World Happines Report (WHR) 2017

• Data source: World Happiness Report, using only the year 2016.
• Goal: understand the drivers behind happiness, not just the final score.
• Reference points: two hypothetical countries, "utopia" and "dystopia".
• Interpretation: utopia takes the best observed value for every feature, while dystopia takes the worst.

Many countries have missing values in 2016, so the project required a dedicated imputation strategy.

Why imputation was needed

• Several countries are missing one or more features in 2016.
• First pass: use previous years and average historical records when they exist.
• Limitation: countries without historical records cannot be handled in that first step.
• Temporary decision: leave those countries aside until a second-stage model is available.

How the integrated score was built

• Standardize the usable data so each feature lies between 0 and 1.
• Train an Autoencoder (AE) and use the encoder to project countries into a two-dimensional non-linear latent space.
• Use Principal Component Analysis (PCA) in that latent space to define the main one-dimensional ordering.
• Project utopia and dystopia onto that ordering after training.
• Define an integrated score from 0 to 10, where dystopia = 0 and utopia = 10.

How the missing countries were recovered

• Train a Random Forest (RF) on countries that already have an integrated score.
• Use that RF to estimate the score of countries with missing values and no historical records.
• Feed the final scores through the decoder part of the AE.
• Use the decoder reconstruction as the final imputation for the missing 2016 values.

This approach is more informative than simply averaging historical values, because the AE can capture more complex non-linear relationships among countries. An imputed version of the data was created as part of this project.

Workflow summary

1. Train an AE and use its encoder.
2. Use PCA in the latent space to assign an integrated score to each country.
3. Use a RF to estimate the integrated score of countries with missing values.
4. Use the AE decoder to reconstruct the missing features.

Main takeaways

• Global inequality is enormous.
• Qatar reaches a GDP per capita of roughly 136,000 USD, while much of the world remains below 20,000 USD, and Burundi does not reach 400 USD.
• In parts of America and Africa, household income inequality is especially severe.
• Some countries combine low freedom with high sadness and anger.
• Large areas of the map turn red when we look at perceived corruption or generosity.
• There are also more hopeful signals.
• Countries such as Rwanda or Somalia stand out for their efforts against corruption.
• Much of the Americas appears stronger in happiness and enjoyment.
• Across much of the world, people still report that they have someone to count on in times of trouble.