World Happiness Analysis (2015-2019)

Statistical analysis of World Happiness Report data to identify key drivers of national well-being across 158 countries.

This is Phase 1 of a multi-phase project:

Phase 1 (Current): Regression analysis and variable importance modeling - it is the engine of the machine learning model

Phase 2 (Upcoming): Interactive Streamlit dashboard for data exploration

Phase 3 (Vision): Decision support tool for evidence-based policymaking

Long-term goal: Support GovTech and policymakers in making statistically-informed governance decisions that improve citizen well-being.

Overview

This project analyzes five years of World Happiness Report data to understand which factors most strongly predict national happiness scores. Through regression modeling and cross-validation, we identify the relative importance of economic, social, and governance factors.

Key Findings

• GDP per capita is the strongest predictor (standardized β = 0.65)
• All six factors are statistically significant (p < 0.01)
• Social support and healthy life expectancy show substantial effects
• Model achieves test RMSE of 0.42 with strong generalization
• Critical insight: Raw coefficients are misleading - standardization reveals true variable importance

Project Structure

world-happiness-analysis/

├── notebooks/

├── Phase_3&4_Visual_Analysis&_Statistical_Estimation.ipynb

├── Phase_5_Regression_Analysis.ipynb

├── Phase_6_Model_Selection_&Standardized_Variable_Importance_Analysis.ipynb

└── Phase_7_Regularization.ipynb

├── data/

└── happiness_clean_2015_2019.csv

├── World Happiness Analysis.docx

└── README.md

Methodology

Data Cleaning:

• Schema harmonization across 5 years
• Country name standardization
• Outlier filtering (IQR method, k=3.0)

Modeling:

• Multiple linear regression (OLS)
• Train-test split (70/30)
• 5-fold cross-validation
• Heteroskedasticity-robust standard errors (HC3)

Variables:

• Outcome: Happiness score (0-10 scale)
• Predictors: GDP per capita, social support, healthy life expectancy, freedom, generosity, corruption perceptions

Tools & Technologies

• Python 3
• pandas, numpy
• statsmodels, scikit-learn
• matplotlib, seaborn
• Jupyter Notebook

Results

Model 5 selected as optimal balance of accuracy and interpretability:

• Test RMSE: 0.4200
• Adjusted R²: 0.76
• All predictors significant at p < 0.01

Standardized Coefficients (Variable Importance):

1. GDP per capita: 0.65
2. Social support: 0.45
3. Healthy life expectancy: 0.38
4. Freedom: 0.32
5. Corruption (negative): -0.18
6. Generosity: 0.12

Key Insights

Standardization matters: Without standardizing predictors, freedom appeared to be the strongest factor due to its larger raw coefficient. Standardization reveals GDP's true dominance.

Balanced development needed: All six dimensions contribute significantly - no single factor drives happiness alone.

Author

Charles Tsoi
Data Analytics Project
November 2025

License

This project is available for educational and portfolio purposes.