The Aggregation Problem: Why No Wellbeing Index Can Be Neutral

The Measurement Problem No One Talks About

Every year, governments, journalists, and policy analysts pore over a single number: a country's GDP. Gross Domestic Product tells us how much an economy produces. It does not tell us whether that production makes anyone happy, whether it depletes the natural resources future generations will need, or whether the gains flow to everyone or concentrate at the top.

We have known this for decades. Robert Kennedy's speech at the University of Kansas in 1968 captured it with characteristic elegance: GDP measures everything, in short, except that which makes life worthwhile. Yet GDP remains the dominant metric around which global policy turns. Nations are ranked by it. Funding is allocated based on it. Success is defined by it.

The search for alternatives has produced dozens of contenders. The Human Development Index. The Genuine Progress Indicator. The Social Progress Index. The OECD's Better Life Index. Each captures something GDP misses. Each collapses a multidimensional reality—health, education, environment, community, purpose—into a single number. And here is the uncomfortable truth that nobody wants to confront: the method you use to collapse that complexity into one number matters enormously. Two sensible methods applied to the same underlying data can produce radically different rankings of which countries are "doing well."

A new paper published on arXiv in July 2026 confronts this problem directly. Titled "A Comparative Review of Methods to Create a Composite Index for Sustainable and Inclusive Wellbeing," it assembles what may be the most comprehensive examination ever undertaken of how we turn many indicators into one. The authors—Ricardo da Silva Vieira, Mario Biggeri, Robert Costanza, and sixteen co-authors spanning economics, ecology, neuroscience, and data science—catalog thirteen different aggregation methods, evaluate each against nine theoretical conditions that any serious wellbeing index should satisfy, and demonstrate empirically that method choice changes country rankings in ways that cannot be dismissed as noise.

Their conclusion is neither surprising nor comforting: no single method satisfies all nine conditions. But the authors do not leave us there. They sketch a path toward a more honest composite indicator—one that acknowledges its own limitations, that combines approaches rather than claiming false universality, and that moves the conversation from "which country ranks highest" to "what does this particular measurement tell us about what we are missing."

The Science of Measuring What Matters

Why Wellbeing Defies Simple Measurement

The concept of Sustainable and Inclusive Wellbeing—SIW, in the paper's shorthand—sits at the intersection of several difficult intellectual problems. It draws on needs theory, which asks what every human requires to flourish: not just calories but capabilities, not just survival but dignity. It incorporates strong sustainability, which insists that natural capital cannot be fully substituted by manufactured capital—you cannot replace a stable climate with more factories. And it tries to honor both subjective experience (how people report feeling about their lives) and objective conditions (what people actually have access to and can do).

When the United Nations High-Level Expert Group on Beyond GDP began calling for headline indicators that go beyond economic output, they were responding to a genuine hunger for metrics that capture human flourishing in its fullness. But fullness is precisely the problem. The moment you try to compress a complex, multi-dimensional reality into a single number, you face what statisticians call the aggregation problem.

Consider a simplified example. Suppose you want to measure a country's SIW using three indicators: average lifespan, carbon emissions per capita, and reported life satisfaction. A country might excel on lifespan, fail badly on emissions, and land in the middle on life satisfaction. How do you combine these into one score? You could take a simple average—but does burning more carbon really offset a longer life in any straightforward sense? You could weight them by importance—but who decides what importance means, and from whose perspective? You could use penalties for bad performance—but how severe should those penalties be?

Each of these choices embeds philosophical commitments about what wellbeing means and how its components relate to each other. The paper's central insight is that these commitments are unavoidable, and that pretending otherwise—claiming that your weighted average is somehow neutral—is the first mistake.

The Nine Conditions That Matter

To bring order to the chaos of possible aggregation methods, the authors derive nine conditions that any serious attempt to measure SIW should satisfy. These conditions emerge from needs theory and strong sustainability, and they represent what the literature has identified as non-negotiable requirements for an index that takes human and planetary limits seriously.

