G High EngagementFluctuation over time. User engagement with items on CERN’s social media platforms fluctuated strongly over time. Fig 4 represents the pattern of user interactions with items posted on CERN’s Facebook page over the time period studied, normalized by daily audience size. In total, the audience typically engaged with items at a constant rate of interactions throughout the study, as illustrated by the cluster of observations near the x axis. In addition, several outliers were found, some with z-scores as high as 5 or more, meaning that for certain items, user behaviours occurred as often as 5 standard deviations more than the means for those behaviours on the respective platforms. A similar pattern of user behaviour was found in most platforms studied. High-engagement items. Thirty-five (35) high-engagement items were found in the study, comprising more than 16 of the 214 items included in the sample. These were defined as items with at least one user behaviour statistic scoring z 1.96. As an example, the six highengagement items for CERN’s Facebook page are labelled in Fig 4A. The point labelled “Open Data” in Fig 4A, for instance, refers to click-throughs on a link in the Facebook announcement that CERN had launched an Open Data Portal to make the data of LHC experiments publicly available. This item received high standard scores on Facebook in terms of click-throughs per thousand users (z = 5.05) and shares per thousand users (z = 2.47). Hence, it was considered a high engagement item posted on the Facebook platform. (Since the distribution was strongly right-skewed, no low-engagement items were identified in the study.) The lifetime total reach of these Facebook posts, a measure indicating the number of users who potentially could interact with the post, was similar at 11,490 users (SD 13,900). Most of the posts reached around 10,000 users (Mean 10,300, SD 7,565) except for one outlier, “Dishwasher,” which concerned a dishwasher for circuit boards, and reached over 121,000 users (Fig 4B). This indicates that in not all cases was user engagement necessarily driven by increased reach. Associations between high engagement items and topics. High engagement across platforms is Bay 41-4109 supplier significantly associated with item topic. Six (6) topics were repeatedly popular acrossPLOS ONE | DOI:10.1371/journal.pone.0156409 May 27,12 /Engagement with Particle TAK-385 web Physics on CERN’s Social Media PlatformsFig 4. User engagement with scientific content and reach on CERN’s Facebook page over time, October ecember 2014. (A) User engagement with scientific items over time. Zero represents the mean rate for each user behaviour on Facebook per item per 1,000 Facebook followers on the day of sampling: Likes 1.21 IPI/kU (SD 1.86); Comments 0.0779 IPI/kU (SD 0.17), Shares 0.187 IPI/kU (SD 0.32); Click-throughs 0.256 IPI/kU (SD 0.52). pkU: Per Thousand Users. Z: Z-score. The size of CERN’s Facebook audience size grew from 343,000 to 367,000 over the course of the study. (B) Reach of scientific items over time. Reach is the total number of Facebook users the item was served to. doi:10.1371/journal.pone.0156409.gmultiple platforms (hereafter “recurring” high-engagement topics), representing 19 items of the 35 “high engagement” items. For example, the “Open Data” topic received high engagement scores not only on Facebook but also on Google+, Twitter English and Twitter French, making it a recurring high-engagement topic. By contrast, another 16 high-engagement.G High EngagementFluctuation over time. User engagement with items on CERN’s social media platforms fluctuated strongly over time. Fig 4 represents the pattern of user interactions with items posted on CERN’s Facebook page over the time period studied, normalized by daily audience size. In total, the audience typically engaged with items at a constant rate of interactions throughout the study, as illustrated by the cluster of observations near the x axis. In addition, several outliers were found, some with z-scores as high as 5 or more, meaning that for certain items, user behaviours occurred as often as 5 standard deviations more than the means for those behaviours on the respective platforms. A similar pattern of user behaviour was found in most platforms studied. High-engagement items. Thirty-five (35) high-engagement items were found in the study, comprising more than 16 of the 214 items included in the sample. These were defined as items with at least one user behaviour statistic scoring z 1.96. As an example, the six highengagement items for CERN’s Facebook page are labelled in Fig 4A. The point labelled “Open Data” in Fig 4A, for instance, refers to click-throughs on a link in the Facebook announcement that CERN had launched an Open Data Portal to make the data of LHC experiments publicly available. This item received high standard scores on Facebook in terms of click-throughs per thousand users (z = 5.05) and shares per thousand users (z = 2.47). Hence, it was considered a high engagement item posted on the Facebook platform. (Since the distribution was strongly right-skewed, no low-engagement items were identified in the study.) The lifetime total reach of these Facebook posts, a measure indicating the number of users who potentially could interact with the post, was similar at 11,490 users (SD 13,900). Most of the posts reached around 10,000 users (Mean 10,300, SD 7,565) except for one outlier, “Dishwasher,” which concerned a dishwasher for circuit boards, and reached over 121,000 users (Fig 4B). This indicates that in not all cases was user engagement necessarily driven by increased reach. Associations between high engagement items and topics. High engagement across platforms is significantly associated with item topic. Six (6) topics were repeatedly popular acrossPLOS ONE | DOI:10.1371/journal.pone.0156409 May 27,12 /Engagement with Particle Physics on CERN’s Social Media PlatformsFig 4. User engagement with scientific content and reach on CERN’s Facebook page over time, October ecember 2014. (A) User engagement with scientific items over time. Zero represents the mean rate for each user behaviour on Facebook per item per 1,000 Facebook followers on the day of sampling: Likes 1.21 IPI/kU (SD 1.86); Comments 0.0779 IPI/kU (SD 0.17), Shares 0.187 IPI/kU (SD 0.32); Click-throughs 0.256 IPI/kU (SD 0.52). pkU: Per Thousand Users. Z: Z-score. The size of CERN’s Facebook audience size grew from 343,000 to 367,000 over the course of the study. (B) Reach of scientific items over time. Reach is the total number of Facebook users the item was served to. doi:10.1371/journal.pone.0156409.gmultiple platforms (hereafter “recurring” high-engagement topics), representing 19 items of the 35 “high engagement” items. For example, the “Open Data” topic received high engagement scores not only on Facebook but also on Google+, Twitter English and Twitter French, making it a recurring high-engagement topic. By contrast, another 16 high-engagement.
