, (6) 776-784.
HEALEY, P. (1986) Autumn Books: review of The Urbanisation of Capital and Consciousness and the Urban
Experience (D. Harvey, 1985; B. Blackwell, Oxford), The Planner, Oct., pp. 24-27.
HILL, M. (1968) A goals-achievements matrix for evaluating alternative plans, Journal of the American
Institute of Planners, 34, 19-29.
HILL, M. (1971) Review of Lichfield, N., Kettle, P. and Whitbread, M. (1975) (eds) Evaluation in the
Planning Process, Pergamon, Oxford, In: Journal of the American Institute of Planners, 43, (2) 207.
HOLLIS, M., SUGDEN, R. and WEALE, A. (1985) Riddles of public choice, Times Higher Educational
Supplement, Nov. 25, p. 15.
HUTCHINSON, B. G. (1974) Principles of Urban Transport Systems Planning, Scripta, Washington.
HWANG, C.-L. and YOON, K. (1981) Multi Attribute Decision Making, Springer-Verlag, New York.
JAAKSON, R. (1984) Recreational planning for a small urban lake, Town Planning Review, 56, (I) 90-111.
JOHNSON, C., BEINE, W. and WANG, T. (1979) A note on right-left asymmetry in an eigenvector
ranking procedure, Journal of Mathematical Psychology. 19.
KARSKI, A. (1985) Strategic Planning: some lost lessons from British Planning Methodology, Planning
Studies No. 16, Polytechnic of Central London, School of Planning.
KEENEY, R. 1. (1972) Utility functions for multi-attributed consequences, Management Science, 18,
276-287.
KEENEY, R. 1. (1981) Siting major facilities using decision analysis, In: Operational Research 1981, J. P.
Brans (Ed.) North Holland: Amsterdam, pp. 355-378.
KEENEY, R. 1. and NAIR, K. (1977) Selecting nuclear power plant sites in the Pacific Northwest using
decision analysis, In: Bell, D. E., Keeney, R. 1. and Raiffa, H. (1977) Conflicting Objectives in
Decisions, pp. 298-322.
KEENEY, R. 1. and RAIFFA, H. (1976) Decisions with Multiple Objectives: Preferences and Value Tradeoff,
John Wiley, New York.
83
KEMENY, J. G. and SNELL, L. J. (1962) Mathematical Models in the Social Sciences, Ginn, New York.
KENDALL, M. (1962) Rank Correlation Methods, Hafner, New York.
KRUECKEBERG, D. A. and SILVERS, A. L. (1974) Urban Planning Analysis, John Wiley, New York.
LAKE, R. W. (Ed.) (1987) Resolving Locational Conflict, Rutgers University, New Jersey.
LANGLEY, P. (1974) Evaluating the effectiveness ofalternatives in Structure Planning, Department of the
Environment, Planning Techniques Paper 1/74, London.
LEE, C. (1973) Models in Planning, Pergamon, Oxford.
LICHFIELD, N. (1970) Evaluation methodology of urban and regional plans: a review, Regional Studies. 4,
151-165.
LINSTONE, H. A. (1984) Multiple Perspectives for Decision Making, North Holland, Amsterdam.
LOOTSMA, F. A., MEISNER, J. and SCHELLEMANS, F. (1986) Multi-criteria decision analysis as an aid
to the strategic planning of energy R D, European Journal of Operational Research, 25,216-234.
MARCHET, J.-C. (1980) Decisions en matiere d'environnement, Doctorat de 3e cycle, Universite paris IX,
Dauphine.
MARCHET, J.-C. and SISKOS, J. (1979) Aide a la decision en matiere d'environnement: application au
choix de trace autoroutier, Sistemi Urbani, 1, (2) 65-95.
MASSAM, B. H. (1981) The fire-station location problem in North York, Ontario, Urban Studies Working
Paper No.2, York University, Canada.
MASSAM, B. H. (1980) Spatial Search: applications to planning problems in the public sector, Pergamon,
Oxford.
MASSAM, B. H. (1982) The search for the best route: an application of a formal method using multiple
criteria, Sistemi Urbani. 4, (1/2) 183-194.
MASSAM, B. H. (1984) The site location problem, The Operational Geographer, 5, 16-19.
MASSAM, B. H. (1986) The need for sensitivity tests in multicriteria plan evaluation, The Operational
Geographer, 10, 28-30.
MASSAM, B. H. and ASKEW, I. D. (1982) Methods for comparing policies using multiple criteria: an
urban example, OMEGA, The International Journal ofManagement Science, 10, (2) 195-204.
MASSAM, B. H., AKHTAR, R. and ASKEW, I. D. (1986) Applying operations research to health
planning: locating heaith centres in Zambia, Health Policy and Planning. 1, (4), 326-334.
MASSAM, B. H. and SKELTON, I. (1986) Application of three plan evaluation procedures to a highway
alignment problem, Transportation Research Record, Transportation Research Board, National
Research Council, Washington, D.C., No. 1076, pp. 54-58.
McALLISTER, D. M. (1980) Evaluation in Environmental Planning, MIT, Cambridge, Massachusetts.
McHARG, I. (1969) Design with Nature, Doubleday, New York.
MIDGLEY, J. and PIACHAUD, D. (Eds) (1984) The Fields and Methods of Social Planning, Heinemann,
London.
MUMPHREY, A. J., SELEY, J. E. and WOLPERT, J. (1971) A decision model, for locating controversial
facilities, Journal American Institute ofPlanners, 37, (6) 397-402.
NELKIN, D. (Ed.) (1984) Controversy: politics of technical decisions, Sage, London.
NIJKAMP, P. (1979) Multidimensional Spatial Data and Decision Analysis, John Wiley, New York.
