Understanding the determinants of a country’s happiness

Measuring happiness and understanding the dimensions of the national happiness index would help political leaders and policymakers build effective evidence-based development strategies with a sustainable vision. There are various domains related to national happiness, ranging from socio-economic to political factors. Based on data from the World Happiness Reports from 2012 to 2020 and supplementary data from the Worldwide Governance Indicators, World Development Indicators and World Risk Report, this analysis piece seeks to explore the drivers of national happiness.

The annual World Happiness Reports generally analyze the happiness index across countries by looking at domains such as GDP per capita, social support, generosity in making donations, freedom in decision-making and corruption and conducting a pooled OLS regression of happiness index against those domains. However, dimensions such as political stability, unemployment and disaster risks were not included, mainly due to the lack of comparable data for the full sample (World Happiness Report, 2022). I will try adding controls for national political stability, unemployment and disaster risks from other data sources and check to what extent the impact of these dimensions on the happiness index would change when using a balanced panel data analysis.

I mainly used data from the World Happiness Report database, which provides national life satisfaction and relevant dimensions across countries from 2012 to 2020. To further explore the impacts of political stability, unemployment and disaster risks on the national happiness score, I merged the happiness score data with data from the Worldwide Governance Indicators (WGI, 2021), World Development Indicators (World Bank, 2020) and World Risk Report (Bündnis Entwicklung Hilft, 2022).

Happiness indexes by country in the 2012–2020 period

As seen from the animated choropleth map, countries in North Europe, North America, Australia and New Zealand maintained high happiness scores over the 2012–2020 period. Meanwhile, most countries in Africa, South Asia and Central Asia had happiness indexes below the global average score. Particularly, from the map for 2020, it can be inferred that countries missing data in this year were mostly from Africa, followed by nations in the Middle East and Southeast Asia.

Now let’s take a look at the distribution and variation of global happiness indexes across the world over the 2012–2020 period. The red line shows the historical trend of the average global happiness score between 2012 and 2020. Most high-income countries (the red dots) had happiness scores higher than the global median, while most low-income economies (the pink dots) were recorded with happiness scores below the world median. Interestingly, both of the global mean and median of happiness scores in 2020 were higher than the estimates in the previous years, regardless of the severe hit of COVID-19 in 2020.

Exploring the correlation

Happiness and income

There is a positive correlation between happiness and the log-transformed GDP per capita. Particularly, the red dots tend to gather around the upper-right quadrant, indicating that the high-income countries usually have high happiness scores.

Happiness and political stability

Happiness and disaster risks

Regression

I use the below regression model:

Correlation matrix to check for the possibility of multicollinearity:

Regression results:

Models 1 and 2 are generated using the unbalanced panel data set when some countries miss observations in several years; therefore, I used the pooled OLS estimation with year-fixed effects in these models. In models 3 and 4, I ran the regressions using the balanced panel data set, thus I used the panel OLS regression technique with country and time fixed effects for these models. The independent variables in models 1 and 3 were the parameters included in the original data set. Models 2 and 4 were added more controls for political stability, unemployment and disaster risk.

Model 1 – Pooled OLS regression with no control variables

Model 2 — Pooled OLS regression with control variables

Model 3 — Panel OLS regression with no control variables

Model 4 — PanelOLS regression with control variables

Limitations

• Omitted variables bias may overestimate the coefficients of the independent variables in the pooled OLS model.
• The final data set was joined from different sources and each organization could have applied a different sampling method, thus this may cause a concern of inconsistency across the entire data set.
• Small sample size may increase the margin of error and make the coefficients estimated less precise.