Economic composition predicts national wellbeing more strongly than aggregate growth

The relationship between economic growth and human wellbeing is weaker than is often assumed. Although GDP remains the dominant metric of national performance, cross-country and longitudinal evidence suggests that beyond moderate income levels, additional output is associated with progressively smaller gains in life satisfaction.^1–3 First identified by Easterlin⁶ and recently re-examined using 50 years of data from 166 countries,³ this pattern implies that early income growth secures basic capabilities, whereas later increments increasingly support other forms of economic activity. The key question is therefore not only how much economies grow, but what kinds of activity that growth increasingly comprises, and whether those activities are associated with wellbeing.

We propose that part of the answer lies in economic composition rather than aggregate volume. For analytical purposes, we distinguish between direct capability-enhancing activity and system overhead. By system overhead, we mean the administrative, coordinative, protective and corrective layers required to sustain complex societies. Such overhead is not uniformly wasteful: some forms may build capability and resilience, whereas others may primarily repair preventable harms or absorb value without proportionate social return. We therefore hypothesize that the composition of overhead, rather than its total magnitude, will be differentially associated with national wellbeing.

To examine this possibility, we decompose measured overhead into three analytically distinct categories and test whether their composition, summarized by an Institutional Ratio, is associated with wellbeing outcomes across countries.

Composition predicts wellbeing

We estimated three components of system overhead for 30 countries (see Methods), each expressed as a percentage of GDP: Institutional Infrastructure (I), including selected measures of social provision and healthcare administration; Remediation Load (R), including selected costs associated with preventable disease, climate-related disaster and pollution; and Extractive Dissipation (E), including selected measures of excess military expenditure, finance, advertising, healthcare pricing and monopoly rents. We summarize composition using the Institutional Ratio, defined as:

expressed in percentage points.

The Institutional Ratio is strongly associated with life satisfaction (r = +0.80, P < 0.001; Fig. 1) and social trust (r = +0.90, P < 0.001). By contrast, total measured overhead shows no significant association with life satisfaction (r = −0.18, P = 0.35) and no significant association with the other outcomes examined (Table 1). The central empirical pattern is therefore compositional: the internal allocation of overhead is strongly associated with wellbeing, whereas the aggregate scale of overhead is not.

Regression confirms composition effect

In a multiple regression predicting life satisfaction from the Institutional Ratio and log GDP per capita, both predictors are significant (Table 2). A 10-percentage-point increase in the Institutional Ratio is associated with a 0.27-point increase in life satisfaction on the 0–10 scale, conditional on income. The model explains 74% of the cross-country variance in life satisfaction. This result should be interpreted as a conditional association rather than a causal effect.

Wellbeing Efficiency varies

Given diminishing returns to income, a related question is how efficiently countries translate income into reported wellbeing. We define a descriptive measure of Wellbeing Efficiency as:

Because the logarithmic transformation maps multiplicative income increases onto an additive scale, this measure can be interpreted as life satisfaction per income doubling (Fig. 2). On this metric, Finland scores 1.64 and the United States 1.37, a difference of 19%. The Institutional Ratio is positively associated with Wellbeing Efficiency (r = +0.68, P < 0.001). As shown in robustness checks, this pattern is also directionally consistent using residual-based efficiency measures.

Why composition matters

One possible objection is that any measure emphasizing institutions may simply restate the proposition that better-functioning societies exhibit higher trust and wellbeing. Our claim is narrower. We do not find that more overhead, in aggregate, is associated with better outcomes; rather, we find that different forms of measured overhead show markedly different associations. This distinction is important because it indicates that composition carries more explanatory power than total volume alone. Illustratively, the United States devotes a larger share of GDP to measured overhead than Denmark (~18% versus ~13.5%), yet Denmark reports higher wellbeing. The contrast lies less in total overhead than in composition: Denmark allocates 63% of measured overhead to infrastructure compared with 25% in the United States.

