key: cord-303506-rqerh2u3 authors: Patel, V.; Haunschild, R.; Bornmann, L.; Garas, G. title: A call for governments to pause Twitter censorship: a cross-sectional study using Twitter data as social-spatial sensors of COVID-19/SARS-CoV-2 research diffusion date: 2020-05-29 journal: nan DOI: 10.1101/2020.05.27.20114983 sha: doc_id: 303506 cord_uid: rqerh2u3 Objectives: To determine whether Twitter data can be used as social-spatial sensors to show how research on COVID-19/SARS-CoV-2 diffuses through the population to reach the people that are especially affected by the disease. Design: Cross-sectional bibliometric analysis conducted between 23rd March and 14th April 2020. Setting: Three sources of data were used in the analysis: (1) deaths per number of population for COVID-19/SARS-CoV-2 retrieved from Coronavirus Resource Center at John Hopkins University and Worldometer, (2) publications related to COVID-19/SARS-CoV-2 retrieved from WHO COVID-19 database of global publications, and (3) tweets of these publications retrieved from Altmetric.com and Twitter. Main Outcome(s) and Measure(s): To map Twitter activity against number of publications and deaths per number of population worldwide and in the USA states. To determine the relationship between number of tweets as dependent variable and deaths per number of population and number of publications as independent variables. Results: Deaths per one hundred thousand population for countries ranged from 0 to 104, and deaths per one million population for USA states ranged from 2 to 513. Total number of publications used in the analysis was 1761, and total number of tweets used in the analysis was 751,068. Mapping of worldwide data illustrated that high Twitter activity was related to high numbers of COVID-19/SARS-CoV-2 deaths, with tweets inversely weighted with number of publications. Poisson regression models of worldwide data showed a positive correlation between the national deaths per number of population and tweets when holding the country's number of publications constant (coefficient 0.0285, S.E. 0.0003, p<0.001). Conversely, this relationship was negatively correlated in USA states (coefficient -0.0013, S.E. 0.0001, p<0.001). Conclusions: This study shows that Twitter can play a crucial role in the rapid research response during the COVID-19/SARS-CoV-2 global pandemic, especially to spread research with prompt public scrutiny. Governments are urged to pause censorship of social media platforms during these unprecedented times to support the scientific community's fight against COVID-19/SARS-CoV-2. What is already known on this topic: • Twitter is progressively being used by researchers to share information and knowledge transfer. • Tweets can be used as 'social sensors', which is the concept of transforming a physical sensor in the real world through social media analysis. • Previous studies have shown that social sensors can provide insight into major social and physical events. • Using Twitter data used as social-spatial sensors, we demonstrated that Twitter activity was significantly positively correlated to the numbers of COVID-19/SARS-CoV-2 deaths, when holding the country's number of publications constant. • Twitter can play a crucial role in the rapid research response during the COVID-19/SARS-CoV-2 global pandemic. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted May 29, 2020. . https://doi.org/10.1101/2020.05. 27.20114983 doi: medRxiv preprint Results: Deaths per one hundred thousand population for countries ranged from 0 to 104, and deaths per one million population for USA states ranged from 2 to 513. Total number of publications used in the analysis was 1761, and total number of tweets used in the analysis was 751,068. Mapping of worldwide data illustrated that high Twitter activity was related to high Conclusions: This study shows that Twitter can play a crucial role in the rapid research response during the COVID-19/SARS-CoV-2 global pandemic, especially to spread research with prompt public scrutiny. Governments are urged to pause censorship of social media platforms during these unprecedented times to support the scientific community's fight against COVID-19/SARS-CoV-2. altmetrics, Twitter, spatial maps, COVID-19/SARS-CoV-2 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted May 29, 2020. . https://doi.org/10.1101/2020.05.27.20114983 doi: medRxiv preprint Twitter is a social network created in 2006, that brings together hundreds of millions of users around its minimalist concept of microblogging, allowing users to post and interact with messages known as 'tweets".(1) Twitter has short delays in reflecting what its users perceive, and its principle of "following" users without obligatory reciprocity, together with a very open application programming interface, make it an ideal medium for the study of online behaviour.(2) Tweets can be used as 'social sensors', which is the concept of transforming a physical sensor in the real world through social media analysis. Tweets can be regarded as sensory information and Twitter users as sensors. Studies have demonstrated that tweets analysed as social sensors can provide insight into major social and physical events like earthquakes (3), sporting events (4), celebrity deaths (5), and presidential elections.(6) Twitter data contain location information which can be converted into geo-coordinates and be spatially mapped. In this way tweets can be used as social-spatial sensors to demonstrate how research diffuses within a population. (7) Researchers are increasingly using Twitter as a communication platform, and tweets often contain citations to scientific papers.