a, Coefficients representing changes in the two physical distancing behaviours reported on a five-point scale compared with baseline (0–30 days after first required measure), controlling for containment policies, age and gender. b, Equivalent coefficients using an analogous scale for mask wearing, also controlling for facial covering policies. These were estimated in multi-level models using a pseudo-panel from the 14-country survey data and three levels: time (level 1, defined by 10-day intervals), cohorts (level 2, defined by interaction of country, 5-year age ranges and gender) and countries (level 3), with a first-order auto-regressive residual covariance structure. Dashed lines represent baseline behaviour, and error bars indicate 95% confidence intervals. The non-linear pattern in a is confirmed by a significant quadratic time trend in Table 1, which presents results of models with standard controls and residual models, and in Supplementary Fig. 1a–g,k,l and Supplementary Table 1a,b,d,e,h–m, with different specifications (including fixed-effects models with clustered standard errors at the country level, using both pseudo-panels and individual disaggregated survey data) and sets of control variables.