key: cord-0900191-3o0ych0w authors: Wang, Yan; Hao, Haiyan; Sundahl Platt, Lisa title: Examining risk and crisis communications of government agencies and stakeholders during early-stages of COVID-19 on Twitter date: 2020-09-23 journal: Comput Human Behav DOI: 10.1016/j.chb.2020.106568 sha: 1bffada609b0ea9904f9ee2220859878a1022e96 doc_id: 900191 cord_uid: 3o0ych0w During COVID-19, social media has played an important role for public health agencies and government stakeholders (i.e. actors) to disseminate information regarding situations, risks, and personal protective action inhibiting disease spread. However, there have been notable insufficient, incongruent, and inconsistent communications regarding the pandemic and its risks, which was especially salient at the early stages of the outbreak. Sufficiency, congruence and consistency in health risk communication have important implications for effective health safety instruction as well as critical content interpretability and recall. It also impacts individual- and community-level responses to information. This research employs text mining techniques and dynamic network analysis to investigate the actors’ risk and crisis communication on Twitter regarding message types, communication sufficiency, timeliness, congruence, consistency and coordination. We studied 13,598 pandemic-relevant tweets posted over January to April from 67 federal and state-level agencies and stakeholders in the U.S. The study annotates 16 categories of message types, analyzes their appearances and evolutions. The research then identifies inconsistencies and incongruencies on four critical topics and examines spatial disparities, timeliness, and sufficiency across actors and message types in communicating COVID-19. The network analysis also reveals increased communication coordination over time. The findings provide unprecedented insight of Twitter COVID-19 information dissemination which may help to inform public health agencies and governmental stakeholders future risk and crisis communication strategies related to global hazards in digital environments. The outbreak of the 2019 novel coronavirus disease (COVID-19) caused by severe acute respiratory 24 syndrome coronavirus 2 (SARS-CoV-2) in the U.S. in January 2020, has resulted in explosive and 25 escalating communication across online environments related to the disease, outbreak trajectory, impact 26 on human mortality, and global and local implications faced by government and health system agencies. The quick and exponential spreading of SARS-CoV-2 has ignited social media with a diversity of 28 information. The increasing rate of detected incidents of COVID-19 along with massive amounts of 29 related posts has triggered divergent reactions (Shimizu, 2020) and interactions across government 30 agencies and stakeholders at various levels. 31 Among all media sources, Twitter, the largest microblogging platform nationally and globally, has 32 played a particularly important role in communicating SARS-CoV-2 and COVID-19 information. This is 33 especially apparent in information dissemination on personal protective action inhibiting disease spread 34 (e.g. wearing masks, reducing travel, social distancing, and teleworking). The World Health Organization 35 (WHO) and the U.S. federal and state health agencies (agencies hereafter) and other federal agency 36 stakeholders (stakeholders hereafter) whose operations are related to stemming the COVID-19 outbreak 37 have consecutively published virus and disease-related content through their Twitter accounts. These 38 actors' crisis and risk communication can provide credible sources of information during the unfolding of 39 a crisis (Lin et al. 2016 ). The predominated information from trusted sources can also suppress the 40 propagation of rumors (Aguirre & Tierney, 2001) . Previous studies have identified several other key 41 factors for the best practices of risk and crisis communication, such as the message update speed (Lin et al. 42 2016), and cooperation with the similar organization (Seegar, 2006) . 43 There have been noted insufficient communications, inconsistent and incongruent messages regarding 44 SARS-CoV-2 and COVID-19 and risks from different agencies and stakeholders. This phenomenon was 45 especially salient at the early stage of the outbreak. Twitter users have difficulties in assimilating and 46 making meaningful interpretations of disparate information from multiple sources (Ippolito, et al., 2020) . 