DIBR (Defining Interrelationships Between Ranked Criteria)

DIBR is a weighting method in which the expert ranks criteria from most to least important and states the relative share only between pairs of criteria that follow one another directly. Its output is a crisp weight vector summing to one; it does not rank alternatives and does not evaluate them. Pamučar and colleagues (2021) proposed it to substantially reduce the n(n-1)/2 pairwise comparisons AHP requires; it has been used in urban-transport and environmental applications.

In classical AHP the expert compares every pair of criteria against one another; as the number of criteria grows, the number of comparisons grows rapidly, and it becomes harder for the expert to stay consistent. DIBR's idea is that comparing only consecutive pairs, not every pair, is sufficient: the expert first ranks the criteria, then gives only consecutive ratios such as "how much more important is the first criterion than the second" and "how much more important is the second criterion than the third." Along the way, every other ratio, such as that between the first and third criteria, is calculated by chaining these consecutive shares together.

This carries a philosophical consequence: DIBR guarantees consistency structurally, rather than asking it of the expert. In AHP, an expert can make inconsistent comparisons, and this is checked afterwards with a consistency ratio; in DIBR, because only consecutive ratios are asked for and everything else is derived from them, the possibility of a contradictory comparison is removed from the outset. The cost of this is that it is deemed sufficient for the expert to know only the relationship between neighbouring criteria; the expert's actual perception of, say, the first criterion against the third is never asked, it is assumed via the chained ratio.

The method proceeds through three steps.

First, ranking. The expert arranges the criteria in a complete order from most to least important. This order is the skeleton for every subsequent step.

Second, consecutive share. The expert is asked, only for each pair of criteria that follow one another directly, to state the next criterion's share relative to the previous one; this share is expressed as "the second criterion is worth this much of the first." Every remaining ratio between criteria, for pairs that are not consecutive, is calculated by chaining (multiplying together) these consecutive shares.

Third, normalisation. The relative weights derived by chaining are divided by their own total to convert them into crisp criterion weights summing to one; the ranking set at the outset is preserved.

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

A weight reflects the ranking the expert set and only the shares given between consecutive criteria; it does not directly ask about the real perceived difference between two criteria that sit far apart in the ranking, the first and the fourth, say, it calculates this by chained multiplication. For this reason, the size of a weight depends not only on that criterion's own consecutive share but also on its position in the ranking: even a small difference in share near the top of the ranking can, through chained multiplication, turn into a large difference in weight further down.

Thus instead of writing:

"DIBR proved this criterion to be three times more important than the others"

the report should read:

"In the ranking the expert set, this criterion sits at the top and receives the highest share under the weight derived by chaining the consecutive shares; the ratios between non-adjacent criteria were not asked directly, they were calculated from the consecutive shares"

DIBR works with crisp (numerical) data: a ranking, and share values for pairs of consecutive criteria. DecisionMind holds no separately registered extension member within DIBR's own family; a second-generation method known in the literature as DIBR II is a separate source. You need: a complete importance ranking of the criteria, and a share value for every consecutive pair. This share value is expected to lie between zero and one half, because a criterion that follows next in the ranking can never be considered more important than the one before it. DIBR produces weights; it does not ask for weights from outside. Between three and twelve criteria works comfortably; as the number of criteria grows, it becomes harder to get the ranking itself right, particularly for placing criteria of similar importance in the correct order.

Where an expert can readily rank criteria but has neither the time nor the patience to compare every pair one by one, DIBR is a fitting choice; it requires far fewer questions than AHP. Where the expert also wants to express a genuine perception of the relationship between criteria that sit far apart in the ranking, not only consecutive ones, DIBR does not use this information, and a method that asks about all or selected pairs, such as AHP or BWM, should be preferred instead. Where a clear order of importance cannot be established among the criteria, because several criteria hold equal importance, say, DIBR's requirement of a "definite ranking" is strained.

Criteria can be readily ranked, weights are wanted with few questions → DIBR

All pairs, or a most-important/least-important comparison, are wanted → AHP, BWM, FUCOM

The ranking is not clear-cut, criteria are very close to one another in importance → methods such as SWARA should be reconsidered

Data is crisp, weight should come from the data's own variability → Entropy, CRITIC

DIBR's most important strength is speed: for n criteria, only n-1 share questions are needed, far fewer than the n(n-1)/2 comparisons AHP requires. Because less information is asked of the expert, the time taken shortens and expert fatigue falls. Consistency being guaranteed structurally, derived by chaining consecutive ratios, removes the need, as in AHP, for a separate consistency check and repeated rounds of comparison.

The method's fundamental limitation is that it uses only the information between consecutive criteria; the real perceived relationship between two criteria that sit far apart in the ranking is never asked, it is assumed through chained multiplication. This assumption may not match the relationship actually in the expert's mind. Second, if the position of two criteria in the ranking is reversed even once, the entire chained calculation is wrong from the start; DIBR cannot catch this error on its own, because its consistency check looks at the consecutive shares, not at the correctness of the ranking. Third, the method is newer than AHP; how closely its comparative behaviour matches other weighting methods has so far been examined in only a limited number of studies.

The most frequent mistake is giving a consecutive share value greater than one half or equal to zero; by definition, the next criterion in the ranking is considered less important than, or at most equal to, the one before it, and stepping outside this bound contradicts the ranking. A second mistake is setting the ranking not by consulting the expert but by the analyst's own assumption; the ranking is the foundation of DIBR's entire chained calculation and must come from the expert. A third is mentally factoring the relationship between non-adjacent criteria into a consecutive share while setting it, "adjusting" the share accordingly; DIBR asks only about the consecutive relationship, and derives the distant relationship itself, by chaining.

The governing principle is this:

The DIBR weight is derived by chaining from the ranking the expert gave and only the consecutive shares; this chained assumption must be stated clearly in the report, because the real perception between distant criteria was never asked at all.