Multi-Criteria Decision Analysis

Status: Folded · Evidence: P · Family: Decision and option evaluation · Verdict: fold (2026-06-03)

Use instead: Decision Option Review

What it is

Multi-Criteria Decision Analysis (MCDA, also MCDM, and in its simplest aggregation the weighted sum model / Simple Additive Weighting) is the decision move of laying options against criteria in a matrix, weighting the criteria by importance, scoring each option on each criterion, and combining the weighted scores into a per-option total so the alternatives can be ranked on a single number. The deliverable is a weighted decision matrix: rows of options, columns of criteria, a weight per column, a score per cell, and a roll-up total per row. The claimed payoff is that a choice with several pulling-in-different-directions considerations stops being argued by gut or by whoever is loudest, and instead becomes an explicit, inspectable structure where the trade-offs are written down and the ranking falls out of stated weights and scores.

The honest description has to separate two things the literature blurs. The durable underlying move is “decompose a choice into weighted criteria, score the options against them, and aggregate.” That move is one thing, and it is the same thing whether it wears the name MCDA, weighted scoring matrix, decision matrix, or grid analysis. The popular packaging is a sprawling family of named aggregation algorithms - Simple Additive Weighting, the Analytic Hierarchy Process (AHP), TOPSIS, ELECTRE, PROMETHEE, VIKOR, MAUT/MAVT - each a different recipe for normalizing scores and combining them. Those algorithms differ in the math, not in the cognitive operation. The operation common to all of them, and the only part a thinking library needs to ship, is: name the criteria, weight them, score the options, total, and read the trade-off. The rest is normalization machinery.

When it helps / when it misleads

It helps when a decision genuinely has several criteria that matter at once and no single one dominates - a vendor choice on cost, fit, risk, and support; a hire on multiple competencies; a roadmap call balancing reach, effort, and confidence. Writing the criteria and weights down forces the implicit trade-off into the open, gives a team a shared structure to argue inside (rather than across), and produces an artifact a reviewer can challenge cell by cell. As a discipline against “we just went with our gut,” it earns its keep.

It misleads or wastes effort when:

• The number is mistaken for the answer. A weighted total looks objective and is not. Every total is a function of weights and scores that a human chose, often after a glance at which option they already preferred. The matrix can launder a predetermined choice into a defensible-looking score, and the false precision of a 7.3-versus-6.9 total invites decisions that the underlying judgments cannot actually support.
• The criteria interact or are not independent. Additive aggregation assumes the criteria trade off linearly and independently. When a low score on one criterion should veto an option regardless of the others (a deal-breaker), or when two criteria are really measuring the same thing (double-counting), the weighted sum quietly gets it wrong.
• The ranking is unstable. The choice of normalization and aggregation method can flip the ranking on identical inputs, and adding or removing an option can reorder the survivors - the documented rank-reversal problem (below). A total that changes when you change the arithmetic, not the facts, is not a reliable verdict.
• The decision does not actually have multiple live criteria. If one criterion dominates, use it. If the real question is reversibility, that is a different move; if it is pricing the uncertainty of outcomes, that is a decision tree; if you cannot put options on an absolute scale at all, pairwise comparison ranks them without one. Reaching for a full weighted matrix on a one-criterion or two-option call is ceremony.

What the evidence says

The honest governing grade is P (practitioner). MCDA is a real, named, fifty-year-old decision-analysis family with a deep formal lineage and heavy institutional use (health technology assessment, environmental and infrastructure appraisal, procurement). What it does not have is controlled evidence that running a weighted-criteria matrix produces better decisions than the alternatives, and the literature is unusually candid about its own soft spots.

What the record supports. The method is well-founded as a normative structure: Fishburn (1967) and MacCrimmon (1968) established additive weighting; Keeney and Raiffa (1976) gave it an axiomatic basis in multi-attribute utility theory. It is widely adopted and taught, and authoritative bodies endorse it as good practice. The ISPOR MCDA Emerging Good Practices Task Force (Thokala et al. 2016; Marsh et al. 2016) states that “structured, explicit approaches to decisions involving multiple criteria can improve the quality of decision making.” That is the high-water mark of the supporting claim, and its verb is can: it is an expert task-force position on process quality and transparency, not a controlled outcome study showing the matrix beats unaided judgment.

