#SubjectiveWellBeing #MachineLearning #PolicyResearch #OECD #InterpretableML #WellbeingEconomics #SocialScience

We’d love your thoughts:
➡️ Do non-linear social effects surprise you?
➡️ What is the most interesting data and method to test this causally on the individual scale?
Let’s discuss 👇

Read the full paper 📄: link.springer.com/article/10.1...
With co-authors Marco Quatrosi and Angelika van Dulong.

This study provides:
✅ A multidimensional view (economic, social, environmental)
✅ Regional insight for OECD policymakers
⚠️ But generalizability is limited beyond OECD countries.

We confirm many known patterns (e.g., the role of trust, perceived corruption).
But we also highlight neglected variables, calling for new causal studies and regional policy reflection.

Some interactions surprised us:
⬇️ Low employment and low elderly sex ratio → higher SWB
⬆️ But this reverses at higher levels.
Non-linearities like this challenge conventional wisdom.

Alongside expected factors like income & social support, we find a _novel predictor_:
👉 Sex ratio among the elderly
This rivaled income in predictive power.

We apply this approach in 4 steps:
1️⃣ Expand OECD’s well-being dataset
2️⃣ Use random forests to predict SWB
3️⃣ Study our model via interpretable ML methods
4️⃣ Derive new hypotheses for future research

💡 Why ML over traditional econometrics?
→ Captures _non-linearities_
→ Includes _interactions_
→ Handles _many predictors_
→ Supports _exploratory, hypothesis-generating_ research
(See Mullainathan & Spiess, 2017)

We propose a machine-learning-informed workflow that generates testable hypotheses for SWB research.
This is induction—powered by ML—for complex socio-economic systems.

Why this matters:
SWB is now key for measuring progress — beyond GDP.
But most models use few variables and miss complex dynamics.
We offer a new workflow to present a way forward.