(8) Twitter citations can form part of a rapid dialogue between users which may express and transmit academic impact and support traditional citation analysis. Twitter citations are defined 'as direct or indirect links from a tweet to a peer-reviewed scholarly article online' (3, 8) , and reflect a broader discussion crossing traditional disciplinary boundaries, as well as representing 'attention, popularity or visibility' rather than influence. (9) . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. We use Twitter data as social-spatial sensors to demonstrate how research on COVID-19/SARS-CoV-2 diffuses through the population and to investigate whether research reaches the people that are especially affected by the disease. We used three sources of data in this study: (1) deaths per number of population for COVID-19/SARS-CoV-2, (2) publications related to COVID-19/SARS-CoV-2, and (3) tweets of these publications. All data was retrieved and analysed between 23 rd March and 14 th April 2020. We used deaths per number of population as a measure of severity of the outbreak of the virus in countries and USA states. We used deaths per one hundred thousand population for country specific data, which was retrieved from Coronavirus Resource Center at John Hopkins University.(12) We used deaths per one million population for US state specific data, which was retrieved from Worldometer, a provider of global COVID-19 statistics trusted by . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted May 29, 2020. . https://doi.org/10.1101/2020.05.27.20114983 doi: medRxiv preprint The deaths per one hundred thousand population for countries ranged from 0 (Ethiopia) to 104 (San Marino). The deaths per one million population for USA states ranged from 2 (Wyoming) to 513 (New York). The total number of publications that were used in the analysis was 1761, and the total number of tweets that were used in the analysis was 751,068 (see supplementary material). One of the problems with Twitter data in the context of this study is that Twitter activity is generally high where more research is done (e.g., Western Europe or the Boston region in Figure 2 ). Since this is not the activity which we intended to measure, we inversely weighted . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted May 29, 2020. Notes. *** p<.001 We did not only use the Twitter data as social-spatial sensors to investigate global trends, but also within a single country. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted May 29, 2020. . https://doi.org/10.1101/2020.05.27.20114983 doi: medRxiv preprint We calculated Poisson regression models with deaths per number of population and number of publications as independent variables and number of tweets as dependent variable. Table 2 reports the results. The results are based on 49 USA states (out of 52) since only USA states with at least one tweet were considered. The percentage changes in expected counts in Table 2 point out that deaths per number of population and Twitter activities are negatively correlated: for a standard deviation increase in the deaths per number of population of a USA state, the expected number of tweets in that state decreases by 10.6%, holding the USA state's number of publications constant. The results in Table 2 further show that the influence of the number of publications is significantly greater than that of the deaths per number of population (and positive). In the USA states, there is a strong dependency of Twitter data on the number of publications. Figure S1 demonstrates that at the time of the analysis the USA was an outlier because of lower national deaths per number of population and higher numbers of publications . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted May 29, 2020. . https://doi.org/10.1101/2020.05.27.20114983 doi: medRxiv preprint and tweets, when compared to other countries that were significantly impacted by COVID-19/SARS-CoV-2 (e.g., UK, France, Spain, and Italy). Social media can be an effective tool for broadcasting research both within and beyond the academic community. (26) Twitter is one of the best social media platforms for sharing scientific research and knowledge because it allows users to post links of recent publications, write a . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted May 29, 2020. is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted May 29, 2020. . https://doi.org/10.1101/2020.05.27.20114983 doi: medRxiv preprint disease. Our study suggests that governments should consider relaxing censorship of social media at times of global crisis, such as the COVID-19/SARS-CoV-2 pandemic. Moreover, allowing researchers greater access to platforms such as Twitter during a global pandemic can aid the scientific community's fight against misinformation and pseudoscience. (32) The USA appears to be an outlier in the worldwide data and the country-specific data shows that the USA has a different relationship between tweets and deaths, both of which may be due to the pandemic reaching the USA later than most other countries in the Northern Hemisphere. Before concluding, it is important to consider the limitations of this study. We have analysed tweets mentioning publications in a quantitative manner which does not account for the association of the tweet with the publication (i.e. a tweet may reference a valid study but claim it to be 'fake news' or have another negative overtone). We have not performed any thematic analysis of the tweets in terms of their content (e.g., are tweets referring to testing for COVID-19/SARS-CoV-2, therapies, or vaccines), quality (e.g., whether tweets are referring to randomised controlled trials or letters), or who tweeted these (e.g., individual researchers, members of the public, universities or pharmaceutical industries). Moreover, no distinction was made between tweets and retweets (of original tweets), which raises the question whether a . