• Acar, A., and Y. Muraki, 2011: Twitter for crisis communication: Lessons learned from Japan’s tsunami disaster. Int. J. Web Based Commun., 7, 392402, https://doi.org/10.1504/IJWBC.2011.041206.

    • Search Google Scholar
    • Export Citation
  • Ash, K. D., R. L. Schumann, and G. C. Bowser, 2014: Tornado warning trade-offs: Evaluating choices for visually communicating risk. Wea. Climate Soc., 6, 104118, https://doi.org/10.1175/WCAS-D-13-00021.1.

    • Search Google Scholar
    • Export Citation
  • Berren, M. R., A. Beigel, and S. Ghertner, 1980: A typology for the classification of disasters. Commun. Mental Health J., 16, 103111, https://doi.org/10.1007/BF00778582.

    • Search Google Scholar
    • Export Citation
  • Bonsón, E., S. Royo, and M. Ratkai, 2015: Citizens’ engagement on local governments’ Facebook sites. An empirical analysis: The impact of different media and content types in western Europe. Gov. Inf. Quart., 32, 5262, https://doi.org/10.1016/j.giq.2014.11.001.

    • Search Google Scholar
    • Export Citation
  • Brantner, C., K. Lobinger, and I. Wetzstein, 2011: Effects of visual framing on emotional responses and evaluations of news stories about the Gaza Conflict 2009. J. Mass Commun. Quart., 88, 523540, https://doi.org/10.1177/107769901108800304.

    • Search Google Scholar
    • Export Citation
  • Brotzge, J., and W. Donner, 2013: The tornado warning process: A review of current research, challenges, and opportunities. Bull. Amer. Meteor. Soc., 94, 17151733, https://doi.org/10.1175/BAMS-D-12-00147.1.

    • Search Google Scholar
    • Export Citation
  • Bubeck, P., W. J. Botzen, and J. C. Aerts, 2012: A review of risk perceptions and other factors that influence flood mitigation behavior. Risk Anal., 32, 14811495, https://doi.org/10.1111/j.1539-6924.2011.01783.x.

    • Search Google Scholar
    • Export Citation
  • Cameron, M. A., R. Power, B. Robinson, and J. Yin, 2012: Emergency situation awareness from Twitter for crisis management. Proc. 21st Int. Conf. on World Wide Web, Lyon, France, International World Wide Web Conference Committee, 695–698, https://doi.org/10.1145/2187980.2188183.

  • Carley, K. M., M. Malik, P. M. Landwehr, J. Pfeffer, and M. Kowalchuck, 2016: Crowd sourcing disaster management: The complex nature of Twitter usage in Padang Indonesia. Saf. Sci., 90, 4861, https://doi.org/10.1016/j.ssci.2016.04.002.

    • Search Google Scholar
    • Export Citation
  • Casteel, M. A., and J. R. Downing, 2013: How individuals process NWS weather warning messages on their cell phones. Wea. Climate Soc., 5, 254265, https://doi.org/10.1175/WCAS-D-12-00031.1.

    • Search Google Scholar
    • Export Citation
  • Chatfield, A. T., and U. Brajawidagda, 2013: Twitter early tsunami warning system: A case study in Indonesia’s natural disaster management. Proc. 46th Hawaii Int. Conf. on System Sciences, Wailea, HI, IEEE, 2050–2060, https://doi.org/10.1109/HICSS.2013.579.

  • Chavez, C., M. A. Repas, and T. Stefaniak, 2010: A new way to communicate with residents: Local government use of social media to prepare for emergencies. ICMA Rep., 16 pp., http://www.unapcict.org/ecohub/a-new-way-to-communicate-with-residents-local-government-use-of-social-media-to-prepare-for-emergencies/at_download/attachment1.

  • Chew, C., and G. Eysenbach, 2010: Pandemics in the age of Twitter: Content analysis of tweets during the 2009 H1N1 outbreak. PLOS ONE, 5, e14118, https://doi.org/10.1371/journal.pone.0014118.

    • Search Google Scholar
    • Export Citation
  • Chong, D., and J. N. Druckman, 2007: Framing theory. Annu. Rev. Polit. Sci., 10, 103126, https://doi.org/10.1146/annurev.polisci.10.072805.103054.

    • Search Google Scholar
    • Export Citation
  • Chou, C. H., F. M. Zahedi, and H. Zhao, 2014: Ontology-based evaluation of natural disaster management websites: A multistakeholder perspective. Manage. Inf. Syst. Quart., 38, 9971016, https://doi.org/10.25300/MISQ/2014/38.4.03.

    • Search Google Scholar
    • Export Citation
  • Claeys, A. S., and V. Cauberghe, 2014: What makes crisis response strategies work? The impact of crisis involvement and message framing. J. Bus. Res., 67, 182189, https://doi.org/10.1016/j.jbusres.2012.10.005.

    • Search Google Scholar
    • Export Citation
  • Comstock, R. D., and S. Mallonee, 2005: Comparing reactions to two severe tornadoes in one Oklahoma community. Disasters, 29, 277287, https://doi.org/10.1111/j.0361-3666.2005.00291.x.

    • Search Google Scholar
    • Export Citation
  • Demuth, J. L., R. E. Morss, J. K. Lazo, and D. C. Hilderbrand, 2013: Improving effectiveness of weather risk communication on the NWS point-and-click web page. Wea. Forecasting, 28, 711726, https://doi.org/10.1175/WAF-D-12-00118.1.

    • Search Google Scholar
    • Export Citation
  • Entman, R. M., 2007: Framing bias: Media in the distribution of power. J. Commun., 57, 163173, https://doi.org/10.1111/j.1460-2466.2006.00336.x.

    • Search Google Scholar
    • Export Citation
  • Gladwin, H., J. K. Lazo, B. H. Morrow, W. G. Peacock, and H. E. Willoughby, 2009: Social science research needs for the hurricane forecast and warning system. Bull. Amer. Meteor. Soc., 90, 2530, https://doi.org/10.1175/2008BAMS2606.1.

    • Search Google Scholar
    • Export Citation
  • Grunig, J. D., and T. Hunt, 1984: Managing Public Relations. Holt, Rinehart and Winston, 550 pp.

  • Guan, W., H. Gao, M. Yang, Y. Li, H. Ma, W. Qian, Z. Cao, and X. Yang, 2014: Analyzing user behavior of the micro-blogging website Sina Weibo during hot social events. Physica A, 395, 340351, https://doi.org/10.1016/j.physa.2013.09.059.

