Abstract
<sec> <title>BACKGROUND</title> <p>Behavior is an endogenous component of epidemic dynamics: populations may change their perceptions, trust, contacts, masking, or isolation before or after conventional surveillance indicators change. Yet COVID-19 research has often examined individual behaviours, single countries, or contemporaneous associations, leaving unclear whether multidimensional behavioural signals lead or lag epidemiological and policy indicators across national settings.</p> </sec> <sec> <title>OBJECTIVE</title> <p>To characterize the timing, direction, and cross-country heterogeneity of associations between population-level behavioural responses and COVID-19 epidemiological and policy indicators, and to assess whether countries share multidimensional behavioural–epidemiological response profiles relevant to epidemic intelligence.</p> </sec> <sec> <title>METHODS</title> <p>Weekly behavioral constructs, including threat perception, institutional trust, vaccine trust, masking behavior, contact reduction, and isolation compliance, were derived from longitudinal survey data collected through the Imperial College London–YouGov COVID-19 Behavior Tracker. These behavioral time series were examined alongside weekly COVID-19 cases, deaths, the effective reproduction number (Rt), and the policy stringency index. Temporal associations were investigated using autocorrelation and cross-correlation analyses of pre-whitened ARIMA residuals to account for serial dependence. Statistical significance was assessed using block permutation testing with false discovery rate adjustment for multiple comparisons. Principal component analysis and hierarchical clustering were subsequently employed to identify cross-country patterns in behavioural–epidemiological association profiles.</p> </sec> <sec> <title>RESULTS</title> <p>Across 15 countries, 360 behavioural construct–outcome combinations were evaluated over lags from −8 to +8 weeks. Peak associations occurred at both positive and negative lags, but their direction and magnitude varied substantially across countries and domains. Overall, 31 of 360 associations (8.6%) were supported by block permutation tests and 5 (1.4%) remained significant after false discovery rate correction: institutional trust with cases and deaths in Israel; masking behaviour with deaths and Rt in Canada; and isolation compliance with cases in South Korea. Combined clustering produced clearer cross-country separation than construct-specific analyses, indicating multidimensional response profiles. The sparse corrected findings argue against a universal behavioural marker and support context-specific interpretation.</p> </sec> <sec> <title>CONCLUSIONS</title> <p>This study contributes an autocorrelation-aware, lag-resolved framework that treats behavioural data as a potential layer of epidemic intelligence rather than as a static covariate. After prospective validation, leading signals could complement virological and clinical surveillance, whereas lagging signals could quantify delays in public response and expose communication or implementation gaps. The marked heterogeneity and sparse corrected findings show that behavioural indicators should not be transferred uncritically across countries or future outbreaks; their value will depend on real-time collection, local calibration, and validation of incremental forecasting utility.</p> </sec>