Social media usage patterns during natural hazards - Fig 8

Violin plots showing the distributions of fractional tweet rate change of the users found to be tweeting about Hurricane Sandy as it occurred (a) before and during Hurricane Sandy and (b) the pairs of times collected for the null distribution. Separate violin diagrams are drawn for users whose follower counts fall into each order of magnitude from 100 to 105. On each violin, a black bar indicates a Bayesian 95% confidence interval for the mean of the population distribution given the sample. For the Hurricane Sandy data, the intervals for 100 and 101 are both notably higher than, and don’t overlap with the intervals for any of the higher orders of magnitude in follower count. The same is not true for the null distributions, for which most of the intervals overlap and are generally closer together. The values of endpoints of the intervals are given in S1–S5 Figs.