47 Both consistency and congruence are the key factors to effective communication about SARS CoV-2 48 transmission and COVID-19 (Seeger, 2020) as individuals' sense of the perceived threat rests largely on 49 the information that they have received from agencies and stakeholders. Consistency refers to "similarity 50 between the tone of the message and the information contained therein" (Glik 2007, pp38) . We use this 51 metric to stress the reinforcement of similar messages and attitudes over time. Congruence implies 52 communication agencies and stakeholders have settled on a single, unifying interpretation of the risk and 53 crisis (Sellnow et al. 2008 ). We use congruence to differentiate with consistency by addressing the 54 message uniformity across communication actors during a similar timeframe. Additionally, sufficient 55 communications can lead to higher perceived risk and increased appropriate response while general and 56 vague messages may cause people not to act (Glik, 2007) . 57 Research in communication related to health warnings suggests that message congruence may evoke 58 semantic priming effects that increase processing fluency of message recipients and improve attention-59 recall of relevant information (Lochbuehler et al. 2018 ). Timely and transparent dissemination of accurate, 60 science-based information about the virus and pandemic and the progress of the response can build public 61 trust and confidence (Reynolds 2014 ). Barriers to effectively communicating the situation, risk and 62 controlling actions can also cause confusion. This phenomenon can lead to inappropriate behavioral 63 contagion that can span a continuum of ignoring recommendations for physical distancing and self-64 quarantine on one end of the spectrum to panic buying, aggression, and unnecessary visits to health-care Coordinating risk communication during crises and emergencies among agencies and stakeholders is 162 critical because an individual actor simply does not have all the necessary resources needed to address 163 unanticipated problems for all (Reynolds & Seeger, 2020) . One major concern in response coordination 164 during the disease outbreak is ensuring timely and consistent information sharing as the uncertainty of an 165 emergency increases the need for information by the public (Hughes & Tapia, 2015) because appropriate 166 information could make substantial improvements in the response process (Glik, 2007) . Existing studies 167 that have examined the organizational emergency and crisis response on social media have to date mainly 168 concerning disasters. These studies that consider governmental agencies can be categorized into; (a) 169 coordination within agencies, such as online cross-sector communication behaviors for emergencies on 170 social media (Wukich, et al., 2019) ; (b) coordination between agencies and the public, e.g. challenges in 171 coordination between professional emergency responders and digital volunteers (Hughes & Tapia, 2015) ; 172 and (c) coordination within and across groups, e.g. four-channel communication model (Pechta, et al., 173 2010). However, among prior work, not many studies specifically focus on social media or 174 comprehensively examine the influences and reactions between agencies, stakeholders, and the general 175 public through the lens of social media risk communication. Few have evaluated the risk communication 176 coordination in terms of information consistency and congruency among agencies on social media during 177 a pandemic outbreak. 178 In the research field of public health, social media has been studied for early detection of epidemic 180 outbreaks as part of the web surveillance system and to predict infectious disease outbreaks (Velasco et We focus on Twitter, from which 22% of people living in the U.S. retrieve news (Hughes & Wojcik, 197 2019). We used the Twitter User Timeline API (Twitter, 2019) to query Tweets posted by the official 198 accounts of the WHO, 12 federal agencies, six governmental stakeholders, and 50 state-level public 199 health agencies (i.e. Department of Health or DOH). Two agencies, the Wyoming DOH and the 200 Department of Homeland Security (DHS), were found to post no tweet during the study period and were 201 hence excluded from the analyses. Table 1 listed our studied communication actors and their Twitter user 202 names. 203 The parameters of our study period are from January 1, 2020, the day after the WHO officially 204 announced the presence of the novel virus, to April 27, 2020 (117 days in total) when the virus resulted in 205 nearly one million confirmed cases and claimed about 56,000 lives in the U.S. (Johns Hopkins University, 206 2020). We chose this timeframe for the study period due to the amount of observed conflicting 207 information, misinformation, and other risk communication incongruencies across agencies and 208 stakeholders. These early months are also critical for crisis responders to impact the general public's 209 preventative behaviors (CDC, 2014). 210 First, we filtered tweets using COVID19-relevant keywords including "coronavirus," "corona," "sars-cov-213 2," "ncov," and "covid," and identified 13,598 relevant tweets, which roughly equals one-third of the total 214 tweets' volume posted by the agencies and stakeholders in our study period. Tweets posted over the study 215 period not adhering to the basis of study context that was omitted from the analysis address other diseases 216 such as seasonal flu, HIV, smoking, heart attack, and other health-related events and activities. 