MEREC (MEthod based on the Removal Effects of Criteria)

MEREC is not a ranking method; it does not rank alternatives, it produces criterion weights. Looking at the decision table, it gives every criterion a weight, summing to 1, and these weights then feed into a ranking method such as TOPSIS, VIKOR or SAW. Like Entropy and CRITIC, it is an objective weighting method, but it rests on a different logic. It does not look at a criterion's own spread or its relationship with other criteria; it looks at how much the alternatives' overall assessment would be disrupted if that criterion were removed from the table entirely. Keshavarz-Ghorabaee, Amiri, Zavadskas, Turskis and Antucheviciene proposed the method in 2021; being recent, its literature is still limited.

MEREC's underlying idea is to measure a criterion's importance by asking "what do we lose without it." An overall performance value is first computed for every alternative with all criteria present. Each criterion is then removed in turn, and the overall performance value is recalculated without it. If the alternatives' overall performance changes a great deal when a criterion is removed, that criterion makes a large difference to the assessment. If it barely changes at all, that criterion contributes little distinguishing information to the table.

This idea reaches the same goal, deriving weight from data, by a different route from Entropy and CRITIC. Entropy looks only at a criterion's own internal distribution. CRITIC looks at distribution together with the relationship to other criteria. MEREC instead runs a direct "removal experiment" and measures a criterion's real effect on overall assessment. This approach refers directly to an outcome, overall performance, without separately computing the complex relationships between criteria, such as correlation.

The method proceeds through five steps.

First, logarithmic normalisation. Every cell is reduced to between 0 and 1 according to the criterion's direction, but on a logarithmic scale rather than the vector normalisation used in TOPSIS. This keeps the "overall performance" measure used in the following steps well defined mathematically.

Second, overall performance. A single overall performance value is computed for every alternative, using all criteria together. This value is a summary measure of how well the alternative fares in light of every criterion.

Third, performance with a criterion removed. Each criterion is removed from the table in turn, and the alternative's overall performance is recalculated with the remaining criteria. For n criteria this means n separate "incomplete table" calculations.

Fourth, removal effect. For each criterion, the total change across all alternatives' overall performance is summed when that criterion is removed. The larger this total, the greater the difference that removing the criterion has made.

Fifth, weight. Each criterion's removal effect is divided by the total removal effect across all criteria; the result is a weight vector summing to 1.

The formulas behind each step, the intermediate tables and the citation formats are given on the DecisionMind method page; this card carries no formulas.

A MEREC weight measures how much the overall assessment would be disrupted if a criterion were removed; it does not measure its importance in the decision-maker's eyes. A high weight means "without this criterion, the difference between alternatives largely disappears." A low weight means "even without this criterion, the overall picture stays largely the same." This is a different thing from the "spread" that Entropy measures and the "spread plus lack of correlation" that CRITIC measures. On the same data set the three methods can produce different weights, because all three define "importance" differently.

The weights are tied to the table: when the alternative set or criterion list changes, the removal effects are recalculated and the weights change. If all the alternatives sit very close together on a criterion, taking almost the same value, removing that criterion already changes overall performance little, and its weight comes out low. This is not an error; it is a sign that the criterion is not distinguishing within this alternative set.

Thus instead of writing:

"The MEREC analysis showed that cost is the most important criterion"

the report should read:

"Cost is the criterion that changes the overall assessment most when removed, within this alternative set; its weight of 0.37 reflects this removal effect, not the decision-maker's order of priority"

MEREC works with crisp data: one number per cell, and it accepts no zero or negative value, because logarithmic normalisation requires a positive value. You need alternatives in rows, criteria in columns, a positive number in every cell, no empty cells; and, for every criterion, whether more or less is better. If the direction is wrong, the meaning of the removal effect is reversed. No weight is entered; the method produces the weights. A minimum of two criteria is required, because at least one criterion must remain once one is removed; if all alternatives are identical across all criteria, the removal effect stays undefined. Alongside the base method, DecisionMind holds fuzzy and scenario-fuzzy extensions, three members in total. Because of the logarithmic compression, the result can be sensitive to changes in a single cell; this sensitivity should be reported in large data sets.

MEREC is a suitable choice if you want weights to come directly from data rather than expert opinion, and the question "what would we lose without this criterion" strikes you as more intuitive. This question offers a different view from CRITIC's correlation logic or Entropy's pure spread logic. It can be used in any field requiring objective weighting.

There are three cases where it should not be used. First, where your data contains zero or negative values, which logarithmic normalisation cannot accept. Second, where the number of criteria is very small; with two criteria, "removal" leaves only a single criterion behind and the effect measure becomes crude. Third, where the weights need to reflect expert opinion, such as strategic priority, a subjective method (AHP, BWM, SWARA) is more suitable than an objective one.

Weight should come from data, "removal effect" logic wanted → MEREC

Weight should come from data, spread alone is enough → Entropy

Weight should come from data, spread plus redundancy between criteria should also be counted → CRITIC

Weight should come from expert opinion → AHP, BWM, SWARA

Data contains zero/negative values → the data should first be transformed, or an objective method that does not require logarithms (CRITIC) should be chosen

MEREC's most important advantage is that it derives weight not through indirect statistical measures such as spread or correlation, but by directly answering the question "how much would the overall assessment change without this criterion." This is a logic that is easier to explain to a decision-maker than the weights produced by Entropy or CRITIC. The computational burden is small and the steps can be followed on the table. Because it does not compute correlation between criteria, it does not carry the "correlation runs to extremes" problem that CRITIC faces in tables with few alternatives.

Its limitations also follow from this structure. First, because the method was proposed in 2021, its literature is still limited; the body of independent review is smaller than for Entropy and CRITIC. Second, owing to logarithmic normalisation and logarithmic summation, the result can be disproportionately sensitive to a small change in a single cell. This is a property noted in the method's own literature and requires a sensitivity check in large data sets. Third, logarithmic normalisation does not accept zero or negative values; if the data set contains such values, they must first be transformed. Fourth, if all alternatives are nearly identical on a criterion, that criterion's removal effect comes out close to zero and so does its weight. This is a statistical outcome and should not be confused with the criterion's importance in the field (Saidin, Lee, Marjugi, Ahmad and Seow, 2023).

The most common mistake is applying MEREC without noticing that the data set contains zero or negative values; logarithmic normalisation is undefined in this case. A second mistake is marking a criterion's direction wrongly; if the direction is wrong, the direction of the removal effect is also calculated wrongly. A third mistake is eliminating a criterion that receives a low weight as "unimportant." A low weight shows that the criterion is not very distinguishing within this alternative set; it does not show that the criterion is unimportant to the field. A fourth mistake is applying MEREC with very few alternatives, for instance two, and presenting the result as a precise ratio; the removal effect is a crude estimate with so few alternatives. A fifth is comparing a MEREC weight directly against a CRITIC or Entropy weight and asking "which is correct"; the three methods use different definitions, and there is no single "correct" answer.

The governing principle is this:

A MEREC weight is a measure of how much removing a criterion disrupts the overall assessment. This concerns the criterion's distinguishing contribution within this alternative set, not its importance in the field, and it must be recalculated when the alternative set changes.