ISPRS, Vol.34, Part 2W2, “Dynamic and Multi-Dimensional GIS”, Bangkok, May 23-25, 2001
decision problem in SMART is called a value tree, or objective
hierarchy, which is a true tree structure allowing one subcriteria
to be connected to only one higher level criterion.
For a model with multiple layers of criteria, lowest criteria weight
(model weight, accumulated weights) is the product of
normalized weights along each path from the Goal block to that
criteria, summed over all such paths. The decision score of each
alternative is the sum, over all lowest criteria (attribute) of each
normalized rating multiplied by the model weight.
Gardiner and Edwards (1975) suggested that SMART "is
oriented not towards mathematical sophistication or intimacy of
relation between underlying formal structures and practical
procedures that implement them but rather towards easy
communication and use in environment in which time is short
and decision makers are multiple and busy". SMART is a
powerful and flexible decision making tool. Because of its
simplicity of both responses required of the decision maker and
the transparent manner in which these responses are analysed,
the method is likely to yield an enhanced understanding of the
problem and be acceptable to the decision maker who is
distrustful of a 'black box’ approach. SMART has been widely
applied, and is the focus of many publications on Multi-Criteria
Decision Making.
Edwards (1977a and 1977b) identified the main stages in the
SMART analysis as:
1. Identify the decision maker (or decision makers).
2. Identify the alternative courses of action.
3. Identify the attributes which are relevant to the decision
problem.
4. For each attribute, assign values to measure the
performance of the alternatives on that attribute.
5. Determine a weight for that attribute that reflects how
important the attribute is to the decision maker.
6. For each alternative take a weighted average of the values
assigned to that alternative to give a measure of how well
an alternative performs over all the attributes.
7. Make a provisional decision.
8. Perform sensitivity analysis to see how robust the decision
is to changes in the weights.
3.2 Normalization when using SMART
Attribute normalization converts the different scales of attribute
to a common internal scale. In SMART this is done
mathematically by means of a value function which transforms
attribute values into ratings on a consistent internal scale with
lower limit 0, and upper limit 1. The simplest choice of a value
function is a linear one, and in most cases this is sufficient. The
literature on Utility theory discusses the different ways of
choosing a value function (Edwards, 1977a). Through using the
value function the ratings of alternatives are not relative, so that
varying the number of alternatives will not in itself change the
decision scores. This allows new candidates to be added to the
model or closed branches to be taken out of the model.
In SMART, the normalization weight process results in the
effective importance of a subcriteria by using Direct Rating (DR)
methods, the user-given weight divides the sum of the weights
at the same level against an upper level criteria. The DR
method used in this process forces the numerical values, or
priorities to range between 0 to 1.
3.3 Criteria for determining branch bank viability
There is now a considerable literature dealing with different
aspects of the changes taking place in retail banking. Of
relevance for this research are the studies focussing on the
criteria and performance indicators related to making decisions
about the closure of individual branches and the reshaping of
branch bank networks. Selection of the non-financial criteria
(Table 1) used in developing the SDSS in this study is based on
a review of this literature and the findings of previous successful
research (see for example, ABA, 1965; Morrall, 1996; Avkiran,
1997).
Table 1 The Principle Social and Demographic Criteria
Determining Branch Profitability
Banking population (age 15 & over)
_ M
re (u
m —
" S3
re re
Population (over 15) growth rate
Competitor's branches
*£ C
§ 5
Average annual family income
£ S
SZ «=
Average age
3 ö
re o>
Employed people
O Q.
Total (small) business No.
Catchment area
Small business in 200m buffer zone
u
Competitor's branch No. in 200m buffer
0
1 ®
if) -O
Working population in 200m buffer
Same bank’s branches in 500m buffer
O fc
At shopping centre
8 >
Shopping centers Within 500m buffer
Present of a free car park
Proximity to public transport (within 1km)
The criteria are divided into two higher level categories termed
Catchment Specific and Location Specific criteria. Some of the
individual criteria are taken from published sources while others
have been generated using GIS.
3.4 Criteria collection and management using GIS
In this research, GIS analytical functions facilitate spatial
manipulation and generate data based on spatial analysis. The
database management functions are used for managing data
derived from different sources and formats, and integrating data
based on spatial relations. The results produced by ArcView can
be presented in a standard attribute table (the decision table),
the rows in which contain the set of branch alternatives, the
columns the set of decision criteria, and the table elements the
criteria scores. The decision table can be exported directly to
CDP through DDE and a program written in AVENUE. The multi
criteria decision-making analysis is performed in CDP, a
weighted decision score is then calculated for each candidate
branch bank, and these are exported to ArcView for display.
3.5 Determining criteria weights
The important components of an MCDM analysis are those that
involve interaction with the decision makers, in order to develop
a value function to obtain a set of weights for the evaluation
criteria. Research (Nijkamp et al., 1990) has shown that decision
makers prefer not to give their preference until the result of the
analysis are presented to them, and generally speaking, it is too
demanding for a decision maker to give a precise indication of
his preferences. This point has been elaborated by Malczewski
(1996), who further argued that it is difficult for decision makers
to provide the information required by the weighting methods,
and their preference judgements are often nor reliable enough to
yield precise criteria weights.
Nijkamp (1990) has demonstrated that it is sufficient to use
hypothetical qualitative priority statements, linked with a
particular policy scenario. Malczewski (1996) suggested that this
problem can be overcome by using a series of sensitivity tests,