217 Then we manually annotated the 13,598 relevant messages with 16 categories by reading each 218 message and assigning the category. One tweet can be maximally assigned to two message types based on 219 the conveyed major information. The 16 categories and descriptions are listed in Example strategies can be "wash hands", "wear masks", and "disinfect the house". 2 Order The executive orders, policies, announcements that are issued by governmental agencies in the context of pandemics. For example, many states have issued "social distancing" and "stay at home" order. The federal government also issued orders to restrict national and international traveling. 3 Situational information The message describing the influence or associated risks of the pandemic, which supports situational awareness of the general public. Examples can be the number of infection cases or deaths or the assessed risks by authoritative agencies. The message about the closure of critical facilities, businesses, services, etc., for example, the closure of non-essential businesses. The message about the openings of critical facilities, businesses, services, etc., for example, the openings of testing sites, temporary hospitals, call centers. 6 Operations. The operations that the agency implemented or planned or the change of policies that the agency will follow. For example, some agencies automatically extended expired driver licenses. FEMA constructed new hospitals. 7 Resource provision The message providing information about resources (e.g., research funding, hospital capacities, medical supplies, relief funding and benefits, childcare services for front-line workers) that are available for relieving the pandemic impacts. 8 Clarification The clarification or explanation on the issued orders or policies or the correctness of before statements made by authoritative agencies. For example, some states specifically mentioned that people won't break the law if they do not observe the stay at home order. 9 Rumor/scam management Tweets that are posted to correct the rumor, false news, and warn scams and frauds. For example, some states warned the public to be aware of fake medical supplies/suppliers. 10 Volunteer/donation Tweets calling for volunteers and donations. 11 Employment Tweets advocating job positions or administrative roles. For example, many state agencies have posted tweets to recruit healthcare workers and professionals. 12 Opinion and commentary Tweets that express opinions (e.g. sadness, gratefulness), ideas, or comments toward situations caused by the pandemic, news, policies, events, and so on. 13 External resources/knowledge. Tweets that provide links to external information resources (e.g. websites, hotlines, videos, newsletters, presses, briefings, interviews, articles, fact sheets, science reports, online portals, apps, etc.) or advocate general knowledge or scientific findings related to the pandemic. 14 Guidance on other diseases or events. Tweets that provide information sharing and extra guidance on individuals managing other diseases or health conditions such as asthma, heart attack, mental stress, or other impacted events (e.g. voting, census) during the pandemic. 15 Event schedules and agendas. Tweets providing information on planned events such as media briefing, interview, meeting, conference, etc. For example, many state governors provide routine briefing during the pandemic and the state agencies held meetings to interview experts to talk about the COVID-19. 16 Intelligence gathering Tweets that collect answers, ideas, and solutions from the public. For example, some states encourage recovered patients to share their experiences and coping strategies. We conducted dynamic network analysis to examine the communication coordination among different 226 actors. Dynamic network analysis has demonstrated its effectiveness is investigating cross-actor risk and 227 crisis communications during events e.g. West Virginia water crisis (Getchell & Sellnow 2015) . We average weighted degree, average path length, network diameter, and modularity). In this study, the 241 average weighted degree and density represent the general frequencies of retweeting and mentioning 242 among studied actors. Higher degree or density refers to more coordination between agencies, which 243 suggests more congruent information to the public. The average path length is the mean of links between 244 all actor pairs, and diameter is the maximum number of links that connect two agencies. Shorter path 245 length or diameters suggests a more connected communication network (Tabassum et al., 2018) . We also 246 used the modularity optimization to divide the network into communities, which is determined by 247 comparing the number of edges within communities and expected numbers when the edges are randomly 248 distributed (Blondel et al., 2008) . We have identified four critical topics (wearing masks, assessment of risks, stay at home order, and 303 disinfectant and sanitizer) that significantly impact individuals' preventative behavior under distinct 304 message types that worth more detailed examination in terms of communication congruence, consistency 305 and sufficiency during our investigated time window. These messages are listed in the Supplementary 306 Tables. 