The strength of weaker MCDA methods
Pekka Salminen
University of Joensuu, Department of Economics, Finland
Risto Lahdelma
University of Turku, Computer Science, Finland
Inspired by the interesting discussion in the EWG/MCDA Newsletter, we would
like to contribute with some of our own thoughts of the behavioural aspects
of decision-making and ideas on how to overcome some biases.
During the past ten years we have participated in a number of real-life
MCDA applications, mainly in the field of public environmental decision
making. In our vocabulary 'real-life application' (RLA) is close to the
definition of Kasanen et al. (2000), but most of our applications would
fit well into one or more of the categories B to F that Vincke proposed
in the Newsletter of fall 2000. We have acted in these projects as the
MCDA analysts as well as developed some new methodologies and software
tools in conjunction with these applications.
To begin with, let us discuss some possible answers to the philosophical
question "What is our goal in MCDA?":
-
To assist in making "better decisions". Unfortunately, in general, there
is no objective measure for a claim that one decision is better than another.
At best, we can eliminate dominated alternatives, and try to control the
decision process so that obvious mistakes and oversights are avoided.
-
To assist in making decisions that the DMs (or the public) will be happier
with. Even this weaker goal can be difficult to reach in RLAs because we
cannot know the reference point, i.e., what the solution might have been
with some other MCDA method, or without any method at all.
-
To assist in making decisions that the DMs (or the public) will be satisfied
with. Without a reference point, the DMs can judge qualitatively how well
they think they understood the problem, how satisfied they were with the
method, and how strongly they believe in having made the right decision.
Such satisficing decision aids were discussed by Rauschmayer in the Newsletter
of spring 2001.
-
To save work and other resources in the decision-making process. Good MCDA
methods can streamline or automate parts of the information processing
and reduce the information requirements in decision making (e.g. ordinal
vs. cardinal information, preference information-free methods). These savings
can be assessed relatively easily. In this weakest goal, the decisions
do not necessarily have to be subjectively or objectively "better". Of
course the decision quality may improve if the saved resources can be used
to deepen the analysis.
With these different possible goals in mind, we can try to evaluate how
results from behavioural research should be considered in MCDA. For one
thing, we think it is necessary to point out that the results from behavioural
research typically emerge from only small fragments of behaviour. There
is no clear understanding how these limited observations should be combined
to understand the overall decision-making process.
Nevertheless, even fractional behavioural information is obviously useful.
We believe that the most important function in each goal setting is to
ensure that the decision-making method does not put any unreasonable demands
on the DMs. Having the DMs make holistic evaluations in high-dimensional
spaces may result in arbitrary answers. The DMs may refuse to express tradeoffs
between criteria that are fundamentally incomparable or otherwise alien
to them. Too many pairwise questions (as in large AHP models) may cause
boredom and fatigue and result in increasingly inconsistent answers.
Understanding how humans process information is clearly important when
constructing various decision models. The DMs (and the public) are more
likely to accept the method and the results if they are able to understand
the decision model and find the method somehow "natural".
In the stronger goal settings, it is indeed essential to try to avoid the
various behavioural biases that may have substantial influence e.g. "on
the form of the value function model", as Stewart states in the Newsletter
of fall 2000. He also mentions that "... all methods make use of
direct or indirect weighting of the criteria". Of course, there is also
a category of so-called preference information-free MCDA methods that can
be used without direct or indirect weighting of the criteria. One obvious
advantage of these methods is that they are less susceptible to the framing,
anchoring and availability biases.
Preference information-free methods include e.g. the Hypervolume criterion
method by Charnetski & Soland (1978), Overall compromise criterion
method by Bana e Costa (1986), and SMAA-family of methods by Lahdelma et
al. (1998, 2001). These methods operate by exploring the space of possible
weights internally, and reveal what kinds of preferences favour each alternative.
In particular, the SMAA-methods perform a stochastic weight space analysis,
and compute how large shares of weights would make an alternative the best
one (stochastic efficiency), or place it on a particular rank (SMAA-2).
This descriptive information can then be used to identify the probably
best alternatives, to eliminate inferior alternatives, or to find alternatives
reflecting potential compromises. The DMs can either make the decision
based on this information, or narrow the weight space by providing (partial)
preference information. This approach is less sensitive to the different
behavioural biases, because it can be used completely without or with only
partial preference information.
Preference-information-free decision-making methods can also alleviate
the problem of representing the preferences of non-existent decision actors,
which was mentioned by Rauschmayer in the Newsletter of spring 2001. As
these methods consider all possible preferences, they will also include
the preferences of these missing DMs. Obviously, their interests must be
represented among the set of criteria, and no formal method can ultimately
guarantee this.
References
-
Bana e Costa C.A. (1986). A Multicriteria Decision Aid Methodology to Deal
with Conflicting Situations on the Weights, European Journal of Operational
Research 26, 22-34.
-
Charnetski J.R., Soland R.M. (1978). Multiple-Attribute Decision Making
with Partial Information: the Comparative Hypervolume Criterion. Naval
Research Logistics Quarterly 25, 279-288.
-
Kasanen, E., Wallenius, H., Wallenius, J., & Zionts, S. (2000). A study
of high-level managerial decision processes, with implications for MCDM
research, European Journal of Operational Research 120(3), 496-510.
-
Lahdelma, R., Hokkanen, J. & Salminen, P. (1998). SMAA - Stochastic
Multiobjective Acceptability Analysis. European Journal of Operational
Research, 106(1), 137-143.
-
Lahdelma, R. & Salminen, P. (2001). SMAA-2: Stochastic Multicriteria
Acceptability Analysis for Group Decision Making. Operations Research 49(3),
444-454.
[HOME - EWG MCDA]
[Newsletter of the EWG - MCDA]