    • Search Google Scholar
    • Export Citation
  • Hoekstra, S., K. Klockow, R. Riley, J. Brotzge, H. Brooks, and S. Erickson, 2011: A preliminary look at the social perspective of warn-on-forecast: Preferred tornado warning lead time and the general public’s perceptions of weather risks. Wea. Climate Soc., 3, 128140, https://doi.org/10.1175/2011WCAS1076.1.

    • Search Google Scholar
    • Export Citation
  • Honan, M., 2011: Watch the Virginia earthquake spread across Twitter. Gizmodo, 25 August, http://gizmodo.com/5834048/watch-the-virginia-earthquake-spread-across-twitter.

  • Höppner, C., R. Whittle, M. Bründl, and M. Buchecker, 2012: Linking social capacities and risk communication in Europe: A gap between theory and practice? Nat. Hazards, 64, 17531778, https://doi.org/10.1007/s11069-012-0356-5.

    • Search Google Scholar
    • Export Citation
  • Houston, J. B., B. Pfefferbaum, and C. E. Rosenholtz, 2012: Disaster news: Framing and frame changing in coverage of major U.S. natural disasters, 2000–2010. J. Mass Commun. Quart., 89, 606623, https://doi.org/10.1177/1077699012456022.

    • Search Google Scholar
    • Export Citation
  • Hu, X., X. Wu, P. Yin, and X. Zheng, 2017: An investigation into consumers’ continued social shopping intentions. Proc. Pacific-Asia Conf. on Information Systems, Kedah, Malaysia, Association for Information Systems, 63, https://aisel.aisnet.org/pacis2017/63/.

  • Hyvärinen, O., and E. Saltikoff, 2010: Social media as a source of meteorological observations. Mon. Wea. Rev., 138, 31753184, https://doi.org/10.1175/2010MWR3270.1.

    • Search Google Scholar
    • Export Citation
  • Koenker, R., and G. Bassett Jr., 1982: Robust tests for heteroscedasticity based on regression quantiles. Econometrica, 50, 4361, https://doi.org/10.2307/1912528.

    • Search Google Scholar
    • Export Citation
  • Lee, K. L., R. J. Meyer, and E. T. Bradlow, 2009: Analyzing risk response dynamics on the web: The case of Hurricane Katrina. Risk Anal., 29, 17791792, https://doi.org/10.1111/j.1539-6924.2009.01304.x.

    • Search Google Scholar
    • Export Citation
  • Levin, D. Z., and R. Cross, 2004: The strength of weak ties you can trust: The mediating role of trust in effective knowledge transfer. Manage. Sci., 50, 14771490, https://doi.org/10.1287/mnsc.1030.0136.

    • Search Google Scholar
    • Export Citation
  • Levin, I. P., and G. J. Gaeth, 1988: How consumers are affected by the framing of attribute information before and after consuming the product. J. Consum. Res., 15, 374378, https://doi.org/10.1086/209174.

    • Search Google Scholar
    • Export Citation
  • Li, L., M. F. Goodchild, and B. Xu, 2013: Spatial, temporal, and socioeconomic patterns in the use of Twitter and Flickr. Cartogr. Geogr. Inf. Sci., 40, 6177, https://doi.org/10.1080/15230406.2013.777139.

    • Search Google Scholar
    • Export Citation
  • Lin, X., K. A. Lachlan, and P. R. Spence, 2016: Exploring extreme events on social media: A comparison of user reposting/retweeting behaviors on Twitter and Weibo. Comput. Human Behav., 65, 576581, https://doi.org/10.1016/j.chb.2016.04.032.

    • Search Google Scholar
    • Export Citation
  • Lindell, M. K., and R. W. Perry, 2012: The protective action decision model: Theoretical modifications and additional evidence. Risk Anal., 32, 616632, https://doi.org/10.1111/j.1539-6924.2011.01647.x.

    • Search Google Scholar
    • Export Citation
  • Liu, B. F., J. D. Fraustino, and Y. Jin, 2016: Social media use during disasters: How information form and source influence intended behavioral responses. Commun. Res., 43, 626646, https://doi.org/10.1177/0093650214565917.

    • Search Google Scholar
    • Export Citation
  • Liu, S. H., H. L. Liao, and J. A. Pratt, 2009: Impact of media richness and flow on e-learning technology acceptance. Comput. Educ., 52, 599607, https://doi.org/10.1016/j.compedu.2008.11.002.

    • Search Google Scholar
    • Export Citation
  • Ma, X. J., W. X. Wang, Y. C. Lai, and Z. Zheng, 2010: Information explosion on complex networks and control. Eur. Phys. J., 76B, 179183, https://doi.org/10.1140/epjb/e2010-00208-1.

    • Search Google Scholar
    • Export Citation
  • Meyer, R. J., E. J. Baker, K. F. Broad, J. Czajkowski, and B. Orlove, 2014: The dynamics of hurricane risk perception: Real-time evidence from the 2012 Atlantic hurricane season. Bull. Amer. Meteor. Soc., 95, 13891404, https://doi.org/10.1175/BAMS-D-12-00218.1.

    • Search Google Scholar
    • Export Citation
  • Miao, Q., and D. Popp, 2014: Necessity as the mother of invention: Innovative responses to natural disasters. J. Environ. Econ. Manage., 68, 280295, https://doi.org/10.1016/j.jeem.2014.06.003.

    • Search Google Scholar
    • Export Citation
  • O’Reilly, T., 2007: What is web 2.0: Design patterns and business models for the next generation of software. Commun. Strategies, 65, 1737, https://mpra.ub.uni-muenchen.de/id/eprint/4580.

    • Search Google Scholar
    • Export Citation
  • Ripberger, J. T., 2011: Capturing curiosity: Using Internet search trends to measure public attentiveness. Policy Stud. J., 39, 239259, https://doi.org/10.1111/j.1541-0072.2011.00406.x.

    • Search Google Scholar
    • Export Citation
  • Ripberger, J. T., H. C. Jenkins-Smith, C. L. Silva, D. E. Carlson, and M. Henderson, 2014: Social media and severe weather: Do tweets provide a valid indicator of public attention to severe weather risk communication? Wea. Climate Soc., 6, 520530, https://doi.org/10.1175/WCAS-D-13-00028.1.