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted May 29, 2020. . different handling of retweets could yield different results. These are interesting questions which might be an interesting topic of further research. Despite these limitations, our study has a number of strengths. We have used an evidence-based and robust methodology (see supplementary material) to clean and analyse data, as well as extracting data from several well-established databases containing real world evidence updated in real time.(12-15) Our study comes at a very critical point in time, when a rapid research response is vital to develop therapies and vaccines to slow the COVID-19/SARS-CoV-2 pandemic and lessen the damage caused by the disease. Our study utilising Twitter data as social-spatial sensors can serve as proof-of-concept for future studies on Twitter and the evolving pandemic. COVID-19/SARS-CoV-2 began as a cluster of cases of pneumonia in Wuhan, Hubei Province, but the outbreak quickly progressed from an PHEIC to a pandemic, which highlights the dynamic process of the spread of an infectious disease.(10, 11) Our study has simply investigated a snapshot of the relationship between this pandemic, research outputs, and Twitter activity, and demonstrates the importance of how social media platforms can be crucial to spread research with rapid scrutiny, which may also impede the degree of misinformation. We urge governments to pause censorship of social media platforms such as Twitter during these unprecedented times to support the scientific community's battle against COVID-19/SARS-CoV-2. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted May 29, 2020. . https://doi.org/10.1101/2020.05.27.20114983 doi: medRxiv preprint ACKNOWLEDGEMENTS Meta-data for publications were downloaded via the Dimensions API. Twitter data were retrieved from the Altmetric.com API. Tweets with their location information were retrieved from the Twitter API. The authors thank Rodrigo Costas (CWTS) and Stacy Konkiel (Altmetric.com) for helpful discussions regarding the analysis of location information of Twitter users. The lead author affirms that the manuscript is an honest, accurate, and transparent account of the study being reported. No important aspects of the study have been omitted, and any discrepancies from the study as planned have been explained. The full data set and the statistical code can be obtained, upon request, from the corresponding author. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted May 29, 2020. . https://doi.org/10.1101/2020.05.27.20114983 doi: medRxiv preprint We have read and understood the medRxiv policy on declaration of interests and confirm that we have no conflict of interests. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted May 29, 2020. . https://doi.org/10.1101/2020.05.27.20114983 doi: medRxiv preprint A social network analysis of Twitter: Mapping the digital humanities community Earthquake shakes Twitter users: real-time event detection by social sensors Twitter as Social Sensor: Dynamics and Structure in Major Sporting Events. ALIFE14 TwitterStand: news in tweets. Proceedings of the 17th ACM SIGSPATIAL International Conference on Advances in Geographic Information Systems Tweet the debates: understanding community annotation of uncollected sources Are papers addressing certain diseases perceived where these diseases are prevalent? The proposal of Twitter data to be used as social-spatial sensors How and why scholars cite on Twitter Can alternative indicators overcome language biases in citation counts? A comparison of Spanish and UK research Statement on the second meeting of the International Health Regulations (2005) Emergency Committee regarding the outbreak of novel coronavirus (2019-nCoV) World Health Organization2020. 11. WHO Director-General's opening remarks at the media briefing on COVID-19 -11 Worldometer's COVID-19 data: Worldometer Global research on coronavirus disease (COVID-19): World Health Organization StataCorp. Stata statistical software: release 15 Working with spmap and maps Guide to creating maps with Stata Limited dependent variable models and probabilistic prediction in informetrics Modelling count data Regression models for categorical dependent variables using Stata Detecting influenza epidemics using search engine query data Twitter as a tool for health research: A systematic review Social impact in social media: A new method to evaluate the social impact of research Peer review: Trial by Twitter Tweeting biomedicine: An analysis of tweets and citations in the biomedical literature Academic information on Twitter: A user survey Chinese and Iranian Scientific Publications: Fast Growth and Poor Ethics Pseudoscience and COVID-19 -we've had enough already. Nature. 2020. 33. Crew B, Jia H. Leading research institutions in the Nature Index 2020 Annual Tables: Nature Index