The first three conditions concern how different dimensions of wellbeing relate to each other. Limited substitutability means that a high score on one dimension cannot fully compensate for a low score on another. Living twice as long does not offset having a destroyed ecosystem. Penalization of imbalances means that an index should penalize countries where some people thrive while others suffer—inequality matters, not just averages. Non-linear transformations means that the relationship between indicators and wellbeing is not straightforward: the jump from terrible to adequate matters more than the jump from good to excellent.

The next two conditions emerge from environmental science. Respect for environmental ceilings means that exceeding planetary boundaries—deforesting beyond regeneration, emitting beyond the atmosphere's capacity to absorb—must drag down an index, regardless of how well a country performs on other dimensions. Respect for lower limits means that there is a floor below which no one should fall: if anyone lacks basic nutrition, shelter, or safety, the index should reflect that catastrophic deprivation, not average it away.

The remaining conditions address technical and ethical requirements. A formative measurement model means that the indicators cause the index, not the other way around—an index should reflect the underlying reality, not define it. No correlation requirement means that dimensions should not be assumed to move together; health and civic engagement might be unrelated, and that independence should be preserved. Distributional sensitivity means that the index cares not just about the average but about spread—about who exactly has what. Cross-border spillovers means that a country's wellbeing depends partly on what happens elsewhere: migration, trade, climate impacts, pandemics. Intertemporal aggregation means that how we weight the past, present, and future matters: do we care as much about the wellbeing of people who will be born in 2075 as we care about people alive today?

These nine conditions are not arbitrary. They emerge from decades of work in welfare economics, ecological economics, and sustainability science. The question the paper asks is: which aggregation methods satisfy which conditions?

The Thirteen Methods

The authors examine thirteen aggregation methods, ranging from familiar to obscure. They categorize them roughly by intellectual lineage.

Simple and weighted arithmetic means treat all dimensions equally or with fixed weights, combining them through straightforward addition. These are the most common approaches in practice—the Human Development Index uses a variant of this. They are easy to understand and calculate, but they assume perfect substitutability (a gain in one dimension fully offsets a loss in another) and are silent on imbalances and ceilings.

Geometric means multiply dimensions rather than adding them, which imposes a kind of penalty for any dimension falling to zero. They are common in environmental indices. They impose some limits on substitutability and are more sensitive to balanced performance, but they still assume dimensions move together and do not explicitly handle environmental floors or ceilings.

Harmonic means are even more punitive toward low values—they penalize extreme imbalance more severely than geometric means. They have found some use in education and development indices.

Penalty-based indices explicitly deduct points for inequality or imbalance, even when average performance is high. The作者的 Index of Sustainable Economic Welfare and its successors use this approach. They address the penalization condition but may not satisfy others.

Outranking multi-criteria methods (including ELECTRE and PROMETHEE families) take a different philosophical approach: rather than computing a single score, they compare pairs of alternatives and determine which one "outranks" the other based on majority rule and veto conditions. These methods impose limited substitutability by construction, but they do not produce a continuous score and can be difficult to communicate.

Data Envelopment Analysis (DEA) borrows from operations research, defining the "frontier" of best-performing units and measuring others by their distance from that frontier. It is powerful for identifying who is most efficient, but it does not naturally produce rankings across all units and can be sensitive to outliers.

Network and graph-based methods treat dimensions as nodes in a network, where relationships between nodes carry information. These approaches can capture correlation structures and emergent properties, but they require more data and more assumptions about network structure.

Multi-level and hierarchical methods recognize that wellbeing operates at different scales—individual, household, community, national—and that aggregation must happen at each level with appropriate rules. This approach is theoretically satisfying but computationally demanding.

Non-compensatory methods refuse to allow any trade-off across certain dimensions. You might measure wellbeing as the minimum across multiple thresholds—a floor that all must meet. These satisfy environmental ceilings and lower limits by construction, but they may produce "cliffs" where small changes in one dimension dramatically shift the index.

Insights from ecology—particularly the concept of ecosystem carrying capacity and resilience—suggest that wellbeing indices should incorporate critical thresholds and tipping points. These approaches inform the emphasis on environmental ceilings in the paper's framework.