    • Search Google Scholar
    • Export Citation
  • Rive, G., J. Hare, J. Thomas, and K. Nankivell, 2012: Social media in an emergency: A best practice guide. Wellington Region CDEM Group Rep., 58 pp., https://www.civildefence.govt.nz/assets/Uploads/CDEM-Resilience-Fund/CDEM-Resilience-Fund-2011-03-social-media-in-an-emergency-guide.pdf.

  • Ros-Diego, V. J., and A. Castelló-Martínez, 2011: CSR communication through online social media. Rev. Lat. Comun. Soc., 67, 4767, https://doi.org/10.4185/RLCS-067-947-047-067.

    • Search Google Scholar
    • Export Citation
  • Sakaki, T., M. Okazaki, and Y. Matsuo, 2010: Earthquake shakes Twitter users: Real-time event detection by social sensors. Proc. 19th Int. Conf. on World Wide Web, Raleigh, NC, International World Wide Web Conference Committee, 851–860.

  • Scheufele, D. A., 1999: Framing as a theory of media effects. J. Commun., 49, 103122, https://doi.org/10.1111/j.1460-2466.1999.tb02784.x.

    • Search Google Scholar
    • Export Citation
  • Scheufele, D. A., and D. Tewksbury, 2007: Framing, agenda setting, and priming: The evolution of three media effects models. J. Commun., 57, 920, https://doi.org/10.1111/j.0021-9916.2007.00326.x.

    • Search Google Scholar
    • Export Citation
  • Schumacher, R. S., D. T. Lindsey, A. B. Schumacher, J. Braun, S. D. Miller, and J. L. Demuth, 2010: Multidisciplinary analysis of an unusual tornado: Meteorology, climatology, and the communication and interpretation of warnings. Wea. Forecasting, 25, 14121429, https://doi.org/10.1175/2010WAF2222396.1.

    • Search Google Scholar
    • Export Citation
  • Semetko, H. A., and P. M. Valkenburg, 2000: Framing European politics: A content analysis of press and television news. J. Commun., 50, 93109, https://doi.org/10.1111/j.1460-2466.2000.tb02843.x.

    • Search Google Scholar
    • Export Citation
  • Simmons, K. M., and D. Sutter, 2008: Tornado warnings, lead times, and tornado casualties: An empirical investigation. Wea. Forecasting, 23, 246258, https://doi.org/10.1175/2007WAF2006027.1.

    • Search Google Scholar
    • Export Citation
  • Sorensen, J. H., 2000: Hazard warning systems: Review of 20 years of progress. Nat. Hazards Rev., 1, 119125, https://doi.org/10.1061/(ASCE)1527-6988(2000)1:2(119).

    • Search Google Scholar
    • Export Citation
  • Stafford, T. F., M. R. Stafford, and L. L. Schkade, 2004: Determining uses and gratifications for the Internet. Decis. Sci., 35, 259288, https://doi.org/10.1111/j.00117315.2004.02524.x.

    • Search Google Scholar
    • Export Citation
  • Stapa, S. H., and A. H. Shaari, 2012: Understanding online communicative language features in social networking environment. GEMA, 12, 817830, http://www.ukm.my/ppbl/Gema/gemacurrentissues.html.

    • Search Google Scholar
    • Export Citation
  • Tang, Z., L. Zhang, F. Xu, and H. Vo, 2015: Examining the role of social media in California’s drought risk management in 2014. Nat. Hazards, 79, 171193, https://doi.org/10.1007/s11069-015-1835-2.

    • Search Google Scholar
    • Export Citation
  • Tumasjan, A., T. O. Sprenger, P. G. Sandner, and I. M. Welpe, 2010: Predicting elections with Twitter: What 140 characters reveal about political sentiment. Proc. Fourth Int. AAAI Conf. on Weblogs and Social Media, Washington, DC, AAAI, 178–185.

  • Tversky, A., and D. Kahneman, 1981: The framing of decisions and the psychology of choice. Science, 211, 453458, https://doi.org/10.1126/science.7455683.

    • Search Google Scholar
    • Export Citation
  • Vieweg, S., A. L. Hughes, K. Starbird, and L. Palen, 2010: Microblogging during two natural hazards events: What Twitter may contribute to situational awareness. Proc. SIGCHI Conf. on Human Factors in Computing Systems, Atlanta, GA, ACM, 10791088.

  • Weber, R. P., 1990: Basic Content Analysis. 2nd ed. Sage Publications, 96 pp.

  • Wei, J., J. Xu, and D. Zhao, 2015: Public engagement with firms on social media in China. J. Inf. Sci., 41, 624639, https://doi.org/10.1177/0165551515586712.

    • Search Google Scholar
    • Export Citation
  • Wei, J., F. Wang, and M. K. Lindell, 2016: The evolution of stakeholders’ perceptions of disaster: A model of information flow. J. Assoc. Inf. Sci. Technol., 67, 441453, https://doi.org/10.1002/asi.23386.

    • Search Google Scholar
    • Export Citation
  • Westerman, D., P. R. Spence, and B. Van Der Heide, 2014: Social media as information source: Recency of updates and credibility of information. J. Comput. Mediat. Commun., 19, 171183, https://doi.org/10.1111/jcc4.12041.

    • Search Google Scholar
    • Export Citation
  • Wogalter, M. S., 2006: Handbook of Warnings. CRC Press, 864 pp.

  • Wong, T. F., and Y. Y. Yan, 2002: Perceptions of severe weather warnings in Hong Kong. Meteor. Appl., 9, 377382, https://doi.org/10.1017/S1350482702003110.

    • Search Google Scholar
    • Export Citation
  • Zhang, F., and Coauthors, 2007: An in-person survey investigating public perceptions of and responses to Hurricane Rita forecasts along the Texas Coast. Wea. Forecasting, 22, 11771190, https://doi.org/10.1175/2007WAF2006118.1.

    • Search Google Scholar
    • Export Citation
All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 0 0 0
PDF Downloads 0 0 0

Public Attention to Natural Hazard Warnings on Social Media in China

View More View Less
  • 1 School of International Economics and Trade, Nanjing University of Finance and Economics, Nanjing, Jiangsu, China
  • 2 School of Management, University of Science and Technology of China, Hefei, Anhui, China
  • 3 State Key Laboratory of Fire Science, and School of Management, University of Science and Technology of China, Hefei, and Center for Crisis Management Research, School of Public Policy and Management, Tsinghua University, Beijing, China
© Get Permissions
Full access

Abstract

Hazard warning is vital in disaster management. The rapid development of social media allows warning producers and receivers to exchange warning messages effectively and sufficiently. This study investigates the factors that influence public attention to natural hazard warning information on social media. Drawing from the protective action decision model and framing theory, this study classifies antecedents into three groups, namely, hazard information, publisher’s/reader’s characteristics, and frame setting. To test the hypotheses empirically, we select Sina Weibo, the leading social network in China, as the research context. From this platform, 3452 warning messages issued by authorities in the target area are collected. We code each message based on its attributes that are related to our study for linear regression analyses. Results show that all the factors related to publisher’s/reader’s characteristics exert significant effects on public attention. However, the affected range indicated by a warning message and the formality of the message’s language are not significantly related to public attention to the message.

