Fig. 7.

Joint discrete distribution of the number of user shifts between CCs and user memberships to CCs, with marginal univariate distributions. (A) UK 2019. (B) USA 2020. Notably: i) 51% and 93% of all users belonged to more than one CC for UK 2019 and USA 2020, respectively, ii) memberships and shifts are strongly correlated ($\rho =0.73$) in UK 2019 and moderately so ($\rho =0.38$) in USA 2020, iii) the marginal distributions of memberships and shifts are heavy-tailed in both datasets.