307 Preventative strategies e.g. wash hands and disinfect surfaces have been congruently communicated 309 across actors, but guidance on whether the general public should wear masks and what type of masks they 310 wear presents an idea that has evolved over time (see Supplementary COVID-19 has also evolved over the early stages, from "no evidence that wearing masks would limit the 325 virus spread" in April to "simulations indicate that universal masking that includes non-symptomatic 326 health persons may reduce potential exposure risk from SARS COV-2" in June (WHO, 2020). These 327 changing attitudes across agencies toward wearing masks can incur inconsistent behavior of the public. During the pandemic, agencies have assessed the overall risk of the coronavirus several times (see 337 Supplementary Table 2 ). In the beginning, agencies reported low-risk levels of coronavirus to U.S. 338 citizens. For example, the CDC and WA both posted contents mentioning the low-risk situation of the 339 U.S. after the confirmation of the first infection case. Some agencies showed more concern for the 340 seasonal flu than SARS-COV-2. In late February, it was still believed that the risk of coronavirus to 341 people in the U.S. was low, but the rapidly evolving global situation has arisen many agencies' awareness. 342 The initial relevant messages highlighted that older people and those with underlying health conditions 343 are at higher risk. Guidelines and strategies for health protection are especially recommended to this 344 population. This statement changed as more cases were reported for young adults. ME DOH and SD 345 DOH reminded the public that young people were not immune to coronavirus by sharing the 346 hospitalization data. Another concern about the virus is the presence of community spreading. In the early 347 stage, there was little evidence of community spreading in the U.S. CDC reported this on February 25 in 348 two tweets based on available data during that moment. The two tweets have been retweeted more than 349 14,000 times. The CDC reported the community spread cases in California, Oregon, and Washington on 350 February 29, but still hold optimistic attitudes on the situation. The prevalence of optimistic bias has 351 precedence in recent pandemics including H1N1 and the first SARS outbreaks ( Taylor Stay-at-home orders were given at different levels not synchronously among states (see Supplementary 361 Table 3 for examples). In the beginning, stay-at-home order is only recommended for all sick people. In 362 early March, ND DOH and the CDC recommended people who returned from infectious countries and 363 were possibly exposed to COVID-19 to stay home for 14 days. Later in mid-March, people in good health, 364 especially the non-essential workers, children, and older adults, are also recommended to stay at home to 365 limit their contact with sick people. States started to issue executive orders and make the stay at home a 366 mandated strategy in late March but differed in the starting time. NJ, ID, and OH were among the first to 367 issue the order on March 21 and 22. Some states closed some counties or regions at the beginning but 368 turned to state-wide order later. Accompanied by the stay-at-home order are the closure or reduced 369 services of nonessential businesses, schools, and long-term care facilities, and nursing homes. Some states 370 initially scheduled the order for a while around three weeks but later extended it for a longer time. Agencies also posted external resources to explain and clarify the stay-at-home order to relieve the 372 public's panic. 373 Many agencies advocated cleaning and disinfecting often-touched surfaces, sanitizing hands as effective 376 countermeasures to slow down the virus spreading (Supplementary Table 4 ). However, the 377 communications appear to be insufficient in volume in January and February (see Figures 2 and 4) . Some 378 states forwarded videos on house-made sanitizers in March when experiencing the shortage of 379 disinfectant products. The advocating of sanitizers and disinfectants, unfortunately, increased the 380 exposure of the public to poisonous substances and vapors as well as improper use cases. For instance, the 381 FDA reported increased cases of ingestion of hand sanitizers on March 28 and April 15 respectively. ID 382 and NM posted tweets to avoid ingesting disinfectant as a treatment for the novel coronavirus on April 24 383 following a comment made by the president during a televised press briefing related to COVID-19 status 384 in the U.S. This resulted from several states reporting that their Poison Control Centers had received calls 385 about individuals ingesting household disinfectant as a way to combat COVID-19 after the April 2020 386 White House briefing. A subsequent survey conducted by the CDC on the appropriate use of household 387 disinfectants also indicated that 39% of respondents had misused cleaning agents in some manner that 388 resulted in adverse health effects (Gharpure et al., 2020) . 