Insights from neuroscience suggest that human decision-making under uncertainty is not well-modeled by simple weighted averages. Prospect theory, loss aversion, and reference-point dependence offer more realistic models of how people experience trade-offs.

Machine learning approaches can identify patterns in high-dimensional data that simpler methods miss, but they risk being black boxes—useful for prediction, harder to justify normatively.

This taxonomy is not exhaustive, but it captures the major families of approaches that have been tried or proposed for wellbeing measurement.

What They Found

The Gap Between Theory and Practice

The authors' systematic evaluation reveals a sobering result. They assess each of the thirteen methods against the nine conditions and find that no single method satisfies all nine. The best-performing methods—penalty-based indices and certain non-compensatory approaches—satisfy five or six conditions. The worst-performing—simple arithmetic means—satisfy only one or two.

This is not a surprise, exactly. The nine conditions embody different normative commitments that may themselves be in tension. Perfect substitutability and limited substitutability cannot both hold. Perfect equality weighting and distributional sensitivity push in different directions. The authors acknowledge that trade-offs between conditions may be unavoidable.

But the systematic nature of the evaluation is valuable. It forces explicit comparison rather than letting each method's proponents claim that their approach "of course" handles the important conditions. By laying out a matrix of methods against conditions, the paper makes visible the assumptions that each method embeds.

Rankings Are Not Robust

The paper's most striking empirical finding comes from its illustrative example. Using data on multiple countries (the specific nations are not detailed in the abstract, but the methodology is clear), the authors calculate SIW using different aggregation methods and compare the resulting rankings.

The result is not subtle: aggregation choices change country rankings significantly. The authors note that "compensatory methods create similar rankings"—methods that allow trade-offs across dimensions tend to cluster together. But non-compensatory methods produce substantially different orderings.

Consider what this means in practice. If Method A ranks Country X above Country Y, but Method B ranks Y above X, then the truth about which country has higher SIW depends on which method you trust. If you have no principled basis for choosing one method over another—and the paper suggests such a basis is elusive—then the ranking is not a fact about the world but a product of your methodological choices.

This finding matters because composite indices are used for consequential decisions. Funding flows to higher-ranked countries. Politicians claim credit for climbing in rankings. Researchers draw causal inferences from cross-national comparisons. If the rankings themselves are method-dependent, then all of these downstream uses inherit that instability.

What Each Method Gets Right and Wrong

The paper's evaluation matrix offers a kind of scorecard for each approach. Simple arithmetic means, as expected, fail on substitutability, penalization of imbalances, environmental ceilings, and lower limits. They are theoretically indefensible for a multidimensional wellbeing index, yet they remain common because they are easy to compute and explain.

Geometric means do better on substitutability—they impose a penalty for any dimension falling to zero—but still assume that dimensions move together and do not naturally incorporate distributional sensitivity or cross-border spillovers.

Penalty-based indices explicitly address imbalance and can incorporate environmental ceilings, but they require choosing penalty parameters, and different parameter choices can again change rankings.

Outranking methods impose non-compensatory logic by construction, but they do not produce cardinal scores (only ordinal rankings) and can be computationally complex. They are better suited to identifying dominated alternatives than to creating a headline number.

DEA identifies frontier performers and is distributionally sensitive, but it does not naturally aggregate across multiple dimensions into a single score, and its results depend on the choice of inputs and outputs.

The paper suggests that the most promising approaches combine elements from multiple families. A future SIW composite indicator might use non-linear normalization of raw indicators, non-compensatory aggregation for certain core dimensions, and distributional sensitivity built in throughout. This is not a single method but a design philosophy.

Why This Changes Things

The Illusion of Objectivity

Composite indices carry an aura of scientific authority. They produce a number, often to two or three decimal places. They rank countries with crisp precision. They are cited in policy documents and academic papers with the implicit assumption that they measure something real.

The paper's analysis punctures this aura. The choice of aggregation method is not a technical detail to be left to statisticians. It is a normative decision about what wellbeing means, what trade-offs are acceptable, and whose interests matter. Different methods embody different theories of wellbeing. There is no method-free way to combine health, environment, and happiness into a single score.

