Pandemic echoes: The influence of covid-19 on music characteristics on Spotify charts

The COVID-19 pandemic reshaped multiple aspects of day-to-day life, including an increase in people’s reliance on streaming platforms. It raises the question of possible changes in music consumption patterns. This thesis examines whether Spotify audio features shifted in Spring 2020 compared to pre-pandemic years, and whether any observed changes were associated with the timing of COVID-19 restrictions. It also compares Portugal and Finland to assess whether these pat-terns differed across countries. The empirical setting uses Spotify Top 200 daily charts for March–May in 2017–2020 and focuses on three audio features: valence, energy, and danceability. The track-level observations are aggregated into country-day series by computing stream-weighted averages of each audio feature. COVID-19 policy measures are taken from the ECDC dataset with three restriction categories chosen for each country to avoid overlapping start dates. These restrictions’ timing is represented using daily binary indicators, which are then merged with the Spotify series by country and date. The analysis then evaluates two modelling strategies: a lag-based time-series regression and a Prior-Year Weekday Average (PYWA) baseline approach (including a hybrid PYWA + lag specification). Week-day and restriction variables are included in all models. Candidate models are compared using time-series diagnostics (ACF/PACF) and model fit across both countries and all three features, narrowing down to one final model for inter-pretation. In the selected final model (Lag 1 + Lag 365), the one-day lag is large and highly significant for all features in both countries, indicating strong day-to-day persistence, while an annual component (lag 365) is more informative in Finland than in Portugal. Weekday effects are also statistically significant and generally positive relative to Sunday, though with less substantial coefficients. In contrast, the restriction indicators are mostly statistically insignificant and add little explanatory power once lag dependence and weekdays are accounted for. This suggests that shifts in daily audio-feature levels do not correspond to the restriction timing within this modelling framework.