Properties of the data sets ordered by the Pearson correlation coefficient between length and rate of messages .

The event type denotes whether an event is unfolding over a non-singular period of time like the (o)ngoing snow storm, if there is an incisive, temporally (s)ingular happening like the winning move in the golf event, or if there is (n)o particular event. Suffixes for ranges and time steps stand for (m)inute, (h)our, (d)ay, (y)ear. The fraction of messages posted in the peak hour, compared with hourly fractions 12 hours prior and afterwards, is denoted by and can be interpreted as the immediacy of the response to singular events. Exponents and measure the best least-squares fit slopes between and , and between and , respectively. The symbol denotes the imposed length limitation in the respective medium. Although the fraction of peak hour messages is highest for the golf event, correlation is stronger in the presidential election forum, possibly due to the length limitation in data set 1. All other correlations are consistent with the type of event, i.e. correlations are less strong when there is only an ongoing event. We checked for robustness of the parameters in Section S4 in File S1.