Percentage errors for popularity change predictions in Twitter database.

For a fixed L = 25 and different values of K (first and second column), we estimated, from ten different Twitter sub-samples (p* = 5%), the number of species having abundance at least K at the unobserved scale 1 − p* = 95% given that they have abundance at least L at the sampled scale p* via estimator (4). The average among the ten sub-samples of the true numbers of species, S1−p*(≥K|≥L), and of the ones predicted by our method, , among the ten sub-samples are displayed in the third and fourth columns, respectively. Finally, in the last two columns, we inserted the mean and the variance of the relative error obtained in the ten predictions. Similar results have been obtained for other values of L and K (see S4 Table in S1 Appendix).