Social support, not income, leads
The blue bar (the share of respondents saying they have someone to count on in tough times) is the single largest contributor in almost every Nordic country. Income comes third.
BEE2041 · Data Science in Economics · April 2026
Beyond GDP.
What really drives national happiness?
The World Happiness Report has crowned Finland the happiest country on earth for seven years running. But why? National wealth is the obvious suspect, yet Afghanistan sits at the bottom of the ranking with under 3% of US income per head. Money cannot be the whole story.
This piece pulls together three live data sources, integrates them in a small SQL database, fits both classical OLS and a modern causal forest, and asks: once we control for everything else, when does income still buy happiness?
7.8Finland's Ladder score
1.9Afghanistan's Ladder score
136countries in the analysis
1,500words, code on GitHub
1 · The headlines
Finland's nearly six-point lead
Finland's Ladder score sits four times above Afghanistan's. The Cantril Ladder question (a 0-10 scale of life satisfaction) is the same instrument worldwide, so the gap is comparable. A six-point gap on a ten-point scale is enormous; any explanation has to grapple with it.
2 · Money matters - with diminishing returns
Below $30k PPP each doubling of income buys roughly the same Ladder bump; above it, the curve flattens
Plotting the Ladder score against GDP per capita on a log axis gives a clear upward relationship. The smoother bends noticeably above around $30,000 PPP. Costa Rica nearly matches the United States on the Ladder (6.6 versus 6.9) on roughly a third of US income, a small-scale version of the Easterlin paradox.
The vertical scatter is also wide: countries with similar income levels routinely differ by close to two Ladder points. That residual variation is where this post lives.
3 · Decomposing the Ladder
What carries the rich-country premium?
The WHR ships six explanatory variables alongside the Ladder. To see how each one contributes to the gap above the world average, I fit a plain OLS regression of the Ladder on the six and then multiply each country's covariate (centred on the sample mean) by the fitted coefficient.
Freedom is universal
Yellow is large and stable across all 15. The highest-ranked country with low freedom is Israel (4th), which makes up the difference through unusually strong social support and life expectancy.
Trust does the heavy lifting
The gap between the world average and Finland's score is mostly not bought with money. The pattern lines up with Robert Putnam's social-capital research and Bo Rothstein's Nordic "high-trust equilibrium" account.
4 · OLS, six ways
Adding social support roughly halves the log-GDP coefficient
I fit six progressively richer specifications on the 2023 cross-section, all with HC3 robust standard errors. The coefficients on log GDP and the social-capital variables tell most of the story.
| Term | m1 | m2 | m3 | m4 | m5 | m6 |
|---|---|---|---|---|---|---|
| R² | ||||||
| N |
***p < 0.01 **p < 0.05 *p < 0.10
All six specifications use HC3 robust standard errors. (m6) uses z-scored variables for direct comparability. (m5) swaps WHR's bundled log-GDP for the World Bank's PPP series. N = 134 because two countries are missing the corruption variable.
The log-GDP coefficient drops from 0.80 on its own to 0.27 once social support, freedom and corruption are controlled for. Most of the apparent income premium runs through those other channels.
This lines up with a long literature on the economics of well-being. Easterlin (1974) first noted that within rich countries the income-happiness link flattens once basic needs are met. Later work (Stevenson & Wolfers, 2008; Layard, Mayraz & Nickell, 2009) finds the gradient does survive in cross-country data but shrinks a lot once mediating institutions are added. Specification (6) makes the magnitudes comparable: a one-SD rise in social support is worth roughly as much as a one-SD rise in log GDP, and freedom is not far behind.
Specification (5) is a quick cross-source check: swapping WHR's bundled log GDP for the World Bank's PPP series leaves the coefficient almost unchanged (0.27 vs 0.26), so the headline is not driven by which income series we use.
5 · A causal forest
When does income still buy happiness?
Cross-country comparisons are not a randomised trial: rich and poor
countries differ in many ways at once. To explore heterogeneity I fit a
causal forest (Wager & Athey, 2018) with
econml.dml.CausalForestDML. The "treatment" is being above
the median income, and X covers six potential moderators. The forest
runs on n = 130: six of the 136 countries fall out because internet
or urban-share data are missing.