389 Stakeholders 391 To identify the spatial disparities in communication frequency and timeliness on Twitter, we mapped 392 frequencies of relevant messages over 50 states in the U.S. (See Figure 3) . In general, states in the 393 Northeastern area of the U.S. have posted more tweets than other states. The Massachusetts DOH tweeted 394 the most while the Wyoming DOH did not post any Twitter messages during the study period. We have 395 also summarized the first dates of the actors communicating the pandemic on Twitter in Figure 4 . Each 396 colored grid represents the first date (x-axis) when the actor (y-axis) started to post the type of message 397 (legend) on Twitter. We found that most state agencies started to post COVID-related tweets during the 398 week of January 20 to 26 while several federal health agencies (i.e., CDC, IDSA, NIAID, and HHS) 399 started the discussion before January 20. We employed dynamic network analysis to investigate how information flows among the investigated 424 agencies and stakeholders over time. In total, we have 67 nodes representing the 67 investigated agencies 425 including federal stakeholders, health departments, state health departments, and the WHO. 426 We constructed the aggregated communication network over our study period ( Figure 5) agencies mainly connected with state governors, and health professionals that are not discussed in this 440 paper. 441 We further partitioned the networks into communities that are densely connected internally. We 442 identified four closely connected communities in the aggregated network ( Figure 6 ). Specifically, 443 Community 1 includes 42 agencies with the CDC as the central node. Community 2 has four nodes, three 444 of which are federal health departments that concern research and laboratory tests of the novel 445 coronavirus (i.e. NIH, NIMH and NIAID We further examined the dynamics of the weekly communication networks among actors over the 16 456 consecutive weeks (Figures 7 and 8) . The network is very sparse in the first three weeks (Figure 8 ). In 457 Week 1, only two U.S. agencies, IDSA (posted the first COVID-19 related tweet on January 9) and CDC, 458 started to post the COVID-19 information. In Week 2, additional two federal health departments (NIAID 459 and HHS) and two state DOHs (RI DOH and NJ DOH retweeted the WHO and CDC's posts) started to 460 communicate the situation. Federal stakeholders including FEMA and FAA began to communicate the 461 crisis and risk in Week 3. Other stakeholders (e.g. DOT, FAA and FTA) did not disseminate the 462 crisis/risk of the pandemic using their social media accounts until Week 9. We also noticed that federal 463 stakeholders retweeted information from the CDC and HHS. Some transportation-related stakeholders, 464 i.e., DOT, FAA, and FTA more frequently communicate with each other. The network connectivity 465 reaches the first small peak in Week 4 (see network density and average weighted degree in Figure 7 ). In 466 that week, the CDC tweeted an order that advised the public to cancel nonessential traveling to China. 467 The network connectivity then decreases over the following three weeks (January 27 -February 17), 468 though more agencies started to post relevant messages independently (evident in the increased network 469 diameter and average path length). The COVID-19 pandemic has been collectively considered as a 470 pandemic since Week 4 due to increasing reported cases across countries. The connectivity in agencies 471 and stakeholders' communication network also kept growing since late February to early April as more 472 suspected or confirmed cases were reported. Large-scale infectious disease outbreaks are a devastating public health "disaster" around the world. The 480 epidemiology of viruses such as SARS CoV-1, MERS, and SARS-CoV-2 has been varied making the 481 virulence trajectories of emergent coronaviruses difficult to predict. More uncertainties of the disease will 482 be revealed, and more response actions will be implemented. It is urgent and time-critical that we track 483 and understand the dynamics and influences of risk communications of agencies and stakeholders on 484 social media. This study analyzed the risk and crisis communication in terms of sufficiency, timeliness, 485 congruence, consistency, and coordination among public health agencies and federal