NIJKAMP, P. (1982) Soft multicriteria analysis as a tool in urban land-use planning, Environment and
Planning B, 9, 197-208.
NIJKAMP, P. and SPRONK, J. (1981) Multiple Criteria Analysis, Gower, Aldershot.
NOWLAN, D. M. (1975) The use of criteria weights in rank ordering techniques of project evaluation,
Urban Studies, 12, 169-176.
ODUM, E. P., BRAMLETT, G. A., IKE, A., CHAMPLAIN, J. R., ZIEMAN, J. C. and SHUGART, H. H.
(1976) Totality indexes for evaluating environmental impacts of highway alternatives, Transportation
Research Record. 561,57-67.
ONTARIO WASTE MANAGEMENT CORPORATION (1985) Annual Report, Toronto.
RADFORD, K. J. (1980) Categories of decision problems and their resolution, Infor, 18, (I) 82-88.
RADFORD, K. J. (1986) Strategic and Tactical Decisions, Holt McTavish, Toronto.
RICHELSON, J. (1978) A comparative analysis of social choice functions, Behavioral Science, 23, 169-176.
RIETVELD, P. (1980) Multiple Objective Decision Methods and Regional Planning, North Holland,
Amsterdam.
RIETVELD, P. (1984) Public choice theory and qualitative discrete multi-criteria decision methods, Chap.
24, In: Bahrenberg, G., Fischer, M: and Nijkamp, P. (Eds) (1984) Recent Developments in Spatial
Data Analysis, Gower, Aldershot.
ROSENBAUM, S. P. (Ed.) (1975) The Bloomsbury Group, University of Toronto Press, Toronto.
ROWE, M. D. and PIERCE, B. L. (1982a) Some tests of analytical multiobjective decision-making
methods, Socio-Economic Planning Sciences, 16, (3) 133-140.
84
Progress in Planning
ROWE, M. D. and PIERCE, B. L. (1982b) Sensitivity of the weighting summation decision method to
incorrect application, Socio-Economic Planning Science, 16, (4) 173-177.
ROY, B. (1968) Classement et choix en presence de points de vue multiples, Revue Francaise d'informatique
et de Recherche Operationne//e, 6, (8) 57-75.
ROY, B. (1978) ELECTRE III: un algorithme de c1assement fonde sur un representation floue de
preferences en presence de criteres multiple, Cahie~s du Centre Etudes Recherche Operatione//e, 20, (1)
3-24.
ROY, B. and BOUYSSOU, D. (1986) Comparison of two decision-aid models applied to a nuclear power
plant siting example, European Journal of Operational Research, 25, 200-215.
ROY, B., SUSSMAN, B. and BENAYOUN, R. (1966) A decision method in presence of multiple viewpoints: ELECTRE, Study session on Methods of Calculation in the Social Sciences, Rome.
ROY, B., PRESENT, M. and SILHOL, D. (1986) A programming method for determining which Paris
metro stations should be renovated, European Journal of Operational Research, 24, 318-334.
SAATY, T. L. (1980) The Analytic Hierarchy Process, McGraw-Hili, New York.
SAGER, T. (1981) Evaluation methods in local pllrticipatory planning, Town Planning Review, 52, (4)
417-432.
'
SCHLAGER, K. (1968) The rank-based expected value method of plan evaluation, Highway Research
Record, 238, 153-158.
SELF, P. (1975) Econocrats and the Policy Process: the Politics and Philosophy of Cost-Benefit Analysis.
Macmillan, London.
SISKOS, J. and HUBERT, P. (1983) Multi-criteria analysis of the impacts of energy alternatives: a survey
and a new comparative approach, European Journal of Operational Research, 13, (2) 278-299.
SMITH, D. M. (1977) Human Geography: a welfare approach, (Ed.) Arnold, London.
SOLOMON, B. D. and HAYNES, K. E. (1984) A survey and critique of multiobjective power plant siting
decision rules, Socio-Economic Planning Sciences, 18, (I) 71-79.
STEWART, T. J. (1979) A Descriptive Approach to Multi-Criteria Decision-Making, (TWISK 128), Pretoria,
South Africa: National Research Institute for Mathematical Sciences, Technical Report.
STEWART, T. J. (1981) A descriptive approach to multi-criteria decision-making, Journal Operations
Research Society, 32, 45-53.
STOKEY, E. and ZECKHAUSER, R. (1978) A Primer for Policy Analysis, Nowton, New York.
SZIDAROVSZKY, F. and DUCKSTEIN, L. (1986) Dynamic multiobjective optimization: a framework
with application to regional water and mining management, European Journal of Operational
Research, 24, 305-317.
THIRIEZ, H. and ZIONTS, S. (Eds) (1976) Multiple Criteria Decision Making, Springer-Verlag, New York.
VINCKE, P. (1986) Analysis of multicriteria decision aid in Europe, European Journal of Operational
Research, 25, 160-168.
VOOGD, H. (1982) Multicriteria evaluation with mixed qualitative and quantitative data, Environment and
Planning B, 9,221-236.
VOOGD, H. (1983) Multicriteria Evaluationfor Urban and Regional Planning, Pion, London.
WILSON, A. G. (1974) Urban and Regional Models in Geography and Planning, John Wiley, New York.
WILSON, A. G. and KIRKBY, M. J. (1975) Mathematicsfor Geographers and Planners, Clarendon, Oxford.
ZELENY, M. (Ed.) (1976) Multiple Criteria Decision Making, Kyoto 1975, Springer-Verlag, New York.
ZELENY, M. (Ed.) (1984) MCDM: Past Decade and Future Trends, a sourcebook ofmu1tiple criteria decision
making, Jai Press, London.