This is descriptive, not causal. Cross-country data does not give us unconfoundedness: countries are not randomly assigned to be rich, and X is not an exhaustive conditioning set. So I read the CATEs as heterogeneity in conditional means, not as what would happen to Ladder if a poor country suddenly became rich.
The average treatment effect is statistically indistinguishable from zero once the moderators are controlled for: the 95 % CI spans both signs. Most of the apparent income premium runs through the mediators, and the conditional effects vary across countries.
The forest's split-importance ranks internet penetration first among the moderators, ahead of social support, corruption, healthy life expectancy, urbanisation and freedom. Once we know whether a country's residents are online, the leftover role of GDP per capita shrinks sharply, lining up with recent work on digital connectivity and the WHR 2026 edition's own pivot to social-media research.
| Moderator | Importance |
|---|---|
| Internet (%) | 0.314 |
| Social support | 0.165 |
| Corruption | 0.153 |
| Healthy life exp. | 0.138 |
| Urban (%) | 0.129 |
| Freedom | 0.100 |
Split-importance from cf.feature_importances_. Sums to 1.0; full table in output/tables/cf_importance.csv.
Robustness check. The "above median" cut is convenient but arbitrary. Refitting at the 75th percentile (around $44k PPP) sharpens the result: the ATE is about +0.28 Ladder points with a 95% CI clear of zero. The income effect concentrates in the richest quartile rather than at the median.
6 · A note on the scrape
The Report has been growing
Alongside the rankings spreadsheets, I scraped chapter metadata for
every WHR edition from 2020 to 2026 (titles, authors, affiliations,
reading times, DOIs). This isn't visualised on the page but it
sits in data/raw/whr_chapters.csv and feeds the
whr_chapters table in the SQLite db.
- 51 chapters scraped
- 7 editions, 2020-2026
- 27 -> 35 mean reading time (min)
- +72% longer reads vs 2020
The 2026 edition leans heavily into social-media and digital
connectivity, which is what motivated putting internet_pct
and urban_pct into the causal forest in the first place.
7 · Take-aways
Four things I take from this
- Money matters, but mostly indirectly. Once you control for social support, freedom, life expectancy and trust, the marginal contribution of income to subjective well-being is small. This is the modern empirical reading of the Easterlin paradox: GDP shows up in well-being not because cash is happiness, but because richer countries tend to have the institutions that produce it.
- The link is non-linear and concentrated at the top. Splitting at the median gives an ATE indistinguishable from zero, but moving the cut to the 75th percentile (~$44k PPP) lifts it to about +0.28 Ladder points with a 95 % CI clear of zero. So the income effect has not vanished, it has migrated into the richest tail of the distribution.
- Social capital is at least as potent as material wealth. In every Nordic country the biggest single contributor to the Ladder gap above the world mean is social support, not GDP.
- Internet penetration is the dominant moderator. Split-importance ranks it ahead of social support, corruption, life expectancy, urbanisation and freedom.
Replication
One command rebuilds everything
Every figure on this page is generated by a Python pipeline tracked under git. Clone, install, run:
git clone https://github.com/maksimkitikov/Data-Science-in-Economics.git
cd Data-Science-in-Economics
pip install -r requirements.txt
make all
The Makefile walks the eight numbered scripts in scripts/,
regenerates the SQLite database, the OLS table, the figures and the
JSON consumed by this page.
Sources & further reading
- World Happiness Report 2020-2026, Sustainable Development Solutions Network: worldhappiness.report
- World Bank, World Development Indicators, queried via
wbgapi - Wager, S. & Athey, S. (2018). Estimation and Inference of Heterogeneous Treatment Effects. JASA 113(523).
- Athey, S., Tibshirani, J. & Wager, S. (2019). Generalized Random Forests. Annals of Statistics 47(2).
- Davis, J. & Heller, S. (2017). Using Causal Forests to Predict Treatment Heterogeneity. AEA P&P.