© 2018 American Meteorological Society. For information regarding reuse of this content and general copyright information, consult the AMS Copyright Policy (www.ametsoc.org/PUBSReuseLicenses).

Corresponding author: Jiuchang Wei, weijc@ustc.edu.cn

Abstract

Hazard warning is vital in disaster management. The rapid development of social media allows warning producers and receivers to exchange warning messages effectively and sufficiently. This study investigates the factors that influence public attention to natural hazard warning information on social media. Drawing from the protective action decision model and framing theory, this study classifies antecedents into three groups, namely, hazard information, publisher’s/reader’s characteristics, and frame setting. To test the hypotheses empirically, we select Sina Weibo, the leading social network in China, as the research context. From this platform, 3452 warning messages issued by authorities in the target area are collected. We code each message based on its attributes that are related to our study for linear regression analyses. Results show that all the factors related to publisher’s/reader’s characteristics exert significant effects on public attention. However, the affected range indicated by a warning message and the formality of the message’s language are not significantly related to public attention to the message.

© 2018 American Meteorological Society. For information regarding reuse of this content and general copyright information, consult the AMS Copyright Policy (www.ametsoc.org/PUBSReuseLicenses).

Corresponding author: Jiuchang Wei, weijc@ustc.edu.cn

1. Introduction

Natural hazards are overwhelming, despite the advancements in disaster management systems and technologies. According to a report issued by the China National Commission for Disaster Reduction, 243 million people in China were affected by severe weather and natural events in 2014. These events led to 1583 deaths and CNY 337 billion worth of direct economic losses. Seasonal typhoons usually hit coastal areas in China, resulting in extensive damage. An example is Typhoon Rammasun in October 2014, which struck a large area in the southeast, affecting 10 million people and causing at least 46 deaths. Moreover, in July 2012, a disastrous rainstorm in Beijing took 79 lives and affected the entire city. Various systematic risk management mechanisms have been developed by governments and specialists to cope with natural hazards. A critical part of these processes is a warning system that can significantly reduce injuries (Simmons and Sutter 2008). Studies have revealed that people tend to seek weather and warning information constantly from all types of information sources (Zhang et al. 2007; Lee et al. 2009). Traditional methods to disseminate warning messages to the public involve sirens, television, and short message services. For instance, the government can use cell base stations to send mass texts to the public. Despite this method being the most direct and effective for delivering emergency messages to people in a certain area, it can be obstructed by many technical factors and huge costs. Thus, it is usually adopted for only the most destructive and urgent hazards. With the development of media techniques, new platforms have emerged that can efficiently and economically support the exchange of natural and weather information (Schumacher et al. 2010; Chatfield and Brajawidagda 2013).

With the advent of web 2.01 techniques, warning information is now available in social media and online communities, through which authorities can post messages so that a huge number of people can see them directly. Online social networks also allow individuals to spread information with ease. The warning process is one-way communication (Höppner et al. 2012), wherein bidirectional ties between the initial sender and receivers are not necessary when warnings are disseminated. Therefore, governments usually select unidirectional relationship-building social media (such as microblogs) instead of bidirectional relationship-building social media (such as Facebook) as information dissemination channels (Chavez et al. 2010). For instance, Guangdong Province has established a professional account on Sina Weibo (Weibo), the largest microblog platform in China. Meteorological agencies publish warnings and related messages on the platform and share related explanations and knowledge, especially when typhoons approach. However, information overload usually occurs on social media, and it can easily bury useful messages for individual users (Ma et al. 2010). Warnings, as urgent real-time information, must draw public attention immediately when they are issued. Thus, this necessity brings about the question of what factors could influence public attention to online warnings.

Taking the aforementioned facts into account and exploring how the public reacts to warnings on social media are meaningful for both research and practice. Given the significance of this topic and the lack of academic and empirical insights on the issue, this study aims to investigate the direct reactions of social media users to online warnings, the elements within a warning that can influence public attention, and the mechanism of the influence process. Specifically, Sina Weibo is chosen as the current research context. We identify and classify the potential factors into three categories based on the protective action decision model (PADM) and framing theory: hazard attributes, framing factors, and publisher’s/reader’s characteristics. Since Weibo enables users to directly react to a post via two functions (i.e., comment and repost), we measure public attention to each post by the number of comments and reposts. The research is expected to expand the application of framing theory and enrich the literature on social media–based risk information management.

2. Research background

a. Disaster management on social media

As an emerging and evolving medium that facilitates effective communication, social media has become a desirable platform to study public response in many disciplines (Chew and Eysenbach 2010; Sakaki et al. 2010; Tumasjan et al. 2010). Both government officials and scholars have acknowledged its potential in risk and disaster management (Cameron et al. 2012). Scholars have examined the particular affordances of social media and suggested that this platform can support various risk management practices that range from natural hazard detection (Hyvärinen and Saltikoff 2010) and warning dissemination (Vieweg et al. 2010) to instantaneous disaster status updating (Rive et al. 2012) and rescue seeking (Acar and Muraki 2011). The spread of information on social media can be extremely fast, and it can even be faster than the development speed of the hazard (Honan 2011).

