PAMSSEM I (Procedure for Multiple Criteria Aggregation for Mixed Evaluations)

PAMSSEM I is an outranking method for ranking alternatives ("which supplier", "which project", "which region") once you hold a decision table scored on several criteria. Its output is a preference-flow value for every alternative, but the ranking is not complete: the method declares, for some pairs of alternatives, "these two are incomparable," and does not force them into an order. PAMSSEM I emerged from work in the 1990s by Martel and colleagues to combine the concordance-discordance-veto logic of the ELECTRE family with the flow logic of the PROMETHEE family. The clearest and most verifiable publication of these steps is the paper Ben Amor, Jabeur and Martel published in 2007, introducing the multiple criteria aggregation procedure for mixed evaluations.

PAMSSEM I's underlying idea is to bring two different notions of outranking under one roof. On one hand it asks the ELECTRE family's question: does one alternative genuinely outrank another, or is it so poor on one criterion that this single weakness invalidates everything else? This second question is answered through the veto threshold. On the other hand it carries the PROMETHEE family's view: it computes how much each alternative outranks the others on average (outgoing flow) and how much it is outranked by them (incoming flow).

This dual structure carries a philosophical consequence. PAMSSEM I is neither fully compensatory like TOPSIS nor a strict elimination method like classical ELECTRE. It allows limited compensation between criteria, but halts that compensation once the veto threshold is crossed. It is also an honest method: if two alternatives send mixed signals relative to each other, one good in one direction, the other good in another, the method does not force an order on them; it prefers to say "I cannot decide about these two."

The method proceeds through seven steps.

First, pairwise local comparison. For each criterion, the difference between two alternatives is computed and signed according to the criterion's direction. If the difference is small (below the indifference threshold), the local comparison is zero; if it is large (above the preference threshold), the local comparison is one; between the two, an intermediate value is taken in proportion to the difference.

Second, concordance. These local comparisons are weighted-summed using the criterion weights to yield a degree of concordance. A high degree of concordance means most criteria prefer the first alternative to the second.

Third, discordance. For each criterion, how far the second alternative falls behind the first is scaled against that criterion's full range in the table. This answers the question: "on the criterion being lost, how large is the loss."

Fourth, comprehensive outranking degree. DecisionMind applies a veto gate at this step: if the user has specified a veto threshold and a reverse difference on one criterion exceeds it, the outranking degree is reset to zero; the first alternative can never outrank the second in any way. If the veto threshold is not crossed, the comprehensive outranking degree equals the concordance degree directly. If no veto threshold has been specified at all, the default, this gate never closes, and outranking always remains equal to the concordance degree.

Fifth, outgoing and incoming flow. The average of how much each alternative outranks all others gives its outgoing flow; the average of how much the others outrank it gives its incoming flow.

Sixth, net flow. Incoming flow is subtracted from outgoing flow.

Seventh, partial ranking. One alternative outranks another only when it is equal to or better in outgoing flow, equal to or better in incoming flow, and strictly better in at least one of the two. If two alternatives send mixed signals, one ahead in outgoing flow, the other ahead in incoming flow, no order is established between them; these alternatives remain incomparable.

The formulas behind each step are given on the DecisionMind method page; this card carries no formulas.

Net flow shows how far an alternative stands out from the others on average; it is not a percentage or a degree of certainty. A positive value shows the alternative stands out on the whole; a negative value shows it lags behind. But PAMSSEM I's real distinguishing output is not net flow but the partial ranking: which pairs are actually comparable, and which remain incomparable.

Thus instead of writing:

"PAMSSEM I ranked all the alternatives"

the report should read:

"PAMSSEM I ranked these pairs relative to one another and could not decide about these pairs; net flow is only an additional reference value"

PAMSSEM I works with crisp data: one number per cell. DecisionMind holds no separate data-type extension of this method; it stands alone in its base form.

You need: alternatives in rows, criteria in columns, one number per cell; direction information for every criterion; criterion weights summing to 1. You also need indifference and preference thresholds for every criterion; if unspecified, DecisionMind uses extreme values such as zero and one, which reduces the local comparison to a simple "greater or smaller" question. The veto threshold is optional but important: if unspecified, no criterion can apply a veto, and the method runs without the safety mechanism it was designed to have. PAMSSEM I does not produce weights, it takes them from outside.

PAMSSEM I is a sound choice if your criteria are numerical, you can accept limited compensation, but you do not want a very poor performance on a particular criterion ever to be covered by the other criteria, and you want it stated explicitly whenever two alternatives cannot be decided between.

It should not be used in the following cases: if you cannot justify the veto threshold, the method runs with a safety mechanism that quietly falls out of service. If you need a complete ranking, PAMSSEM I does not guarantee one; if full compensation is acceptable, a simpler method is sufficient.

Limited compensation, a justifiable veto threshold, partial ranking acceptable → PAMSSEM I

A complete ranking is essential → PAMSSEM II or a fully compensatory method

Full compensation acceptable, no veto needed → TOPSIS, SAW

Weights, not thresholds, are needed → AHP, BWM, SWARA (subjective); Entropy, CRITIC (objective)

PAMSSEM I's greatest strength is its honesty: it does not force a ranking on alternative pairs that send mixed signals, and so does not make a decision look more certain than it is. Once a veto threshold is defined, it prevents an unacceptable weakness on a single criterion from being masked by the other criteria, a protection that fully compensatory methods (TOPSIS, SAW) cannot offer. Using both concordance and flow calculations together gives a richer picture than methods relying on only one logic, veto alone or flow alone.

Its limitations stem from its dependence on parameters. First, the result depends directly on the indifference, preference and veto thresholds; if these thresholds are chosen without justification, the result becomes arbitrary. Second, if the veto threshold is unspecified, the method quietly falls out of service, and the user may have trusted a safety mechanism that in fact never operated. Third, the partial ranking does not always answer the decision-maker's question of "who is better"; some pairs remain uncertain, and this must be resolved with an additional method, such as PAMSSEM II. Fourth, the common limitation of outranking methods applies here too: when the alternative set changes, the concordance and discordance degrees are recomputed and the ranking can be affected (Roy, 1991).

The most common mistake is assuming, without ever specifying a veto threshold, that PAMSSEM I will "automatically eliminate poor alternatives"; if no threshold is given, no elimination happens at all, and the method falls back on a purely weighted concordance calculation.

A second mistake is reading a pair left incomparable in the partial ranking as "equal"; incomparability does not mean equality, it means a mixed signal. A third is leaving the indifference and preference thresholds at their default extreme values without examining the data; this reduces the local comparison from a fine gradation to a crude "does it pass or not" question. A fourth is presenting net flow alone as the final decision and failing to report the uncertainties the partial ranking reveals.

The governing principle is this:

A PAMSSEM I result is a reflection of the thresholds you set; if the thresholds are unjustified, the result is unjustified too, and the incomparable pairs in the partial ranking must not be hidden from the report.