Quantifying the drivers of protective behaviours in epidemics: Mask wearing during COVID-19

Epidemic dynamics are shaped by behavioural responses to the risk of contagion, which are in turn influenced by beliefs, disease awareness, compliance with government mandates, and other drivers. However, quantitative assessments of the importance of drivers of protective behaviours remain scarce, largely due to limited and fragmented datasets. This limitation hinders the development of empirically-grounded models that couple the dynamics of human behaviour and of epidemic spread, which would have the potential to substantially improve the prediction of disease transmission and the effectiveness of pharmaceutical and non-pharmaceutical interventions. Existing behavioural-epidemiological models typically rely on strong assumptions regarding the linear and direct dependence of behavioural responses on awareness, opinion dynamics, epidemic progression, or other factors that can only be partially validated against small-scale datasets. Here, we address this challenge by integrating multiple datasets to quantify the impact of different drivers of protective behaviours, focusing on mask-wearing during the COVID-19 pandemic. Combining multivariate regression, Granger causality and random forests for covariate analysis reveals that beliefs regarding the effectiveness of protective behaviours and compliance with government interventions explain most of the variability of mask-wearing behaviour worldwide and are substantially more predictive than peer influence or disease awareness. Our findings provide quantitative evidence on the primary drivers of mask wearing, establish a reproducible framework to identify the drivers of other protective behaviours, and empirically test key assumptions underlying widely adopted behavioural-epidemiological models, thereby supporting their refinement and interpretability.