LBWA (Level-Based Weight Assessment)

LBWA is a method that assigns weight to criteria by expert opinion. Its output is a weight vector that sums to one; it does not evaluate alternatives and does not produce a ranking. AHP asks for a list of pairwise comparisons that grows rapidly as the number of criteria increases. LBWA instead asks the expert for a single ordering operation: first, identify the most important criterion; then split the remaining criteria into a few coarse levels (first-degree important, second-degree important, third-degree important, and so on). The final step is only to score the small differences within the same level. Proposed by Žižović and Pamučar in 2019, the method is preferred in fields such as multi-criteria supplier selection, sustainability assessment and public-policy prioritisation, because its comparison burden barely grows as the number of criteria increases.

The idea behind LBWA is that people are more comfortable grouping criteria roughly than making fine-grained pairwise comparisons between them. Instead of asking an expert "which matters more, speed or safety, and by how many times," LBWA asks which criterion is the most important. It then asks which criteria sit almost at the same level as that one (first level), which sit one step less important (second level), and so on down this order. A small distinction is also made within each level: one of two criteria at the same level may be slightly more influential than the other. This two-stage grouping is closer to human intuition than AHP's fully ordered pairwise comparison.

This has a consequence: in LBWA, the entire comparison burden gathers around the most important criterion. The other criteria are related not directly to one another but only to the most important criterion and to their own levels. This makes the method fast and undemanding, but it also makes it heavily dependent on the most important criterion having been chosen correctly; if that choice is wrong, the whole weight series is built on that error.

The method proceeds through three steps.

First, level and influence assessment. The expert first selects the most important criterion from among all of them; this criterion is accepted at the first level with an influence score of zero. The remaining criteria are distributed across increasing levels, first, second, third, and so on, as their importance falls further behind the most important one. For criteria at the same level, the expert gives a small integer influence score (from 0 up to the number of criteria at that level) showing how far it stands from the most important criterion. In this step a scale-width parameter is also set; this parameter is chosen larger than the highest level number.

Second, the influence function. For every criterion, its level and its influence score within that level are converted, by a formula, into a single figure, the influence value. As the level number grows, as the criterion moves further from the most important one, this value shrinks; within the same level, the value also shrinks as the influence score grows.

Third, converting to weight. The most important criterion's weight is calculated using an anchor value derived from the sum of all the other criteria's influence values. The other criteria's weights are found by multiplying their own influence values by this anchor. In the end, all the weights sum to one.

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

The weight is a direct consequence of the level-and-influence structure the expert built around the most important criterion; it reflects not the fine differences between criteria but the rough groups the expert drew. If two criteria sit at the same level with close influence scores, their weights also come out close to each other; this closeness means the criteria were seen as "almost equally important," not evidence of a fine mathematical difference. The most important criterion's weight is always the highest, because the method is built by definition to make it so; this is not a computed result but a direct reflection of the decision the expert made in the first step.

Thus instead of writing:

"LBWA proved that this criterion is the most important"

the report should read:

"The expert selected this criterion as the most important; LBWA converted this choice, together with the remaining criteria's level-and-influence structure, into a consistent weight series"

LBWA works with crisp data: level numbers and influence scores are integers. DecisionMind carries no separately registered extension member in LBWA's own family; although fuzzy, grey and interval rough number extensions have been defined in the literature, these do not appear as a separate member in DecisionMind. You need at least two criteria, a clear choice of which criterion is most important, each criterion's level, the influence scores within each level, and the scale-width parameter. The method produces weight and does not ask for weight from outside. Three to twelve criteria is typical; even as the number of criteria grows, the comparison burden does not grow as fast as with other methods, because every criterion is evaluated only with a small score within its own level.

If the number of criteria is large and asking the expert for a full pairwise comparison is impractical, while the expert can nonetheless split the criteria roughly into a few importance groups, LBWA is a suitable choice. If the expert can already make clear ratio comparisons between criteria, such as "this one matters three times as much as that one," AHP or BWM gives a more detailed weight structure. If the most important criterion itself is disputed, that is, if the experts cannot agree which criterion is most important, the method becomes fragile, since LBWA's entire structure rests on that single choice. In that case, agreement on the most important criterion should first be reached through a consensus method such as Delphi.

Many criteria, a rough level grouping is enough → LBWA

Few criteria, a fine ratio comparison is wanted → AHP, BWM

The most important criterion is disputed, consensus is needed first → Delphi, Fuzzy Delphi

Not weight but the data's own variability matters → Entropy, CRITIC

LBWA's most important strength is its speed: as the number of criteria grows, the comparison burden grows far more slowly than AHP's n(n-1)/2 pairwise comparisons, because criteria are compared only within their own level. The judgement asked of the expert is simple and easily justified: "this criterion is the most important," "these three sit almost at the same level as it." The method also includes a consistency check and, through the scale-width parameter, allows fine-tuning of how sharply the results separate.

The method's greatest fragility is that its entire structure rests on a single criterion, the most important one; if this choice is wrong, or disputed among experts, the whole remaining weight series is built on that error. Second, parsimonious methods that ask for few comparisons, like LBWA, BWM and FUCOM, use a precise numerical scale; this cannot capture the linguistic uncertainty, the hesitancy, in expert judgement (Ayan, Abacıoğlu and Basilio, 2023). Third, the choice of the number of levels and the scale-width parameter is left to the expert, and this choice can change the result; the same set of criteria with a different number of levels can produce different weights. Fourth, LBWA defines no built-in aggregation rule for multi-expert assessments; if more than one expert is involved, a consensus must first be reached.

The most frequent mistake is choosing the most important criterion hastily without noticing that this choice determines the entire result; if the choice is disputed, the method does not sit on solid ground from the start. A second mistake is being inconsistent when splitting criteria into levels; for instance, if two criteria are seen as similarly important they should be placed at the same level, and spreading them across different levels creates an artificial difference. A third mistake is choosing the scale-width parameter by looking at the results; this parameter must be fixed before the analysis begins. A fourth mistake is behaving as though LBWA itself can combine more than one expert's level-and-influence structure; the method contains no multi-expert aggregation rule.

The governing principle is this:

LBWA's weight is a direct reflection of the level-and-influence structure the expert built around the most important criterion; the more disputed that structure is, the more disputed the weight is.