In practice, many social media platforms have been applied in disaster management. For example, Twitter and Weibo, two popular social networking sites worldwide, usually support public information dissemination and discussions in crisis events (Lin et al. 2016). However, although disaster management on social media has been studied, little is known about how the public focuses on the warning messages on such platforms. To the best of our knowledge, Ripberger et al.’s (2014) work is one of the few studies that has investigated whether tweets could indicate public attention to severe weather signals. They analyze public attention to messages about tornadoes based on a comparison of daily tornado-related tweet counts and the number of tornado watches and warnings issued. They find that tweet counts are systematically related to the communication of tornado watches and warnings. Thus, social media data can be used to examine the relationship among risk communication, attention, and public reactions to severe weather. Moreover, in comparison with traditional methods such as field surveys, interpreting public attention based on online behaviors can be more accurate and in real time (Ripberger 2011). On this foundation, we conduct our research on public attention to warning messages on Weibo. Different from Ripberger’s (2011) work, we first collect warnings on Weibo following a prespecified screening criterion. Specifically, to improve the generalizability of our research, instead of focusing on one particular natural hazard event, all types of natural hazards’ warnings that meet our prespecified criterion are considered. This approach makes it difficult to measure how people react to each type of hazard by counting the general number of posts on Weibo that contain the focal keyword. Instead, we analyze public attention to each warning individually by counting the number of comments on and reposts of each warning because these are closely related to the warning message. We likewise analyze the factors in the warnings to examine their effects on public attention.

b. Theoretical background

When a warning is issued, paying attention is the first step of a receiver’s information-handling process (Wogalter 2006). To understand public attention, we draw from the classic PADM, which has been extensively applied in risk management research. The model suggests that individuals experience three sequential phases before a protective decision is made, namely, warning exposure, attention, and comprehension (Lindell and Perry 2012). Public attention to emergencies is affected by two aspects: environmental/social context and hazard characteristics (Lindell and Perry 2012). Different hazards raise different responses; for example, people usually pay more attention to severe and more perceptible hazards, specifically, the hazards that they can see, hear, or experience in person (Wei et al. 2016). The social context involves the source of the warning message and the receiver’s properties, both of which affect the receiver’s attention. People have more trust in authentic and familiar sources and thus show a higher level of attention to warning messages from them. Furthermore, the extent of attention can be influenced by individual features; for instance, educational level can affect the attention through the interpretation of warning content.

In addition to the abovementioned factors, the different ways by which publishers write and issue warnings can also lead to different levels of public attention. To include this factor in our study, framing theory is used to investigate how the framing of warnings influences public attention to the message.

Framing theory, a branch of communication theory, has been extensively applied in many fields, including news and political events (Levin and Gaeth 1988; Semetko and Valkenburg 2000; Entman 2007). Framing relates to the subjective transmission process of a piece of information. Research posits that framing can significantly shape an audience’s response to a message (Chong and Druckman 2007). Scheufele (1999) concluded that framing involves two aspects: media frames and audience frames. Media frames refer to how news events are identified and presented. Audience frames refer to ideas and routines that guide individuals to interpret the messages. A media frame is formed by two processes: frame building and frame setting. Frame building denotes how a media frame is determined by the styles and preferences of the journalists and editors. Frame setting means that the publishers selectively post or highlight the content to achieve a certain effect. The most classic example of the framing effect is the presentation of two similar messages that emphasize different aspects of an event, which can lead to the “half empty or half full” perception (Tversky and Kahneman 1981). Media can manipulate the wording, tone, focal point, and coverage to affect a target receiver’s manner of thinking and thus affect his/her decision-making process (Scheufele and Tewksbury 2007). Framing theory is frequently used in communication research, but it is seldom employed in studies on natural hazard warning management.

According to framing theory, the means by which a warning message is organized can affect readers’ perceptions and responses to it. Moreover, as suggested by the PADM, the warning source, content, formation, and receivers’ attributes are key antecedents of the public’s attention to warning messages on social media. Thus, we classify the influencing factors into three categories: hazard attributes (semantic content), media framing (syntactic content), and characteristics of publisher and receiver.

3. Predictors of public attention and hypothesis development

a. Hazard attributes

1) Hazard type

Previous studies suggested that people hold different risk perceptions and perform different protective behaviors when dealing with different types of natural hazards (Sorensen 2000; Chou et al. 2014; Miao and Popp 2014; Tang et al. 2015). The most widely accepted classification method of natural hazards is based on natural phenomena, such as the cause of formation and the injured party (Berren et al. 1980). However, these classifications do not reflect the social reactions associated with the hazard. Normally, the public is not familiar with scientific terminologies. When they perceive natural hazards, they rely heavily on their personal experiences and intuitions. For example, numerous studies indicate that people in different districts respond differently to the same hazard, and the responses are associated with the specific hazard history and experience in the focal area (Lee et al. 2009). Therefore, we propose that occurrence frequency, which indicates how often a specific type of hazard occurs in the focal area, is a critical measure of the hazard type when investigating public reactions. Hence, we suggest the following:

Public attention to a hazard warning on social media is significantly and positively associated with the occurrence frequency of the hazard in this warning message.

2) Level of severity

Severity or rate is another important aspect of a hazard. It refers to the level of the predicted damage that the public may experience and the level of emergency of this damage. A rational person cares about potential property damage and bodily injury and is naturally more concerned about hazard warnings when the weather situation becomes tough. In addition, people become highly sensitive to warning information when events become urgent and they experience danger in person. A positive and significant relationship between hazard severity and response has been verified by many researchers (Bubeck et al. 2012; Miao and Popp 2014; Sorensen 2000). Thus, we posit the following:

Public attention to a hazard warning on social media is significantly and positively associated with the level of severity of the hazard in this warning message.

3) Affected range

Affected range of a warning message refers to the estimated area that the hazard will impact. The range can be a specific district (e.g., city and province) or a general area (e.g., southeast). The area that may possibly be affected by a natural hazard is detected and computed by professional institutions on the basis of real-time data and analytical methods. In general, a large area is more likely to contain a high number of residents. Therefore, the affected range is positively related to the affected population. A warning that refers to a larger range likewise indicates that the hazard may lead to consequences that are more destructive. Previous studies suggested that the public has different responses to hazard messages that aim at different ranges (e.g., local or national) on an information platform (Brotzge and Donner 2013; Liu et al. 2016). Thus, we initially propose the following:

Public attention to a hazard warning on social media is significantly and positively associated with the affected range of the hazard in this warning message.

b. Frame setting

1) Timing

Scholars have noticed public responses to social media posts are different during different days and hours (Li et al. 2013). Similarly, people exhibit varying responses to the same warning message at different times (Lee et al. 2009; Carley et al. 2016; Sorensen 2000). For example, a person may pay significant attention to a meteorological message when planning to go outdoors, which is more possible on rest days than on working days. Moreover, the sleep–wake cycle affects one’s response to warnings. An individual’s timetable of using social media is also a critical antecedent of his/her attention toward the warning messages. Therefore, information publishers usually pay attention to the timing of releasing information on social media. Similar to the so-called “golden time” of television or radio broadcasts, social media also has periods when the audience exerts optimum attention. Public attention is different during different days and different hours. Thus, we propose the following hypothesis:

Public attention to a hazard warning on social media is significantly higher if the warning message is issued during the working day or during television and radio golden time periods (0600–0830 and 1830–2100 LST; LST = UTC + 8) than during other hours.

