Figure 2
The rows in both the heat maps and the table correspond to retweet count, retweet h-index, and retweet PageRank, from the top row to the bottom row. (a, b) The heat maps show the enrichment-depletion patterns of users in the top-decile with respect to each measure, with the same vertical order as in the table, in (a) apolitical and (b) divisive political discussions. User groups A (high-CAP), B (mid-CAP), and C (low-CAP), correspond to the three types with respect to our account automation typology. User group D indicates deactivated or suspended accounts. For each cell, the quantities are centered around the values with a randomly assigned distribution, conditional on the observed population of each row and column, and normalized to a \(-1\) to \(+1\) range. The annotations within each cell show the raw quantities. (c) The ‘Deactivated’ column in the table shows the odds ratios of deactivation or suspension in divisive political discussions compared with apolitical discussions, for users belonging to the top decile by each of the three influence measures—retweet count, retweet h-index, and retweet PageRank. Similarly, the column labeled ‘High CAP vs. Low Cap Users’ in the table shows the odds ratios of belonging to the high-CAP group of users (A), as opposed to the low-CAP group (C), for the same topics and with respect to the same influence metrics. Chi-square tests are performed to verify statistical significance. ‘*’, ‘**’, and ‘***’ indicate significance levels 0.1, 0.05, and 0.01, respectively.
