ELECTRE - TRI multicriteria decision making method requires the definition of veto, preference and indifference thresholds. My question is how to fix them for a specific problem ? can this task be acheived manually by an expert ??
Usually a good way to set thresholds is by a survey of experts in the field pertaining to the decision problem. However, this requires an extensive effort.
Alternatively, you may want to do some trial and error. You could start with a larger value and progress downward reducing until you feel that the preference is not so much well established.
1. Mousseau, V., Slowinski, R., & Zielniewicz, P. (2000). A user-oriented implementation of the ELECTRE-TRI method integrating preference elicitation support. Computers & operations research, 27(7-8), 757-777.
2. Mousseau, V., Figueira, J., & Naux, J. P. (2001). Using assignment examples to infer weights for ELECTRE TRI method: Some experimental results. European Journal of Operational Research, 130(2), 263-275.
3. Dias, L., Mousseau, V., Figueira, J., & Clı́maco, J. (2002). An aggregation/disaggregation approach to obtain robust conclusions with ELECTRE TRI. European Journal of Operational Research, 138(2), 332-348.
4. Almeida-Dias, J., Figueira, J. R., & Roy, B. (2010). Electre Tri-C: A multiple criteria sorting method based on characteristic reference actions. European Journal of Operational Research, 204(3), 565-580.
Fiddling with ELECTRE TRI thresholds is rather tricky, and defining reference alternatives is even more. It is really case-specific and requires some experience. I suggest you simulate some data and play around with different thresholds and references before calibrating the method to your real problem.
Also, ELECTRE TRI has two procedures: pessimistic and optimistic (it has to do with the way alternatives are compared to the reference ones). In all real-life applications I've seen, the optimistic procedure was *really* optimistic, to the point where most alternatives ended up in the top classes. So I recommend using the pessimistic procedure.
If you set the veto threshold to a very high number, the electre discordance index will drop to zero for all pairwise comparisions. That's just it. In practice the method will become more similar to a compensatory method.
If the veto thresholds on every criteria are very large, the discordance index goes to zero (no single criteria may oppose to aSb). If the veto thresholds are very close to the strict preference thresholds, the discordance index goes to 1 (every time a strict preference occurs in the opposite sense to aSb, that criteria will object to aSb). You can check this out in the textbooks.
As explained by Mousseau et al. (2000), threshold values have to be established by the analyst in order to calibrate the method in adherence with the particular problem under study. For example, the veto threshold can be assumed as equal to the width of classes, whereas the preference and indifference thresholds respectively as a half and a quarter of the veto threshold. Larger values can be first simulated for thresholds by leading various attempts and then progressively reduced until considered appropriate for each criterion.
Mousseau, V., Słowiński, R., Zielniewicz, P. (2000) A user-oriented implementation of the ELECTRE TRI method integrating preference elicitation support. Computers & Operations Research, 27(7-8), 757–777.