2) Syntactic content

A hazard warning is usually posted and authorized by official organizations. To maintain credibility and accuracy, hazard warning messages are usually written in formal and professional styles. However, some studies reveal that casual or visual messages can capture intensive attention (Ash et al. 2014). The language of Internet-based media differs from that of traditional ones in richness and linguistic style (Stapa and Shaari 2012). In each medium, the audience has a specifically preferred manner of information presentation, which affects their acceptance of the message. Social media users are diversified, and technology allows them to express themselves in many styles. In addition to language and tone, the rich presentation methods offered by social media (e.g., pictures and videos) also have a potential effect on public response. Brantner et al.’s (2011) study on visual framing effects suggests that the image in an article can significantly influence a reader’s perception of the whole story. According to media richness theory, the richness of a message is positively associated with the attention and perception of a participant (Liu et al. 2009). Prior studies have shown that for microblogs and other online platforms, a richer message can elicit more attention (Guan et al. 2014; Chou et al. 2014; Demuth et al. 2013). However, in terms of warning messages, richer messages do not always attract more attention. Casteel and Downing (2013) indicate that people show the same level of attention to warning texts that the National Weather Service sends with or without pictures. Given the significance of information richness in public attention and the lack of a unanimous opinion on this issue, the present study further explores the issue and proposes the following hypothesis:

Public attention to a hazard warning on social media is significantly higher if the warning message is worded in formal language and is rich in presentation methods (e.g., pictures and videos).

c. Characteristics of publisher and receiver

1) Activeness of the publisher

Information source is an important aspect of the warning, and it contributes to individuals’ perceptions of the content. The style and feature consistency of a sender can significantly affect receivers’ perceptions when the sender publishes a warning (Stafford et al. 2004). On social media, activeness is a critical trait of a publisher that significantly affects its interaction with receivers. Active updates could attract and retain constant public attention, increase audience engagement, and enhance the credibility of information (Wei et al. 2015; Westerman et al. 2014). Thus, we propose the following hypothesis:

Public attention to a hazard warning on social media is significantly and positively associated with the activeness of the publisher.

2) Region of receivers

Warnings are usually issued to recipients in a certain area, but the specific attributes of each receiver are not considered. Thus, the general properties of residents within a particular area are highly important. Many district properties have been previously investigated, and most studies have focused on location (e.g., the distance of the district from the hurricane center) and hazard history (e.g., whether the district has previously experienced a similar hazard; Comstock and Mallonee 2005; Lee et al. 2009). The conclusions are often inconsistent. Given that we do not focus on one typical hazard but on general natural hazards, the aforementioned factors are not suitable. Research on the demographics of receivers has revealed that income level, educational level, technological level, and other relevant factors significantly affect people’s response to online warnings (Wong and Yan 2002; Meyer et al. 2014). If these individual factors are attributed to a general regional property, then the economic situation can be approximated. Therefore, the economic development level of the recipients’ area is considered in this study.

Public attention to a hazard warning on social media is significantly and positively associated with the regional economic development level of the receivers.

To summarize the above discussions, the research framework is depicted in Fig. 1.

Fig. 1.
Fig. 1.

Research model.

Citation: Weather, Climate, and Society 11, 1; 10.1175/WCAS-D-17-0039.1

4. Methodology

a. Dataset

In this study, we aim to understand public attention to social media by analyzing users’ behaviors. In particular, Weibo, one of the leading social media platforms in China, is adopted as the research context.

Sina Weibo is a leading social media platform that supports information sharing, dissemination, and acquisition based on the personal online social network ties of its users (Hu et al. 2017). It can be accessed via websites and mobile applications. It was established in August 2009, and by 2015, Weibo had 222 million active users per month and 100 million active users per day. Weibo has a 43.6% penetration rate in China.2 Weibo users focus on topics such as entertainment, news, and social events. A number of well-known social events were originated and amplified by Weibo, which revealed its power in information production and diffusion. Weibo has become a useful tool for users to search for information and comment on public affairs. It has changed Chinese people’s daily lives. Similar to Twitter, because of the large number of celebrity users and famous bloggers and the platform’s unidirectional relationship-building mechanism, Weibo tends to be more of an information channel than a relationship-building platform (Guan et al. 2014). It has a 140-character limit for each message, and links, pictures, videos, or long texts can be attached to the message. Similar to Twitter, Weibo has two functions that enable interactivity among users: comment and repost. The “comment” mechanism allows users to write opinions about a post, and the comments are shown under the original post. A user can further reply to a comment directly under a post, which can create a discussion. “Repost” allows a user to display another post as his/her own post with reference to the original author.

Given Weibo’s high information diffusion efficiency, it has become the most popular choice for authorities to issue official declarations during social events, especially when the focal event is initially developed on Weibo. Most government departments in China have been required to create a Weibo account, and Weibo has even offered a special government version for them. By 2015, nearly 13 million verified official accounts, including weather-related accounts, had been established on Weibo, offering abundant information every day. For instance, if the keyword “meteorology” is entered to search for a verified organization, 1719 results can be obtained, most of which are from meteorological administrations. Following common rules, China has established its own mechanisms and standards for the warning process. A warning usually contains the hazard name, affected area, predicted occurrence time, severity, signal, and other necessary elements. Given its message length allowance, Weibo is considered a proper medium for issuing warning messages. Figure 2 shows a screenshot of a warning post on Weibo.

Fig. 2.
Fig. 2.

Screenshot of a Weibo hazard warning post.

