Multiple criteria decision-making (MCDM) approa-ches are major parts of decision theory and analysis.They seek to take explicit account of more than onecriterion in supporting the decision process . The aimof MCDM methods is to help decision-makers learnabout the problems they face, to learn about their ownand other parties' personal value systems, to learn aboutorganizational values and objectives, and throughexploring these in the context of the problem to guidethem in identifying a preferred course of action[5,12,20,32,34,35]. In other words, MCDA is useful incircumstances which necessitate the consideration of dierent courses of action, which can not be evaluatedby the measurement of a simple, single dimension .Hwang and Yoon  published a comprehensivesurvey of multiple attribute decision making methodsand applications. Two types of the problems that arecommon in the
that best ®t MCDAmodels are evaluation problems and design problems.The
is concerned with the evaluationof, and possible choice between, discretely de®nedalternatives. The
is concerned with theidenti®cation of a preferred alternative from a poten-tially in®nite set of alternatives implicitly de®ned by aset of constraints .
3. The analytical hierarchy process (AHP)
Belton  compared AHP and a simple multi-attri-bute value (MAV), as two of the multiple criteriaapproaches. She noticed that both approaches havebeen widely used in practice which can be considered asa measure of success. She also commented that thegreatest weakness of the MAV approach is its failure toincorporate systematic checks on the consistency of judgments. She noticed that for large evaluations, thenumber of judgments required by the AHP can besomewhat of a burden.A number of criticisms have been launched at AHPover the years. Watson and Freeling  said that inorder to elicit the weights of the criteria by means of aratio scale, the method asks decision-makers mean-ingless questions, for example: `Which of these two cri-teria is more important for the goal? By how much?'Belton and Gear  and Dyer  pointed out that thismethod can suer from rank reversal (an alternativechosen as the best over a set of
, is not chosen whensome alternative, perhaps an unimportant one, is exclu-ded from
). Belton and Gear  and Dyer and Wendel attacked the AHP on the grounds that it lacks a®rm theoretical basis. Harker and Vargas  and Perez discussed these major criticisms and proved with atheoretical work and examples that they are not valid.They commented that the AHP is based upon a ®rmtheoretical foundation and, as examples in the literatureand the day-to-day operations of various governmentalagencies, corporations and consulting ®rms illustrate,the AHP is a viable, usable decision-making tool.Saaty [24±27] developed the following steps forapplying the AHP:1. De®ne the problem and determine its goal.2. Structure the hierarchy from the top (the objec-tives from a decision-maker's viewpoint) throughthe intermediate levels (criteria on which sub-sequent levels depend) to the lowest level whichusually contains the list of alternatives.3. Construct a set of pair-wise comparison matrices(size
) for each of the lower levels with onematrix for each element in the level immediatelyabove by using the relative scale measurementshown in Table 1. The pair-wise comparisons aredone in terms of which element dominates theother.4. There are
judgments required to developthe set of matrices in step 3. Reciprocals are auto-matically assigned in each pair-wise comparison.5. Hierarchical synthesis is now used to weight theeigenvectors by the weights of the criteria and thesum is taken over all weighted eigenvector entriescorresponding to those in the next lower level of the hierarchy.6. Having made all the pair-wise comparisons, theconsistency is determined by using the eigenvalue,
, to calculate the consistency index, CI as fol-lows:
is thematrix size. Judgment consistency can be checkedby taking the consistency ratio (CR) of CI with theappropriate value in Table 2. The CR is accep-table, if it does not exceed 0.10. If it is more, the judgment matrix is inconsistent. To obtain a con-sistent matrix, judgments should be reviewed andimproved.7. Steps 3±6 are performed for all levels in the hier-archy.
Table 1Pair-wise comparison scale for AHP preferences [24±27]Numerical rating Verbal judgments of preferences9 Extremely preferred8 Very strongly to extremely7 Very strongly preferred6 Strongly to very strongly5 Strongly preferred4 Moderately to strongly3 Moderately preferred2 Equally to moderately1 Equally preferred20
K. M. A.-S. Al-Harbi/International Journal of Project Management 19 (2001) 19±27