Identifying Objectives and Criteria
Decision makers' and stakeholders' values or preferences may be explicitly included in a MCA model through a set of criteria against which the impact of the potential policy options is evaluated. This may include environmental criteria, such as the protection of natural habitats for certain species, or economic criteria, such as the job and economic development opportunities or costs arising from an infrastructure development policy. Evaluation criteria can be built in two ways. The top-down approach starts from a main objective and builds a hierarchical tree structure of fundamental objectives (Keeney 1992) or key concerns (Bana e Costa and Beinat 2005). The bottom-up approach starts from the impacts of policy options and builds a consistent family of evaluation criteria (Roy 1996) by partial synthesis of related and non-conflicting items. In practice, a combination of the two approaches may prove the most efficient (Bana e Costa and Beinat 2005).
Numerous studies in the MCA literature have addressed the desirable properties of a good set of criteria: the most important include (1) exhaustiveness (the criteria selected characterize completely the evaluation of any policy option); (2) cohesiveness (partial preferences with respect to each individual criterion have to be consistent with the global preference); and (3) non-redundancy (elimination of any criterion from the chosen set of criteria leads to violation of at least one of the previous properties) (Roy 1996). The second in this list means for instance that an improvement of an option with respect to some criteria should not lead to a worse 'global' evaluation. Roy and Bouyssou (1993) give an example involving a 'reliability' criterion, proving that in some cases this property might not be valid: two cost values might be indifferent if they both have a low reliability, but the lower cost could be strictly preferred if the higher cost is obtained with a higher reliability. Keeney (1992) argues that the evaluation criteria should also be operational, that is, allowing impact assessment for the available policy options within reason given available time and resources. Related to the latter, one should note that in the presence of uncertainty, the evaluation of an option with respect to a criterion might not be a unique element, but rather an interval, a distribution or a fuzzy set.