stakeholders at the 486 early stage of an infectious disease outbreak. The analysis results reveal that agencies and stakeholders, 487 though underestimating the pandemic risk at the beginning, have paid increasing attention to the crisis 488 over the study period. Substantial efforts on the part of US health agencies are being made to convey 489 situational awareness and to educate the public on preventative strategies. The analysis of agencies and 490 stakeholders' tweets identifies insufficiency, incongruency and inconsistency across critical message 491 types. The dynamic network analysis showed a changing communication pattern among agencies and 492 stakeholders with an increased level of connectivity and coordination during the study period (early-stage 493 response). Disentangling the interactive influences of risk communication actors is instrumental in 494 furthering information and education about the communication science of virus transmission and 495 prevention on social media. The study also has a few limitations that will be addressed in future research. 496 First, this research focuses on risk and crisis communication of agencies and stakeholders, so we only 497 studied tweets posted by their official accounts. Future empirical studies can also investigate public 498 behavioral impacts responding to insufficient, inconsistent and incongruent risk communication over the 499 full cycle of the pandemic. Because this research is an observational study of Twitter-based 500 communication related to COVID-19, it is not appropriate to draw inferences about behavioral impacts 501 related to platform or message type dissemination of specific social media (e.g. Twitter) users. Second, 502 the research conducts an analysis for COVID-19. Future research can be extended to different types of 503 infectious disease outbreaks to identify general strategies for sufficient, congruent, and effective risk 504 communication. Third, we used data collected from Twitter only as Twitter is one of the largest short 505 blogging platforms in the U.S. and it provides open APIs. In the future, a cross-platform investigation 506 may generate more comprehensive findings once data from other social media become available. Lastly, 507 future work can also compare findings of communication congruency from public health crises and 508 natural hazards disasters to compare the commonalities and differences in persuasive communication 509 strategies. 510 The ultimate success of a public health campaign in the wake of a pandemic such as COVID-19 511 is dependent on effective population interaction with health agency communication and uptake of ideas. 512 Considering that both individual and community responses to health communication are emergent in 513 nature it can be difficult to ascertain the effectiveness of such a health crisis campaign with the 514 immediacy needed to ensure alignment of health information provision and consequent human reaction. 515 Using Twitter message dissemination analysis provides an important basis for the understanding of health 516 crises and risk communication of official agencies and stakeholders. This would offer potential for insight 517 generalizability on information dissemination attributes (e.g. frequency, timing, message types and 518 coordination) helpful in guiding future global emergent health crisis communication strategies. 519 Furthermore, this study of social media-based crisis risk communication related to a global health event 520 such as COVID-19 provides an opportunity for these findings to be parsed and evaluated further and more 521 extensively within the public response. This would be valuable in interpreting to what extent sufficient, 522 congruent, consistent, or coordinated risk and crisis communication can generate meaningful and 523 effective reaction from messaging targets. The research findings lead to fundamental knowledge of social 524 media risk and crisis communication in large-scale hazards (e.g. pandemic and disasters) by bridging 525 public health and disaster and emergency management. The research also provides public health agencies, 526 first responders and other government stakeholders with an updated understanding of their role in 527 disseminating crisis and risk information on social media. The outcomes will have the potential to 528 understand how to better reduce the risk of inappropriate behaviors and preventable deaths caused by 529 insufficient risk communication, incongruent and inconsistent information harmful to human health and 530 wellbeing as well as how to improve existing crisis risk communication plans, critical health information 531 dissemination efficacy and coordination during unprecedented health crises in the fragmented 532 communication world. 