Extraction and inequality

Extractive Dissipation is moderately negatively associated with life satisfaction (r = −0.51, P < 0.01) and strongly positively associated with inequality (r = +0.69, P < 0.001; Fig. 3). These patterns are consistent with the interpretation that more extractive forms of economic activity operate, at least in part, through transfer rather than broad-based capability creation.

Cross-country patterns

Table 3 presents illustrative composition and outcome values for selected countries. In this sample, Nordic countries allocate 59–63% of measured overhead to infrastructure, compared with 25% in the United States.

Robustness

The main results remain stable across alternative specifications (Extended Data Table 1). The association between the Institutional Ratio and life satisfaction remains strong when excluding Nordic countries (r = +0.73), using Spearman rank correlation (r_s = +0.85), restricting the sample to OECD countries only (r = +0.86, n = 23), replacing the ratio with a log-ratio specification log(I/(E+R)) (r = +0.80), and removing outcome-adjacent terms from component construction (r = +0.77). A residual-based version of Wellbeing Efficiency yields directionally consistent results (r = +0.38, P = 0.04). The non-OECD subsample is directionally consistent but statistically underpowered (r = +0.71, P = 0.07, n = 7).

Discussion

These findings suggest a compositional interpretation of the Easterlin paradox. If economic expansion increasingly takes the form of remediation and extraction rather than capability-building infrastructure, then additional GDP need not translate into higher wellbeing. On this reading, the puzzle is not simply that growth ceases to matter, but that the composition of growth may matter more than its aggregate scale.

The framework can be understood as distinguishing among activities that build capability, activities that repair preventable harms, and activities that primarily transfer or absorb value without proportionate capability gains. Institutional infrastructure may generate compounding social returns by strengthening capability and coordination. Remediation addresses harms that are often necessary to treat but do not necessarily expand underlying capability. Extractive activity may redirect value without corresponding broad-based gains in wellbeing.

We emphasize that the present cross-sectional design does not permit causal inference. Reverse causation is plausible: societies with higher trust and wellbeing may be better able to sustain effective institutions and lower remediation burdens. Omitted variables, including culture, history, political structure and state capacity, may also contribute to the observed associations. We therefore do not claim that the Institutional Ratio causes wellbeing; rather, we show that measured composition is more strongly associated with wellbeing outcomes than income or total overhead in this sample. Establishing causal direction remains a task for future work.

Several limitations should be noted. First, the component estimates rely on proxy measures and threshold assumptions, and although the results are robust to reasonable parameter variation (Extended Data Table 3), measurement uncertainty remains. Second, the sample is small and OECD-weighted (23 of 30 countries), limiting external validity. Third, some within-category correlations are substantial (VIF 1.7–6.9), raising the possibility that the framework partly captures broader institutional clustering rather than sharply separable mechanisms. Although single-proxy and alternative-specification tests support the overall pattern, the categories should be interpreted as analytically useful approximations rather than definitive ontological partitions.

These findings suggest that, under ecological and fiscal constraints, improving wellbeing may depend less on aggregate expansion than on redirecting economic activity toward capability-building institutional capacity.

Methods

Sample

We analysed 30 countries selected ex ante on the basis of data availability and broad regional diversity: 23 OECD members together with Brazil, Costa Rica, India, South Africa, Chile, Mexico and Turkey. The sample is not globally representative, and external validity should therefore be interpreted cautiously, particularly for underrepresented regions.

Wellbeing Efficiency

We define Wellbeing Efficiency as WE = H / log₁₀(GDP_pc), where H is life satisfaction (Cantril ladder, 0–10) from the World Happiness Report 2024⁴ and GDP_pc is PPP-adjusted GDP per capita in 2023 US dollars from the World Bank.⁵ This is intended as a descriptive scaling measure rather than a structural welfare parameter.

The logarithmic transformation is used for four reasons. First, prior literature indicates that the relationship between income and wellbeing is concave, such that earlier gains matter more than later ones. Second, the log scale transforms multiplicative income increases into an additive scale, making the measure interpretable as wellbeing per income doubling. Third, cross-country rankings are invariant to log base choice. Fourth, the substantive pattern is directionally consistent when efficiency is estimated using residual-based approaches.