This does not mean that all methods are equally good. Some satisfy more of the nine conditions. Some are more transparent about their assumptions. Some have been validated against outcomes we care about. But the choice should be explicit, discussed, and defensible—not buried in a footnote about weighting procedures.

The authors are careful not to claim that their nine conditions are the last word. They derive them from needs theory and strong sustainability, which are themselves contested frameworks. But by making these conditions explicit, they create space for debate. If you think perfect substitutability is acceptable, you should say so and explain why. If you think environmental ceilings are too strict, you should defend a different threshold. The paper forces the conversation into the open.

What GDP Gets Wrong

The motivation for all this methodological sophistication is the inadequacy of GDP. The paper does not belabor this point—it's well-established in the literature—but the stakes are worth restating.

GDP counts everything that is exchanged in markets. It rises when we convert old-growth forest into timber, when we ship sick patients to hospitals, when we remediate pollution we caused, when we increase incarceration. It is blind to unpaid labor, to ecosystem services, to the distribution of gains, to the depletion of non-renewable resources.

GDP's dominance is not just a measurement problem. It shapes incentives. If a country's success is measured by GDP growth, then governments pursue GDP growth, even when that growth comes at the expense of wellbeing. The paper's nine conditions—including environmental ceilings and distributional sensitivity—can be read as a specification of what a measurement system would need to reward if it wanted to guide policy toward genuine wellbeing rather than mere production.

The UN High-Level Expert Group on Beyond GDP has called for headline indicators that go beyond economic output. The paper is explicitly positioned as a contribution to that project. But the authors are warning that the path from "call for alternatives" to "credible headline indicator" requires solving the aggregation problem—not just collecting better data.

The Method Determines the Message

Perhaps the most important insight from the paper is that method choice is not neutral. Different methods send different messages about what matters.

A compensatory method—allowing trade-offs across dimensions—says that a country can "buy" environmental protection by being richer, or "buy" health by being more equal. This may or may not be true, but it embeds an assumption about the relationship between dimensions.

A non-compensatory method—refusing to allow trade-offs on certain dimensions—says that some things matter absolutely. No amount of economic growth justifies destroying the climate. No level of average wellbeing justifies letting some people fall below a floor. This is a stronger claim about rights and limits.

Penalty-based methods say that inequality is bad, even if the average is high. They encode a concern for fairness that pure efficiency metrics ignore.

Network methods say that the relationships between dimensions matter—that health and civic engagement might be connected in ways that simple averages miss.

When a country uses a particular index to guide policy, it is implicitly adopting the philosophy embedded in that index. The paper suggests that policymakers should choose their aggregation methods deliberately, with full awareness of what those methods imply.

Comparisons Are Harder Than We Thought

Cross-national comparisons of wellbeing are seductive but dangerous. We want to know how we are doing relative to other countries. Rankings give us that. But if rankings are method-dependent, then the question "are we doing better than Country X?" may not have a determinate answer.

This does not mean all comparisons are meaningless. It means we should be more humble about what comparisons can tell us. "We rank higher on this index, which uses compensatory geometric means and weights dimensions by expert judgment" is a defensible claim. "We are doing better than Country X" is not, unless you have shown that the comparison is robust to alternative reasonable methods.

The paper suggests that we might need to report not just rankings but uncertainty: how much would rankings change if we used a different aggregation method? A country that is robustly in the top tier across methods is different from one that hovers near the boundary, depending on which method happens to be chosen.

What's Next

A Design for a More Honest Index

The authors do not leave us with a negative result. They sketch a constructive approach: a future SIW composite indicator will need to combine methods across levels. Specifically, it should incorporate:

Non-linear normalization of raw indicators. The relationship between, say, income and wellbeing is not linear—big gains at low income levels matter more than equivalent gains at high levels. Normalization should reflect this. Functions like the log transform or power transforms can impose this non-linearity, but the specific choice of transform matters and should be justified.

Non-compensatory aggregation for certain core dimensions. Environmental ceilings and lower limits should be treated as hard constraints, not dimensions to be traded off. No amount of economic growth offsets exceeding planetary boundaries. This means the index structure should include dimensions where the minimum drives the result, not just weighted averages.

