Project
Open one of the verified example problems, open your own file, or start a new problem from scratch. Then give it a name and, if you like, some notes.
Open an example
The 23 verification problems of the dissertation and the published studies. Each opens with its criteria, weights and values.
Open your own file
A project file (.mcewr) saved here or in the desktop edition, a spreadsheet export (.csv), or a 1995 .MCE file.
Start a new problem
An empty table with the number of alternatives and criteria you choose (all criteria maximizing, weight 1). You fill in the names, weights and values in the next steps.
How the evaluator works
- Define the problem (steps 1 to 4): the alternatives, the criteria with their weights and limits, and the decision matrix of values.
- Set up the evaluation (steps 5 and 6): how the criteria are weighted, which method and which settings.
- Read the results (step 7) and, when needed, test them (steps 8 to 10): compare all methods, vary the weights, check for rank reversal.
Every number shown is produced by the verified MCEWR engine. The steps are also explained in the Web Tutorial and the Quick Start.
Alternatives
The options to be ranked: projects, sites, technologies, management plans. Name them, add or remove them, and untick used to leave one out of the evaluation without deleting it.
Alternatives
Edits are stored as soon as you leave a cell (Enter or Tab). A problem needs at least two alternatives.
Criteria
Each criterion has a direction (max or min), a weight, and the settings some methods need: the Upper and Lower limits (Compromise Programming, ELECTRE II), the PROMETHEE preference function with its thresholds p and q, and the ELECTRE III veto. Criteria can be grouped in a hierarchy: add a sub-criterion to make a criterion aggregated.
Criteria
The share column is the weight divided by the sum of the weights of its group - what the methods use when the weights are normalized (step 5). Saaty marks weights or values given as pairwise comparisons in the file; they are used by AHP and cannot be edited here.
Decision matrix
The value of every alternative on every leaf criterion, in the units of the criterion. Rows are alternatives, columns are the leaf criteria (with their direction: ↑ larger is better, ↓ smaller is better).
Values
Enter moves down, Tab moves right. You can paste a block of numbers copied from a spreadsheet into a cell.Weights
Choose how the criteria are weighted in the evaluation: the weights you entered, equal weights, or weights derived from the data (entropy, CRITIC, MEREC). The table below shows, for every group of criteria, the weights each choice gives.
Weighting method
Ticked (the usual choice): weights 4, 2, 4 become 0.4, 0.2, 0.4. Unticked: the weights are used exactly as given - only for files whose weights are already meaningful as they are.
The weights each method gives
Per group of sibling criteria; each column sums to 1 within a group. The chosen method is highlighted.Method and settings
Choose the evaluation method, how the values are standardized, and the options of the method. Then run the evaluation.
Evaluation method
Standardization of the values
Options of the method
Advanced: text-book switches and exclusions
All switches off = the behaviour of the 1995 program (the audited results). Each switch applies the text-book formulation instead.
Ready to run
The result appears in step 7. Running again with other settings replaces it.
Results
The ranking of the alternatives under the chosen method, the ranking value of each, and the domination percentage (how far each is ahead of the next). Download the report or go on to the analyses.
Compare the methods
Run several methods on the same problem and see where they agree: the rank of every alternative under every method, the mean rank, a consensus order, and Spearman's rho between the rank vectors.
Methods to compare
The standardization, the weights, the switches and the method options of steps 5 and 6 apply to every method.
Sensitivity
How much can one input change before the winner changes? Each analysis moves one input at a time - a weight, a value of the decision matrix, or a setting of the method - re-runs the method of step 6 at every step, and records the winner. The critical points are located by bisection.
The weights of one group
The share of each criterion in the total weight of its group is moved from 0 to 1 while the other criteria keep their proportions. The result: the winner in every interval of the share, and the critical share - the smallest change of that weight that changes the decision. The criterion with the smallest relative change is the most sensitive.
The values of the decision matrix
Each chosen value is moved from value - range to value + range of its criterion (the range over the alternatives) in equal steps. The result: the winner in every interval, the nearest critical value, and the change it represents as a percentage of the criterion's range. The cell with the smallest change is the most sensitive value of the study.
"All alternatives" runs every cell of the matrix (at most 400 cells); with a large problem choose the winner only.
The settings of the method
One numeric setting - the standardization code, or an option of the method such as the VIKOR v, the CP power or the ELECTRE III thresholds - is given several values; the method is re-run for each. The result: the winner and the whole order for every value, and Spearman's rho of the positions against the base run (1 = the same positions).
Robustness (Monte Carlo)
How stable is the ranking when the data and the weights are uncertain? Every value of the decision matrix, and if you wish every weight, is multiplied by a random factor 1 + u; the method is re-run many times and the positions are counted. The result is the rank acceptability of every alternative - how often it takes each position - and the winner stability.
Settings
0 % switches a perturbation off. The triangular distribution makes small errors more likely than large ones. The seed makes the drawings repeatable: the same seed gives the same table on any computer. Values given as Saaty comparisons and weights given as Saaty priorities are not perturbed.
Rank reversal
Does the order of the alternatives change when one is removed, duplicated, or when a dominated one is added? A method that keeps the order is rank-reversal free on this problem; Compromise Programming with the provided limits (SPOTIS) is so by construction.
Methods
Tests
remove: each alternative left out in turn; duplicate: a copy of each added; dominated: one alternative worse than all the others added.