Citation: Weather, Climate, and Society 11, 1; 10.1175/WCAS-D-17-0039.1

Data collection

The first sampling procedure was to select several provinces for the data collection boundary. Given that economic regional development level is an antecedent, we used the GDP per capita rank of 2014 at the provincial level in China to choose target provinces. The interval sampling method was initially used. If a selected province did not conduct warnings actively on Weibo, it was eliminated, and a nearby province in the GDP rank would be chosen instead. Afterward, following the same method, two to three cities in each province were selected by interval sampling based on the development level. As a result, four municipalities, 11 provinces, and 31 cities (selected from the 11 provinces) were chosen in this study. Based on these regions, we selected 47 Weibo official accounts established by meteorological bureaus and administrations, including the National Meteorological Center, as the warning data sources. Authorized bloggers usually post warnings within their administrative regions. For example, a national blogger may issue hazard information that develops at the province or city level. Accounts at the provincial level can issue warnings that affect their administrative coverage, including provincial and lower-region levels. Thus, numerous messages at all district levels can be acquired. Given that China has complex geographical and meteorological conditions that give rise to various natural hazards at different times, two periods that could cover most hazard types were selected: July–September 2014 and January–March 2015. With “warning” used as a keyword, warning messages were searched in each account’s posting history. Thus, 3452 records were gathered, which included 30 types of natural hazards.

b. Data characterization and measurement

1) Predictor variables

The aforementioned predictors were all detected from each record. Each type of hazard in each district was counted to measure the hazard occurrence frequency. In the Chinese warning system, two measurements are commonly used to indicate the severity of a hazard. The coloring classification is the most applied method. White, blue, yellow, orange, and red stand for severity rates (potential damage and arrival time) in increasing order. For example, a red storm warning means that an urgent and dangerous storm at the highest severity level is approaching. The other scale uses Roman numerals, in which a small rank indicates a high damage risk (IV, III, II, I). This method can be replaced by color indicators. We used these two scales to measure the severity of hazards in the warnings from 1 to 5, with 5 being the highest severity. To measure the affected range, the areas were divided based on the administrative divisions, ranging from town level to national level (evaluated as 1–6).3

Weibo supports diverse methods of presenting posts. A warning usually consists of text and pictures; links and videos are less frequently used. Two variables are used to measure syntactic content, namely, media richness and formality of language. If a warning message is text only, the media richness is 0. Messages that contain pictures or video are rated 1. If the warning language is formal, it is marked as 0 (otherwise, 1). Regarding the warning day, the Chinese national legal holiday standard was adopted to code work days and holidays, namely, 0 for working days and 1 for holidays and weekends. The television and radio time standard of China was used to determine whether a warning is issued in the golden time period: “yes” was input for dummy 1 (0600–0830 and 1830–2100 LST), and “no” was input for 0.

Weibo updates were computed per day during target periods to derive the value of publishers’ activity levels. An important activity statistic is the variance of daily updates. Meteorological information pertains to real-time messages, which means that nearly all accounts publish weather-related information frequently without much variance. Thus, indicating the updates per day is a sufficient activity proxy. The economic level of receivers’ regions was measured by GDP per capita of 2014, and the data were accessed from yearly economic statistics reports.

2) Outcome variables

Public attention to each post was measured by the number of comments and reposts. Survey answers often involve postprocess and artificial factors. However, online secondhand data can provide insights into the readers’ actual responses induced by their attention, thereby ensuring accurate evaluation.

Both the comments and reposts of a post can reflect the readers’ attention to it, though there are differences between the two. On Weibo, users can comment on a post, and these comments will be listed below the post. Furthermore, users can comment on others’ comments and form a dialogue. The comments a user produces are not traceable in his/her personal homepage. However, a repost means one collects the posts on his/her personal homepage, and these posts are visible to visitors. Thus, reposting behaviors are somehow affected by one’s concern for his/her self-image. The comments a user produces show his/her emotions, such as worry or disappointment; questions about the message; and suspicions about the correctness of the warning. After reading a post, users usually browse its comments to obtain additional information and write comments to exchange opinions with other users. Therefore, the number of comments and reposts of a post can indicate how the public pays attention to it. Given that public attention is an aggregate variable (Grunig and Hunt 1984; Wei et al. 2015), we measure public attention by counting the total number of reposts and comments.

3) Control variables

Considering the variety of population and the Internet penetration within each district, control was exercised over permanent residents and the network penetration rate of warning districts. The former was acquired from the 2013 Statistical Yearbook of selected districts, and the latter was obtained from the statistics of the 35th Report of China Internet Development by the China Internet Network Information Center. We controlled for the followers of each account as well. Finally, warning messages are usually updated to retain public attention. Thus, public attention may weaken regarding a repeated/updated warning. We controlled this effect by using 0 to code an initial warning message and 1 to code a repeated/updated warning message.

4) Data coding

Two of the coauthors of the current research read and coded full-text versions of all warning messages separately. For each message, they recorded the variables of level of severity, affected range, with pictures, and informal language. Most of the warning messages clearly included the above information, and the two coders agreed 87% of the time, which suggested reasonable intercoder reliability (Weber 1990). Disagreements were resolved by discussions. The final coding was verified by a colleague who coded the 480 warning messages from 2014 (with 84% agreement).

c. Analysis techniques

Linear regressions were conducted using Statistical Package for the Social Sciences (version 18). Multicollinearity was checked, and no evidence was found. The variance inflation factors for the study’s independent variables were less than 1, which was far below the standard cutoff of 10. The sum of reposts and comments was primarily investigated. In consideration of the heteroscedasticity issue, referring to the circumstance in which the variability of a variable is unequal across the range of values of a second variable that predicts it (Koenker and Bassett 1982), all regressions proceeded with a logarithmically transformed item of each continuous variable, given that some warnings contained no comments, that is, public attention for comment = log (1 + comments) for regression (Levin and Cross 2004). The same is applied to reposts and the sum. Given that reposts and comments are at different levels of attention, the two outcomes were also studied separately, as shown in Table 1.

Table 1.

Linear regression results II.a

Table 1.

5. Results

a. Descriptive report

Table 2 presents the descriptive report of all variables. The mean number of followers of each account is over 0.2 million, revealing that the official meteorological blogger elicits high attention from users who could potentially produce a significant influence on the platform. The accounts publish an average of seven posts daily, including weather forecasts, meteorological knowledge, warnings, and news. Most of the warnings are about emergent hazards. A few are related to slow hazards, such as drought. In total, 92% of the warnings are only effective within 12 h, and each warning is updated fewer than five times. Therefore, the current message flow is assumed to be sufficient for providing the required real-time information and revealing the differences among accounts. As for the warning range pattern, most warnings target the city level (over 50%). The average sum of comments and reposts of each Weibo warning message is 12.93, indicating that warnings elicit moderate attention. Similar to users of Twitter and other social media platforms, Weibo users prefer to browse messages rather than create content. In effect, many people open an account only to read news or comments from others. Although the number of Weibo users is huge, the numbers of reposts and comments are relatively low. Most warning messages have a formal tone plus graphical symbols or icons of disaster. Over 90% of the messages are written in formal language and are posted with pictures (which often show warning signals), indicating that authorities tend to adopt a traditional manner when issuing warnings on social media.

