Abstract
We propose a new measure of national valence based on the emotional content of a country's most popular songs. We first trained a machine learning model using 191 different audio features embedded within music and use this model to construct a long-run valence index for the UK. This index correlates strongly and significantly with survey-based life satisfaction and outperforms an equivalent text-based measure. Our methods have the potential to be applied widely and to provide a solution to the severe lack of historical time-series data on psychological well-being.
Introduction
Measuring a nation's collective well-being is a cornerstone of modern policy, yet researchers face a significant "data gap." Systematic surveys of life satisfaction, such as the World Happiness Report, only began relatively recently. To look further back into history, researchers have previously turned to "text valence"—analyzing the sentiment of words in millions of digitized books. However, text-based measures have flaws: language evolves, publication lags create delays, and many regions lack massive digitized literary corpora. This paper explores whether music—often called the "universal language of emotions"—can provide a more immediate and robust way to track national mood.
Methodology
The researchers developed a Music Valence Index (MVI) by combining music information retrieval (MIR) with econometric analysis:
- Feature Extraction: Instead of analyzing lyrics, the team focused entirely on audio, extracting 191 different audio features covering harmony, timbre, rhythm, and melody.
- Machine Learning: They trained a Support Vector Regression (SVR) model using the emoMusic dataset, which contains songs previously annotated by humans for emotional valence (positivity vs. negativity).
- The "Horse Race": To test the MVI, they identified the most popular UK songs from 1973 to 2010 and compared the index against two benchmarks: Eurobarometer surveys (self-reported life satisfaction) and the Text Valence Index (TVI) derived from Google Ngrams.
Key Results
The study reveals that the audio features of chart-topping hits are a remarkably accurate proxy for national life satisfaction. Interestingly, the single most popular song of the year (the #1 hit) was found to be a better predictor of national mood than an average of the Top 10, suggesting it captures a unique cultural zeitgeist.
As shown in the time-series data, the MVI successfully captures dramatic shifts and local peaks in national mood—such as those in 1980 and 1989—which the smoother, text-based TVI often misses. Unlike literature, which is a "slow" medium, music is ephemeral and reactive, showing a much tighter temporal link to life satisfaction.
In a direct "horse race" comparison, the MVI emerged as a stronger and more significant predictor of national happiness than the TVI ($p = 0.007$ when controlling for economic factors like GDP, inflation, and unemployment). The MVI correlates significantly with survey-based life satisfaction ($r = 0.38, p = 0.02$), proving its utility as a high-frequency well-being metric.
Discussion and Limitations
By ignoring lyrics and focusing on raw audio, the MVI avoids the "language evolution" trap and offers a potentially universal tool for tracking well-being across different cultures. However, the authors note several limitations. The study relies on historical record sales, which skewed toward younger demographics, and the rise of streaming may change how popularity reflects genuine national mood. Furthermore, while the MVI reflects the public mood, the researchers remain agnostic on whether music simply reflects or actively influences national sentiment.
Summary updated