Ents are over, tends to dissipate naturally with time. We point out, however, that there is probably also an element of statistical `regression to the mean’ occurring. We did not choose our communities at random: we chose five of them because they were among those with the most extreme sentiment in the autumn period.7 This introduces a bias andNamely: 4 (weddings), 7 (Islam), 9 (Islam versus atheism), 10 (Madeleine McCann) and 11 (religion).1.nursingIsorhamnetin biological activity nursing Scottish politics Islam versus atheism religion Islam0.9 0.8 0.7 conductance 0.6 0.5 0.friends chattingrsos.royalsocietypublishing.org R. Soc. open sci. 3:…………………………………………wildlife and animals dogs astronomy human resources smoking/e-cigarettes housing sector friends chatting weddingsMadeleine McCannIndian politics0.3 0.2 0.1 0 0.`GamerGate’1.0 1.1 1.2 1.3 1.4 1.5 1.6 user loss factor (no. users active in 28-day autumn period divided by number active in later 28-day PD168393MedChemExpress PD168393 spring period) Louvain weighted Louvain k-clique-communities1.Figure 8. Communities with higher conductance tended to lose more of their users over time.1.2 1.0 mean community sentiment based on (SS) 0.8 0.6 0.friends chatting astronomy nursingfriends chattingweddingshousing sector nursingdogshuman resources0.2 0 0.90 ?.2 ?.4 ?.6 ?.wildlife and animalsIndian politics1.1.1.1.`GamerGate’1.religion1.1.1.Scottish politicssmoking/e-cigarettesIslam versus atheism Islam Madeleine McCannuser loss factor (no. users active in 28-day autumn period divided by number active in later 28-day spring period)Figure 9. Communities with more negative sentiment, measured by (SS), tended to lose more of their users over time.makes it more likely for the sentiment in these five communities to become more moderate by the spring period (which it does, in all five cases). This bias is unavoidable when one disproportionately selects communities with extreme sentiment for study. The correlation coefficient in figure 11 is -0.71. The relationship was less apparent using the other sentiment measures, though still present, with corresponding correlations of -0.59 for (SS) and -0.32 for (L).1.2 mean sentiment over 28-day spring period based on (SS)line of equality1.weddingsrsos.royalsocietypublishing.org R. Soc. open sci. 3:…………………………………………0.human resources dogs y = 0.878x + 0.0146 housing sector nursing friends chatting wildlife and animalsfriends chatting0.6 0.4 0.smoking/e-cigarettes religion astronomy Indian politicsnursing0 ?.1 ?.`GamerGate’?.?.0.0.0.0.0.1.Islam versus atheism scottish politics Islam?.?.6 mean sentiment over 28-day autumn period based on (SS) Louvain weighted Louvain k-clique-communitiesFigure 10. Graph showing that community sentiment was very stable over the 19-week period. The solid line shows where the autumn and spring sentiments are equal.0.3 change in sentiment from autumn to spring using (MC)0.Islam versus atheism Islam religion0.Madeleine McCann Scottish politics wildlife and animals human resources dogs nursing smoking/e-cigarettes?.5 ?.0.`GamerGate’1.1.2.2.astronomy friends chatting housing sector?.Indian politicsfriends chattingweddings?.nursing?.4 mean sentiment over 28-day autumn period based on (MC) Louvain weighted Louvain k-clique-communitiesFigure 11. Graph showing that the sentiment in 16 of the 18 communities became more moderate over time. This plot uses the (MC) measure.The robustness of the weekly sentiment measures suggests that only a limited am.Ents are over, tends to dissipate naturally with time. We point out, however, that there is probably also an element of statistical `regression to the mean’ occurring. We did not choose our communities at random: we chose five of them because they were among those with the most extreme sentiment in the autumn period.7 This introduces a bias andNamely: 4 (weddings), 7 (Islam), 9 (Islam versus atheism), 10 (Madeleine McCann) and 11 (religion).1.nursingnursing Scottish politics Islam versus atheism religion Islam0.9 0.8 0.7 conductance 0.6 0.5 0.friends chattingrsos.royalsocietypublishing.org R. Soc. open sci. 3:…………………………………………wildlife and animals dogs astronomy human resources smoking/e-cigarettes housing sector friends chatting weddingsMadeleine McCannIndian politics0.3 0.2 0.1 0 0.`GamerGate’1.0 1.1 1.2 1.3 1.4 1.5 1.6 user loss factor (no. users active in 28-day autumn period divided by number active in later 28-day spring period) Louvain weighted Louvain k-clique-communities1.Figure 8. Communities with higher conductance tended to lose more of their users over time.1.2 1.0 mean community sentiment based on (SS) 0.8 0.6 0.friends chatting astronomy nursingfriends chattingweddingshousing sector nursingdogshuman resources0.2 0 0.90 ?.2 ?.4 ?.6 ?.wildlife and animalsIndian politics1.1.1.1.`GamerGate’1.religion1.1.1.Scottish politicssmoking/e-cigarettesIslam versus atheism Islam Madeleine McCannuser loss factor (no. users active in 28-day autumn period divided by number active in later 28-day spring period)Figure 9. Communities with more negative sentiment, measured by (SS), tended to lose more of their users over time.makes it more likely for the sentiment in these five communities to become more moderate by the spring period (which it does, in all five cases). This bias is unavoidable when one disproportionately selects communities with extreme sentiment for study. The correlation coefficient in figure 11 is -0.71. The relationship was less apparent using the other sentiment measures, though still present, with corresponding correlations of -0.59 for (SS) and -0.32 for (L).1.2 mean sentiment over 28-day spring period based on (SS)line of equality1.weddingsrsos.royalsocietypublishing.org R. Soc. open sci. 3:…………………………………………0.human resources dogs y = 0.878x + 0.0146 housing sector nursing friends chatting wildlife and animalsfriends chatting0.6 0.4 0.smoking/e-cigarettes religion astronomy Indian politicsnursing0 ?.1 ?.`GamerGate’?.?.0.0.0.0.0.1.Islam versus atheism scottish politics Islam?.?.6 mean sentiment over 28-day autumn period based on (SS) Louvain weighted Louvain k-clique-communitiesFigure 10. Graph showing that community sentiment was very stable over the 19-week period. The solid line shows where the autumn and spring sentiments are equal.0.3 change in sentiment from autumn to spring using (MC)0.Islam versus atheism Islam religion0.Madeleine McCann Scottish politics wildlife and animals human resources dogs nursing smoking/e-cigarettes?.5 ?.0.`GamerGate’1.1.2.2.astronomy friends chatting housing sector?.Indian politicsfriends chattingweddings?.nursing?.4 mean sentiment over 28-day autumn period based on (MC) Louvain weighted Louvain k-clique-communitiesFigure 11. Graph showing that the sentiment in 16 of the 18 communities became more moderate over time. This plot uses the (MC) measure.The robustness of the weekly sentiment measures suggests that only a limited am.