What the record does NOT support. There is no body of controlled or comparative trials I can locate showing that MCDA scoring yields more accurate or better-outcome decisions than a structured discussion without the matrix. Worse, a substantial methodological literature documents that the answer is method-dependent: Triantaphyllou and Mann’s experiments on hypothetical multi-criteria problems found that different MCDM methods produced different rankings on the same data in a majority of test cases (the “decision-making paradox”), and Belton and Gear (1983) showed that AHP in its standard mode could reverse the ranking of existing options merely by introducing a near-duplicate alternative. The rank-reversal phenomenon is reviewed at length in the MCDM literature (for example Maleki and Zopounidis, and the Brazilian Pesquisa Operacional review). This is a stronger caution than mere absence of evidence: it says the tool can return different verdicts depending on the arithmetic you pick, which bounds how much weight any single total deserves.

Transfer caveat (required). All of the supporting record - the task-force endorsements, the application studies, the formal results - comes from human analysts and human decision bodies. None of it studies an MCDA matrix produced by or with an AI agent. The evidence is transferred from human organizational contexts and is not validated for AI-augmented use; the conservative grade is P.

Excluded figures (required). Vendor and blog write-ups routinely attach percentage improvements to MCDA tooling (“better decisions by N%,” “reduces decision time by X%”). I could trace none of these to a nameable primary study, so under this library’s evidence rule they are excluded as fact and do not influence the grade. The Triantaphyllou and Mann “majority of cases reversed” figure is reported here only as a limitation finding (rankings disagree across methods), not as a performance claim, and the rank-reversal literature should be read as the bound on the method’s reliability, not as a number in its favor.

Why it is / is not a skill here

Verdict: Fold into decision-option-review. The registry reasoning is blunt - “Subsumed: equivalent to decision-option-review” - and the fold is a structural identity rather than a close call.

The Build burden is to name one distinct, durable cognitive move that no shipped skill already produces above the roughly 20% overlap ceiling. MCDA’s move is “compare options against weighted criteria, score, total, rank.” That is, verbatim, what think-decision-option-review ships: its one-line is “compare options against weighted criteria,” and its alias is literally “Weighted decision matrix.” The shipped skill’s own registry note states that it “Absorbs Multi-Criteria Decision Analysis.” There is no daylight between the candidate’s mechanism and the shipped skill’s mechanism - the options-by-weighted-criteria matrix is the same artifact, built by the same steps, read the same way. Overlap is at or near 100%, not below the ceiling.

The only thing MCDA brings that a generic weighted matrix does not is the aggregation algorithm choice (SAW versus AHP versus TOPSIS versus PROMETHEE). But that is precisely the part the evidence section flags as unstable and method-dependent, and it is normalization machinery, not a separate cognitive operation. A library that ships distinct thinking moves does not ship a second weighted-matrix skill in order to expose a menu of normalization formulas whose main documented property is that they disagree with each other. Picking an aggregation recipe is a configuration of the move, not a new move.

The neighbors confirm there is nothing left for a standalone MCDA skill to own. The AHP pairwise variant - rank by judging items two at a time when you cannot put them on an absolute scale - is a genuinely different operation, and the library already ships it separately as think-pairwise-comparison (which names AHP among its aliases). So MCDA’s one arguably-distinct sub-method already has its own home, and the remaining, dominant move is the weighted matrix, which decision-option-review owns outright.

Why fold rather than reject. The move is real, useful, and worth locating, so the honest service is to point the reader to where it already lives rather than to dismiss it. And the fold sharpens, not weakens, the shipped skill: decision-option-review’s enrichment lineage (the Questions/Options/Criteria design-rationale work of MacLean, Young, Bellotti and Moran 1991) reinforces exactly the caveat MCDA most needs - record why a cell scores as it does, not only the total - which is the discipline that guards against the false-precision and rank-instability failure modes above. The learning value of the NO: a famous, formally-deep, institutionally-endorsed decision family can still be a fold, because its fame is in the aggregation algorithms (AHP, TOPSIS, PROMETHEE) while its cognitive move is the plain weighted matrix the catalog already ships. Folding it keeps the catalog from shipping a second decision matrix under a more impressive acronym.