533 Testing Shibutani's prediction of information seeking behavior in 535 rumor Can companion animals become infected with Covid-19 Quality of online information on breast cancer treatment options Gephi: an open source software 541 for exploring and manipulating networks Fast unfolding of communities in 544 large networks Development of an education campaign to reduce delays in pre-hospital response to stroke CDC Advises Travelers to Avoid All Nonessential Travel to China Stop the Spread of Rumors Impact assessment of non-pharmaceutical 554 interventions against coronavirus disease 2019 and influenza in Hong Kong: an observational 555 study To mask or not to mask: Modeling the potential for face mask use by the general 558 public to curtail the COVID-19 pandemic Knowledge and Practices Regarding Safe Household 560 Cleaning and Disinfection for COVID-19 Prevention -United States A network analysis of official Twitter accounts during the West 564 Virginia water crisis Face masks for 566 the public during the covid-19 crisis Risk communication for public health emergencies The impact of social and cultural environment on health Behavior, and the Social Environment: Moving Beyond the Nature/Nurture Debate COVID-19 pandemic: stop panic abandonment of household 573 pets Social media in crisis: When professional responders meet digital 575 volunteers Toning down the 2019-nCoV 577 media hype-and restoring hope Cumulative Cases Uncertainty in a Time of Coronavirus Universal masking is 584 urgent in the covid-19 pandemic: Seir and agent based models, empirical validation Exploring optimistic bias and the integrative model of behavioral 587 prediction in the context of a campus influenza outbreak The biggest pandemic risk? Viral misinformation Crowdsourcing data to mitigate epidemics. The Lancet Digital Health Effect of message congruency on attention and recall in pictorial health warning labels Crisis communication, learning and 595 responding: Best practices in social media Globalization of Public Health Communication: 597 Preparing Local Leaders using the Supercourse. Pandemics and Bioterrorism: Transdisciplinary 598 Information Sharing for Decision-making Against Biological Threats Coronavirus: the spread of misinformation Many college students underestimate 601 diabetes risk COVID-19 A systematic framework for designing and evaluating 605 persuasive systems 10 facts about Americans and Twitter The Effects of Social Media Use on Preventive Behaviors during 610 Infectious Disease Outbreaks: The Mediating Role of Self-relevant Emotions and Public Risk 611 Perception Understanding the dynamics of emergency 613 communication: Propositions for a four-channel model Covid-19 Exposes Digital Divide in Healthcare Fake and spam messages: Detecting misinformation during natural 618 disasters on social media CERC Manual Crisis Emergency Risk Communication. (CERC) Rhode Island Commuting Patterns Crisis communication researcher shares 5 key principles that officials should 625 use in coronavirus Best practices in crisis communication: An expert panel process Effective risk communication: A 631 message-centered approach 2019-nCoV, fake news, and racism. The Lancet The airborne lifetime of small speech 634 droplets and their potential importance in SARS-CoV-2 transmission Rumors, false flags, and 637 digital vigilantes: Misinformation on Twitter after the 2013 Boston marathon bombing Best practices in risk and crisis communication: Implications 640 for natural hazards management The Psychology of Pandemics: Preparing for the Next Global Outbreak of Infectious 642 Disease Social network analysis: An overview Wiley Interdisciplinary Reviews: Data Mining and Knowledge Discoverage GET statuses/user_timeline | Twitter Developers Opening up America Again Social media and 651 internet based data in global systems for public health surveillance: a systematic review. The 652 Milbank Quarterly Institutional trust and 654 misinformation in the response to the 2018-19 Ebola outbreak in North Kivu, DR Congo: a 655 population-based survey Coronavirus disease (COVID-19) advice for the public: When and how to use masks Aggregated Responses of Human Mobility Patterns to Severe 660 Winter Storms: An Empirical Study Coupling Sentiment and Human Mobility in Natural Disasters: A 662 Twitter-based Study of the DUET: A Data-Driven Approach based on Latent Dirichlet Allocation 664 (LDA) Topic Modeling Warning Message Elements and Retweet Counts: An 666 Analysis of Tweets Sent during Hurricane Irma Conceptualizing and implementing an 668 agent-based model of information flow and decision making during hurricane threats Rethinking 671 social amplification of risk: Social media and Zika in three languages Government Social Media Messages across Disaster Phases Cross-Sector Emergency Information Networks on Social 676 Media: Online Bridging and Bonding Communication Patterns. The American Review of Public 677 Administration the Assessment of twitter's 679 potential for outbreak Detection: Avian Influenza Case Study Tracking urban geo-topics based on dynamic topic model. Computers, 681 Environment and Urban Systems