Data Science Project · Group 21

The Wealth Paradox

A data-driven investigation into global suicide rates and economic well-being across 180+ countries, 1990–2022.

180+ countries 32 years of data WHO · World Bank · UNDP Python · statsmodels · scipy
This project analyses suicide rates as a public health metric using anonymised, country-level aggregate data from the WHO. If you or someone you know is struggling, please reach out to a local mental health helpline.

Is economic prosperity synonymous with mental well-being?

This common assumption has been floating around for a while — but how true is it? In an increasingly interconnected world, economic downturns and booms are felt by everyone. Our project uses over three decades of global data to explore the complex relationship between economic conditions and mental health outcomes.

180+Countries
32Years (1990–2022)
3Core Datasets
2Research Questions

We combined the WHO Global Suicide Rates dataset (via Kaggle) with World Bank economic indicators (GDP per capita, employment-to-population ratio, inflation) and the UNDP Human Development Index — capturing health, education, and income dimensions of human development globally.

We also engineered derived variables: SuicideDiff (year-over-year change in suicide rate), GDPperCapitaGrowthPct, and EmpRatioChange to capture dynamic trends rather than static snapshots.

Research Question 1 How do suicide rates respond to macroeconomic crises, such as periods of high inflation or sudden GDP contractions?
Research Question 2 In times of rising unemployment, do countries with stronger welfare systems (reflected in HDI) show smaller increases in suicide rates?

What the data looks like

Before diving into the research questions, we explored the dataset extensively — cleaning missing values, understanding distributions, and mapping regional patterns.

GDP vs Suicide Rate over time
Global average suicide rate (red, left axis) vs mean GDP per capita (blue, right axis), 1990–2022. Note the divergence after 2000 — GDP rose sharply while suicide rates fell, then both converged near 2020.

The headline chart tells a nuanced story. Rising GDP did correlate with falling suicide rates globally — but not uniformly, and not without exceptions. This motivated our deeper regional and crisis-level analysis.

Average suicide rate by region bar chart
Europe has the highest average suicide rate of any region — a counterintuitive finding that motivated the "European V-shape" analysis.
Distribution of suicide rates
The distribution of country-year suicide rates is strongly right-skewed, with most values under 20 per 100k but a long tail of extreme outliers.
Suicide rates by region over time
Suicide rates by region over time. Europe starts highest and declines dramatically. Africa starts lowest but shows volatility near 2020.
Top 10 countries by average suicide rate
Lithuania, Belarus, and Russia consistently rank as the countries with the highest average suicide rates over the full 32-year period.

Correlation analysis

We ran correlation heatmaps to understand how economic indicators relate to suicide metrics — first globally, then zooming into Europe where the patterns are strongest.

Global correlation heatmap
Global correlation heatmap. GDPPerCapita shows almost no linear correlation with DeathRatePer100K (r = 0.01) — a first hint that the relationship is non-linear.
Europe correlation heatmap
Europe-only heatmap. Within Europe, higher GDP is more strongly associated with lower death rates (r = -0.34), suggesting wealth does matter — but only within a certain developmental band.
GDP vs suicide rate Europe V-shape scatter
The "European V-shape": plotting GDP (log scale) against suicide rate for Europe reveals that middle-income countries have the highest rates. Very poor and very wealthy countries both have lower rates — for different reasons. This non-linearity explains why global correlations appear flat.
EDA

The European V-shape

GDP and suicide rate have a non-linear relationship in Europe: middle-income countries show the highest rates. This is likely because rapid industrialisation and social dislocation create stress that very poor countries (with tight community bonds) and very wealthy countries (with robust mental health infrastructure) don't experience in the same way. This finding directly informed our regression modelling choices.

How do crises affect suicide rates?

Our first hypothesis was intuitive: if a crisis hit, suicide rates would spike. To test this, we defined crisis years using two criteria:

📉

GDP Crisis

GDP per capita dropped ≥ 2%, growth went negative, or it was a known global recession year (1998, 2008, 2020).

🔥

Inflation Crisis

Inflation exceeded 10%, or spiked by ≥ 5% from the previous year.

# Flag crisis years cond_gdp_drop = df['GDPperCapitaGrowthPct'] <= -2 cond_recession = df['Year'].isin([1998, 2008, 2020]) cond_inflation = df['InflationRate'] > 10 df['Any_Crisis'] = cond_gdp_drop | cond_recession | cond_inflation

The stagnation effect

We ran a Welch's t-test comparing year-over-year suicide rate changes in crisis vs non-crisis years. The result was statistically significant (t = 2.49, p = 0.013) — but the direction was surprising.

