HELLWIG (Hellwig's Development Pattern Method)

The Hellwig method is a ranking method for when you hold units measured on several indicators ("which district is more developed," "which branch is in better shape") and want them placed in a single order. Its output is a development score close to the 0-to-1 range for every unit, together with the rank that score produces. Unlike methods such as TOPSIS or AHP, it requires no criterion weight; it draws a common scale out of the data itself and holds every indicator equal on it. The Polish statistician Zdzisław Hellwig proposed the method in 1968, and official statistical bodies have used it for over fifty years, especially in regional and cross-country development comparisons.

Hellwig's underlying idea is to judge a unit not against a fixed "perfect" target but against a "development pattern" built from the best values the group under study has itself achieved. On every indicator, the best value the group reaches is taken, and these are gathered into a single hypothetical unit that does not actually exist. Real units are considered more developed the closer they sit to this pattern. No weight is used, because the method first standardises every indicator against its own mean and spread; this way, an indicator with large figures, such as income, and one with small figures, such as a satisfaction score, are compared on the same scale, with an equal say.

This idea carries a philosophical consequence: Hellwig is compensatory, but it does not reward extremes. Because distance from the pattern is a single figure, weakness on one indicator can be papered over by strength on another. However, the method does not reward a unit that shines on a single indicator while trailing on the rest; it rewards the unit that stands balanced across all indicators, close to the pattern as a whole. Leaving out weights reduces subjectivity, but the decision of which indicators enter the analysis is itself a value judgement, and that decision lies outside the method.

The method proceeds through four steps.

First, standardisation. Indicators are expressed in different units: an income score, an education rate, an unemployment rate. Hellwig subtracts each column's own mean and divides by its own standard deviation. Every indicator is thereby carried onto a common scale, independent of its unit, with a mean of zero, so that columns become comparable with one another.

Second, building the development pattern. In the standardised table, the best value the group reaches is taken for every indicator: the highest standardised value for a "higher is better" indicator, the lowest standardised value for a "lower is better" indicator. These values are combined into a single hypothetical unit, a pattern that is best on every indicator and that no real unit may ever reach.

Third, distance from the pattern. Every real unit's straight-line distance (Euclidean distance) to this hypothetical pattern is computed: the difference on each indicator is squared, the squares are summed, and the square root is taken.

Fourth, the development score. The method sets a "sufficiently large" reference distance by adding the mean of all units' distances from the pattern to twice the standard deviation of these distances. Each unit's score is found by dividing the unit's distance by this reference distance and subtracting the result from one. A unit very close to the pattern approaches a score of 1, one that is far shrinks towards a lower score, and on rare occasions can even fall below zero. Units are ranked by this score from the highest to the lowest.

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

The development score tells you how close a unit sits to the pattern relative to the other units in this particular group; it says nothing more than that. A score of 0.44 does not mean "44 per cent developed." The score is not strictly confined to the 0-to-1 range: because the pattern and the reference distance are built from the group's own statistics, an extreme unit can very rarely exceed 1 or fall below 0. More importantly, the score is meaningful only within this group: because both the pattern and the reference distance are derived from the group's own mean and spread, it cannot be compared with the scores of another Hellwig analysis run on a different group of units.

The unit ranked first is often not the outright best on any single indicator; it is the one that is balanced across all of them. A unit that stands out sharply on one indicator while trailing on another can finish behind a unit that sits at a middling level on every indicator, because distance also penalises imbalance across indicators.

Thus instead of writing:

"Hellwig found the most developed unit"

the report should read:

"With these indicators and this set of units, the unit closest to the pattern is this one; the ranking is sensitive to adding or removing a unit from the set"

Hellwig works with crisp data: one number per cell. The method's most distinctive feature is that it uses no externally supplied weight; once standardised, every indicator carries an equal say. DecisionMind holds no extension of this method for other data types (fuzzy, grey, intuitionistic and so on); it stands alone in its base form.

You need: units in rows, indicators in columns, one number per cell with no empty cells; and, for every indicator, whether higher or lower is better. A minimum of two units and two indicators is required; three to twelve indicators work comfortably. If every unit carries the same value on an indicator (its standard deviation is zero), the standardisation step is left undefined; that indicator must either be dropped or the data reviewed. The number of units also matters: because the reference distance is built from the group's own mean and standard deviation, these statistics are not reliable with too few units.

Hellwig is a sound choice if your indicators can be measured numerically, your table is complete, and you either do not want to set criterion weights or find doing so contested. Its typical territory includes regional and cross-country development comparisons, institutional performance ranking, and statistical classification with large indicator sets; it is especially preferred where assigning weights would be politically or administratively contentious.

It should not be used in the following cases: if you hold expert opinion on the relative importance of the indicators and want it reflected in the analysis, a method that accepts weights (TOPSIS, SAW) is more suitable, because Hellwig cannot use this information. If the number of units is very small (three or four units, say), the reference distance becomes statistically unreliable. If one criterion can never be traded away, methods built on elimination logic should be used instead.

A numerical table, no wish to set weights, the goal is ranking → Hellwig

The importance of the indicators is known and is to be used → TOPSIS, SAW, VIKOR (weighted methods)

The weights themselves are also to be determined → AHP, BWM, SWARA (subjective); Entropy, CRITIC (objective)

Too few units → statistical standardisation is unreliable; use simple comparison or expert judgement instead

Hellwig's greatest strength is that it removes the debate over setting weights entirely. In comparisons across regions or institutions, which indicator should count as more important is often itself a political dispute; Hellwig closes this debate from the outset by standardising every indicator and holding them equal. The method has an application history of over fifty years and remains in use in official statistical bodies. It can work with relatively few indicators or with many; its computational burden is light.

Its limitations stem from the pattern and the reference distance both depending on the data. First, the development pattern and the reference distance are built from the statistics of the very set of units under study; adding or removing a unit from the set can change how the other units rank relative to one another. The literature has separately examined how Hellwig can give a different ranking compared with other ranking methods such as TOPSIS (Bąk, 2016). Second, although leaving out weights reduces subjectivity, the decision of which indicators, and how many, enter the analysis is itself a value judgement; this decision lies outside the method and must be reported. Third, because the standardisation step rests on the mean and standard deviation, an extreme value in the set can noticeably shift both the pattern and the reference distance (Bielak and Kowerski, 2019). Fourth, the method treats indicators as independent; this independence is implicitly broken where indicators influence one another.

The most common mistake is marking indicator direction wrongly. Mark a "lower is better" indicator such as the unemployment rate as "higher is better," and the pattern is built from the worst-performing unit, rendering the ranking meaningless.

A second mistake is reading the development score as a percentage or an absolute degree of success, and comparing scores across different analyses; the score is meaningful only within this set of units. A third is including an indicator whose standard deviation is zero (the same value across every unit); in that case the standardisation step is left undefined. A fourth is running the analysis with too few units and assuming the reference distance is reliable. A fifth is trying to use Hellwig as if it were a weight-accepting method, attempting to assign external importance to indicators; the method takes no account of this information.

The governing principle is this:

A Hellwig result is a summary of the shared statistical pattern of the units and indicators included in the study; if the set of units changes, the pattern changes with it, and the ranking can shift accordingly.