Lineage and who to read

MCDA grew out of operations research and decision theory in the 1960s and 1970s, as the field moved from chasing a single mathematically “optimal” answer toward finding a “preferred” option that explicitly carried human values and trade-offs. The additive weighting core was set out by Peter C. Fishburn in 1967 and surveyed and consolidated by Kenneth R. MacCrimmon in a 1968 RAND memorandum; Ralph Keeney and Howard Raiffa gave the family its axiomatic foundation in multi-attribute utility theory in 1976. Thomas L. Saaty developed the pairwise-comparison branch, the Analytic Hierarchy Process, around 1980. The European “outranking” school (ELECTRE, PROMETHEE) and methods such as TOPSIS and VIKOR added alternative aggregation algorithms. For the critical read, Triantaphyllou and Mann and the rank-reversal literature show the rankings are method-dependent, and Belton and Gear (1983) first surfaced AHP rank reversal; Valerie Belton and Theodor Stewart’s textbook (2002) is the standard balanced treatment. For the institutional good-practice position (and its careful “can improve” hedging), read the ISPOR MCDA Task Force reports (Thokala et al. 2016; Marsh et al. 2016). “Multi-Criteria Decision Analysis,” “MCDA,” and “MCDM” are generic descriptive terms in common use - no trademark, no single owner - so this entry is documented descriptively and is not flagged as branded.

Named sources

• Peter C. Fishburn, “Additive Utilities with Incomplete Product Sets: Application to Priorities and Assignments,” Operations Research 15(3) (1967): 537-542. Establishes the additive-utility (weighted sum) representation that underlies SAW/MCDA. Foundational. (P)
• Kenneth R. MacCrimmon, Decisionmaking Among Multiple-Attribute Alternatives: A Survey and Consolidated Approach (RAND Corporation, RM-4823, 1968). Surveys and consolidates additive weighting and the other multi-attribute methods; an early naming of the family. Foundational. (P)
• Ralph L. Keeney and Howard Raiffa, Decisions with Multiple Objectives: Preferences and Value Tradeoffs (Wiley, 1976; Cambridge University Press, 1993). The axiomatic foundation in multi-attribute utility theory. Foundational. (P, normative theory)
• Valerie Belton and Theodor J. Stewart, Multiple Criteria Decision Analysis: An Integrated Approach (Kluwer, 2002). The standard balanced textbook; documents the family and its limits, including aggregation and weighting pitfalls. Practitioner/textbook. (P)
• Evangelos Triantaphyllou (with B. Mann and later work), studies of the “decision-making paradox” - different MCDM methods produce different rankings on identical inputs in a majority of test problems. A limitation finding on method-dependence, not an effectiveness finding. Critical literature. (P, critical)
• Valerie Belton and Tony Gear, “On a Short-coming of Saaty’s Method of Analytic Hierarchies,” Omega 11(3) (1983): 228-230. First demonstration of AHP rank reversal when a near-duplicate alternative is added. Critical literature. (P, critical)
• Praveen Thokala, Nancy Devlin, Kevin Marsh et al., “Multiple Criteria Decision Analysis for Health Care Decision Making - An Introduction: Report 1 of the ISPOR MCDA Emerging Good Practices Task Force,” Value in Health 19(1) (2016): 1-13 (with Report 2, Marsh et al.). The authoritative good-practice position: structured multi-criteria approaches can improve decision quality. An expert task-force endorsement of process, not a controlled outcome trial. (P, practitioner-consensus)

Excluded under the evidence rule: vendor and blog “MCDA improves decisions / saves time by N%” figures trace to no nameable primary source and are excluded; the Triantaphyllou and Mann “majority of cases reversed” result is reported only as a method-dependence limitation, not as a performance number, and is not counted toward the grade.