Crisis vs non-crisis boxplot
Boxplot of year-over-year change in suicide rate. Crisis years (left) show a distribution centred near zero; non-crisis years (right) show a clear downward shift. The global decline in suicide rates effectively pauses during crises.

"Economic crises don't spike suicide rates — they stagnate them. The global downward trend in suicide rates effectively stops during crisis years."

PeriodMean SuicideDiffInterpretation
Non-Crisis Years−0.248Steady global improvement
Crisis Years−0.014Progress nearly stops

The lagged effect

If the immediate effect was stagnation, when did the real damage appear? We ran an OLS regression predicting current suicide rates using GDP growth and inflation from the current year, 1 year ago, and 2 years ago.

Event study average suicide rate before and after crisis
Event study: average suicide rate in the years surrounding crisis events (0 = crisis year). Rates dip at the crisis year itself, then climb back up in the 1–2 years that follow — the psychological toll arrives with a delay.
VariableCoefficientp-valueSignificance
GDP Growth (current year)+0.0390.020Significant
GDP Growth (lag 1)+0.0210.236Not significant
GDP Growth (lag 2)+0.0190.289Not significant
Inflation Rate (current)+0.0050.389Not significant
Inflation Rate (lag 1)−0.0050.367Not significant
Inflation Rate (lag 2)+0.0100.003Significant ★

Key Finding: The 2-Year Delayed Effect

High inflation two years ago is a statistically significant predictor of rising suicide rates today (p = 0.003). People are resilient during the initial shock — but as savings deplete, debts mount, and hope fades, the psychological toll eventually manifests 1–2 years later.

Model fit: R² = 0.011 — expected, since suicide is driven by culture, healthcare access, and stigma, not just economics.

Policy implication: Policymakers cannot declare victory just because suicide rates don't spike in the first year of a recession. Mental health support systems must be maintained well after markets stabilise.

Do stronger welfare systems buffer employment shocks?

We examined whether countries with higher HDI scores showed smaller suicide rate increases during employment shocks — defined as a drop of ≥ 1 percentage point in the employment-to-population ratio. Countries were split at the median HDI into High and Low groups.

HDI group employment shock boxplot
Change in suicide rate during employment shock years, split by HDI group. Low HDI countries (left) show a positive mean change; High HDI countries (right) show a slight decline — suggesting welfare systems do buffer the shock.
GroupMean SuicideDiffMedian SuicideDiffCount
High HDI−0.139+0.065137
Low HDI+0.337+0.098153
Correlation heatmap with HDI
Correlation heatmap including HDI. HDI correlates strongly with GDPPerCapita (r = 0.78) and moderately with DeathRatePer100K (r = 0.21), confirming its relevance as a welfare proxy.
2

Directionally Compelling, Not Conclusive

High-HDI countries averaged a slight decline in suicide rates (−0.14) during employment shocks, while low-HDI countries averaged an increase (+0.34). The OLS interaction model confirmed the direction — but the interaction term did not reach significance (p = 0.137).

This result should be treated as suggestive, not conclusive. The pattern is consistent with the theory that welfare systems buffer unemployment shocks — but the data doesn't confirm it definitively.

import statsmodels.formula.api as smf reg = smf.ols( "SuicideDiff ~ EmpShock * C(HDI_Group)", data=analysis_df ).fit() print(reg.summary())

How we built this analysis

🗂

Data Sources

WHO suicide data (Kaggle, 1990–2022), World Bank economic indicators, UNDP HDI dataset, Gini coefficient data.

🔧

Feature Engineering

Constructed SuicideDiff, GDPperCapitaGrowthPct, EmpRatioChange, lagged inflation/GDP variables, and binary crisis flags.

📊

Statistical Tests

Welch's t-test for group comparisons. OLS regression with lagged predictors and interaction terms via statsmodels.

🧹

Data Cleaning

Forward-fill within country, dropped countries with entire columns missing. Final dataset: ~4,700 country-year observations.

Tech stack

Python 3 pandas NumPy statsmodels scipy matplotlib seaborn Jupyter Notebook

What this analysis cannot tell us


Group 21

This project was a collaboration between five students as part of a data science course.

H
Hafsah Nasir LinkedIn ↗
H
Hamania Asim LinkedIn ↗
E
Elizeh Faisal Group 21
Z
Zainab Raza Group 21
A
Aamna Imran Rana Group 21