File S1 - Using Friends as Sensors to Detect Global-Scale Contagious Outbreaks

Contains the following files: 1 An Analytic Elaboration of the Friendship Paradox (2). 2 The Twitter Data (4). 3 Sensor Performance in a Simulated Infection Model (5). 4 Sensor Performance in Real Data (7). 5 Using the Sensor Method with a Small Set of Samples (8). 6 Using the Sensor Method with Hashtag Networks (9). 7 Reproduction Rates of Hashtags as a Factor Affecting Early Detection (11). 8 Twitter, Sensors in Twitter, and Google Trends (12). 9 Friends vs. Most Connected Nodes and Most Connected Friends as Sensors (14). 10 Differences in Sensor and Control Characteristics That Also Affect Propagation (15). 11 Figures (16): S1 The friendship paradoxes (17). S2 Twitter data overview (18). S3 Hashtags first appearance and popularity (19). S4 Most used hashtags, users and Greatest Connected Component size (20). S5 Variations of lead time of using friends as sensors with sample size for hashtags used by more than 0.01% of all users (21). S6 Variations of lead time of using friends as sensors with sample size for hashtags used by more than 0.04% of all users (22). S7 Variations of lead time of using friends as sensors with sample size for hashtags related to important events (23). S8 Variations of lead time of using friends as sensors with sample size for different hashtags (24). S9 Variations of lead time of using friends as sensors with sample size for hashtags for which the method works (25). S10 Variations of lead time of using friends as sensors with sample size for hashtags for which the method does not work (26). S11 Networks and cumulative distributions for hashtags with the biggest lead times (27). S12 Networks and cumulative distributions for hashtags with big lead times (28). S13 Networks and cumulative distributions for more hashtags with big lead times (29). S14 Networks and cumulative distributions for hashtags with small lead times (30). S15 Variations of sensor lead time with final number of hashtag users (31). S16 Exposure rate of sensor and control populations (32). S17 Variations of sensor lead time and divergence alarms with final number of hashtag users (33). S18 Distribution of number of hashtag users for hashtags that trigger a divergence alarm (34). S19 Distribution of number of users for hashtags triggering a divergence alarm vs. not triggering an alarm (35). S20 Twitter hashtags and Using Friends as Sensors vs. Google searches (36). S20 Continued. Twitter hashtags and Using Friends as Sensors vs. Google searches (37). S20 Continued. Twitter hashtags and Using Friends as Sensors vs. Google searches (38). S21 Lead time of using friends as sensors vs. using sensors by degree (39). S21 Continued. Lead time of using friends as sensors vs. using sensors by degree (40). S21 Continued. Lead time of using friends as sensors vs. using sensors by degree (41). S21 Continued. Lead time of using friends as sensors vs. using sensors by degree (42). S21 Continued. Lead time of using friends as sensors vs. using sensors by degree (43). S21 Continued. Lead time of using friends as sensors vs. using sensors by degree (44) S22 Degree and betweenness differences between controls and sensors (45).
(PDF)