For Finland, for example, H = 7.74 and GDP_pc = 53,000, so log₁₀(53,000) = 4.724 and WE = 7.74 / 4.724 = 1.64.

Component operationalization

We operationalize three overhead components, each expressed as a percentage of GDP. No outcome variable used in the analysis, including life satisfaction or inequality, is directly included in predictor construction. However, because the proxies capture related institutional and social structures, conceptual overlap cannot be ruled out entirely.

Extractive Dissipation (E). We define this component as the sum of selected proxies intended to capture activity that absorbs or transfers value without proportionate capability gains:

E_Military = max(Military − 1.5, 0), where 1.5% of GDP approximates the OECD median benchmark (SIPRI).

E_Finance = max(Finance − 5.0, 0), where 5.0% of GDP approximates the OECD median benchmark (BIS).

E_Advertising = 0.7 × Advertising, reflecting the assumption that a substantial share of advertising expenditure is persuasive rather than informational.

E_Healthcare = max(Health − OECD median, 0) × AdminExcess, intended to capture excess pricing and administrative burden above benchmark levels.

E_Monopoly is proxied using markup rents.⁷

These choices are necessarily approximate and are tested for sensitivity in Extended Data Table 3.

Institutional Infrastructure (I). We define this component as the sum of selected proxies intended to capture capability-building and coordinating social provision:

I_Social = min(SocialSpending, 15), using OECD SOCX social expenditure excluding health, capped to reduce domination by extreme values.

I_HealthAdmin is defined as 0.5 × AdminCosts for countries with UHC service coverage ≥ 80, and 0.2 × AdminCosts otherwise, reflecting the assumption that healthcare administration in higher-coverage systems is more likely to serve integrative and enabling functions rather than fragmented billing or access frictions.⁸

Remediation Load (R). We define this component as the sum of selected proxies intended to capture expenditure associated with repairing or managing preventable harms:

R_NCD = 0.4 × NCDSpending, reflecting an assumed preventable share of non-communicable disease burden based on WHO-informed estimates.

R_Climate is proxied using disaster costs (Munich Re).

R_Pollution is proxied using environmental protection expenditure (OECD).

R_Mental = 0.2 × MentalHealth, using a fixed proportion intended to capture the share plausibly attributable to preventable social and environmental stressors.

As with the other components, these operational choices are approximate and intended for comparative rather than definitive measurement.

Institutional Ratio. We summarize composition using:

interpreted as the percentage of measured overhead allocated to Institutional Infrastructure. For Denmark, for example, E = 3.2, I = 8.5 and R = 1.8, giving IR = 100 × 8.5 / 13.5 = 63 percentage points.

Parameter sensitivity

Extended Data Table 3 shows that varying all major thresholds and coefficients by ±25% leaves the main correlation substantively unchanged (r = 0.78–0.82), suggesting that the headline result is not driven by any single parameter choice.

Statistics

We report Pearson and Spearman correlations with 95% bootstrap confidence intervals based on 10,000 country resamples. Partial correlations control for log GDP per capita. We estimate ordinary least squares models for conditional association, apply Bonferroni and Benjamini–Hochberg false discovery rate corrections, and assess robustness using an alternative log-ratio specification, log(I/(E+R)). Analyses were conducted in R 4.3.2.

Data availability

Open sources include the World Happiness Report, World Bank, OECD, SIPRI and WHO GBD datasets. Restricted data include GroupM (available by request) and Munich Re (derived values reported in Supplementary Table 1). Supplementary Table 1 provides all country-level values used in the analysis.

Code availability

R code is provided in the Supplementary Information.

Acknowledgements

We thank colleagues at the AI+Wellbeing Institute for comments on earlier drafts.

Author contributions

J.R. conceived the study, developed the framework and wrote the manuscript. C.D.B. contributed to data analysis, robustness checks and revision.

Competing interests

The authors declare no competing interests.

Additional information

Supplementary Information is available for this paper. Correspondence and requests for materials should be addressed to J.R. Reprints and permissions information is available at www.nature.com/reprints.

Extended Data