Measurement-level choices for inclusiveness and spillovers. Data should be disaggregated to capture distributional sensitivity—reporting not just averages but variance, not just national figures but regional and demographic breakdowns. And cross-border dependencies should be modeled explicitly, not assumed away.

This is a research agenda, not a finished product. The authors acknowledge that each element—non-linear normalization, non-compensatory aggregation, distributional sensitivity—requires further specification and testing.

The Road to a Beyond-GDP Headline Indicator

The paper is positioned as a step toward the headline aggregated indicator advocated by the UN High-Level Expert Group on Beyond GDP. The High-Level Expert Group has called for indicators that capture sustainability, wellbeing, and resilience in ways that GDP does not. But calling for such an indicator is easier than building one.

The paper's contribution is to map the methodological terrain—to show what choices must be made, what trade-offs are unavoidable, and what the consequences of each choice are likely to be. It is a foundation for more informed deliberation, not a finished blueprint.

The path forward involves several challenges. First, more empirical work is needed to see how robust rankings are to method choices across different datasets and country samples. Second, normative deliberation is needed to decide which of the nine conditions matter most and whether there are conditions the authors missed. Third, communication strategies are needed to present a complex, multi-method index to audiences accustomed to single-number rankings.

Open Questions

The paper leaves several questions unresolved, as any good review should.

Weighting remains contested. Who should decide how much weight to give to health versus environment versus life satisfaction? The paper evaluates methods but does not resolve the ultimate question of whose values should be encoded in the weights. Options include expert judgment, democratic participation, revealed preferences, and capability approaches—but none is obviously superior.

The nine conditions may be in tension. Can an index satisfy both distributional sensitivity and robustness to small changes in data? Can it impose hard environmental ceilings while still differentiating among countries that have met those ceilings? The authors note these potential tensions but do not fully resolve them.

Data availability limits what's possible. The most sophisticated methods require data that many countries cannot provide. Disaggregation by region, demographic group, and time requires statistical capacity that is unevenly distributed. An ideal index may be practically infeasible for lack of data.

Validation is hard. How do you know if an index is measuring what it claims to measure? GDP is criticized for not capturing wellbeing, but wellbeing surveys have their own problems—response bias, cultural differences in reporting, adaptation and habituation. A wellbeing index would need a theory of change connecting its inputs to its outputs, and that theory would need empirical testing.

Political economy matters. Indices that produce inconvenient rankings face political pushback. Countries that rank poorly may contest the data or the method. An index that can be easily manipulated or captured loses its value. The paper focuses on methodological choices but does not address the institutional design needed to keep an index credible over time.

Why This Matters Now

The timing of the paper is not coincidental. The Beyond GDP movement has gained significant momentum, driven by growing recognition that current metrics are driving destructive behavior. Climate change, biodiversity loss, rising inequality, and mental health crises have exposed the limits of GDP as a guide to policy. Governments are looking for alternatives.

But the search for alternatives must be pursued with intellectual honesty. The paper's message is that we cannot simply replace GDP with another single number and call the problem solved. The aggregation problem is fundamental. Any composite index embeds choices about what matters, how dimensions trade off, and whose wellbeing counts. These choices can be made explicit, discussed, and refined—but they cannot be eliminated.

The alternative to a flawed single number is not no number at all. It is a number with a rich explanation of what it captures and what it misses, how it would change under different assumptions, and what it cannot tell you. The most honest headline indicator might be one that presents multiple methods and acknowledges that the "true" ranking may lie within a band of reasonable possibilities.

The nine conditions the paper identifies point toward what such an index would need: respect for limits (both human and planetary), sensitivity to distribution, attention to trade-offs that cannot be compensated, and honesty about uncertainty. These are not just technical requirements. They are ethical commitments about what a decent society owes its members and its descendants.

Building an index that honors those commitments will take years of work. The paper by da Silva Vieira and colleagues is a map of the territory—showing where the paths lead, where they diverge, and where they dead-end. The journey from map to destination remains ahead.