Table 2.

Means, standard deviations, and correlations.a

Table 2.

b. Regression and further analysis

We first input control variables in regression Eq. (1) (see Table 3 for equation results), in which the population of receivers’ district, Internet penetration of receivers’ district, number of followers of the publisher, and initial/updated message are regressed on public attention. Then, hazard characteristics were input in regression Eq. (2), in which hazard type, level of severity, and affected range are regressed in public attention. As shown in Table 3, level of severity is positively related to public attention, indicating that people are significantly more sensitive to more severe warnings (β = 0.13, p < 0.01). However, the affected range of the warning is not significantly related to public attention, which contradicts our proposal. Furthermore, public attention to different levels of warnings in the affected range is compared. Warnings concerning a moderate level of affected range (city level) attract the most public attention (on average, a warning of city-level hazards has 16 comments/retweets). To investigate whether a blogger’s administrative level affects its public attention, we conducted another regression analysis on the followers of each account (controlling for network penetration rate and residents) and found that meteorological accounts at a lower level of management attract more followers. This result reveals that weather messages posted within a local or smaller district gain relatively more attention. As predicted, hazard history and frequency of occurrence significantly affect attention (β = −0.12, p < 0.01), and the negative coefficient demonstrates that people exert higher concerns about natural events that they experience less frequently. Different hazards may also have varying degrees of effect on public attention. For instance, rainstorms are usually random, which hardly allows for preparation. Thus, people may pay more attention to rainstorm notices, compared to warnings of other hazards that are easy to anticipate. The analysis of variance method was used to investigate these unequal effects. The results revealed significant variance among the effects of different hazards’ warnings on public attention (p < 0.01). In other words, the public exhibits significantly varying levels of attention toward warnings of different hazards.

Table 3.

Linear regression results I.a

Table 3.

Third, frame-setting factors were input in regression Eq. (3), in which warning days, warning hour, and syntactic content were regressed on public attention. Both the warning day and hour are significant (p < 0.01). The negative coefficient of warning day is −0.16, thereby indicating that people pay more attention to warning information during work days on social media. The effect of the warning’s issuing hour is consistent with our expectations (β = 0.12, p < 0.01), offering evidence that the public pays more attention to social media within the traditional media golden time. Furthermore, we examined warning publishing and public attention at hours throughout the day to obtain details of the temporal effect. Figure 3 illustrates the dynamic of warning issuing and public attention during daytime on Weibo on the basis of the collected sample (public attention is measured by the total number of comments or reposts). To emphasize the distinction of message sending and attention, we used an average sum of reposts and comments for each message as the public attention score at the hour point, which is shown on the solid line; the dotted line shows warning message issuing dynamics. The attention curve indicates two peak periods in 1 day, at 0500–0600 and 2100–2300 LST. In the daytime, public attention becomes moderate. However, warning messages issued on the platform are significantly inconsistent with public attention pattern. The public pays little attention to related information at 1600–1700 LST, but many warning messages are posted then. Though public response reveals increasing attention at night, fewer messages are provided then. The incongruence reveals that warning publishers have not used social media tools to their full potential yet.

Fig. 3.
Fig. 3.

The dynamics of warning issuing and public attention in daytime.

Citation: Weather, Climate, and Society 11, 1; 10.1175/WCAS-D-17-0039.1

Regarding syntactic factors, our results show the significant effect of media richness. Messages with text plus pictures and other attachments elicit significantly more attention (β = 0.32, p < 0.01). However, formality of language does not significantly affect public attention.

Next, we input two variables, degree of publisher’s activity and economic level of receiver’s region, in regression Eq. (4). Both are found to be positively and significantly related to public attention as predicted (β = 0.13, 0.09, p < 0.01, p < 0.1). This result provides evidence that on social media, the more active the publisher, the more attention it can elicit from the public. Specifically, the activeness of a publisher can somehow reflect its reliability and credibility in providing accurate, timely, and consistent information. Moreover, people who live in an economically developed area pay more attention to warnings on social media. The possible reason for this finding is that they have more access to social media warnings because of the higher penetration of smartphones, availability of convenient Internet connection, and habit of using Weibo.

Similar results were obtained in regressions for reposts and comments as two dimensions of public attention, except for a subtle difference. At the comment level, warnings issued during the traditional golden time do not indicate a striking difference, compared with warnings issued at normal times. Furthermore, informal language positively affects user comments. At the repost level, GDP per capita of receiver area does not significantly affect reposting behavior. Given that we mainly focused on the summary of the two indexes, additional details were not explored.

6. Discussion and conclusions

This study investigated how public attention is paid to meteorological and natural hazard warning messages on social media and how each element contained in a warning message affects public attention to it. Warning messages are usually short and prompt and require accurate description and high dissemination speed. Therefore, social media is an appropriate channel for warning dissemination and amplification. The factors related to a warning were divided into three groups, namely, hazard property, frame setting, and the characteristics of publishers and receivers.

Consistent with previous studies, we found that people pay more attention to more urgent and severe events on social media and exhibit different levels of concern for different hazards. This work contributes to the current literature by revealing the different reactions of social media users to different types of hazards. Specifically, we used hazard occurrence frequency as one indicator of natural hazard type and revealed that people are concerned more about infrequent hazards, such as typhoons in noncoastal areas. According to personal hazard experience, if one type of hazard occurs frequently, people gain sophisticated evaluation and protection knowledge about the hazard. By contrast, when facing uncommon or infrequent hazards, people rely more on warning bulletins and relevant messages because they do not have much experience. Therefore, the public will pay more attention to warnings of infrequent hazards. Our findings also showed that people are not sensitive to affected ranges within a warning. However, among the many publishers, a city-level publisher can attract the most attention. This result indicates that warnings issued by a medial-range platform (the city-level platform in this study) are the most influential. Given that the area affected by a hazard is predicted by data and technological rules and cannot be modified subjectively, the warning issuing platform can be adjusted based on practical needs, and then the coverage of the warning message can be determined. Governments could issue more warning messages through a city-level station or medium immediately once it acquires messages, and specific cities should be indicated when the threatened area is beyond the city level.

The framing features and social elements of a warning are usually the focus of warning research (Gladwin et al. 2009), and these properties can be complicated in terms of timing and media. The results show that people pay more attention to social media warning messages during work days than during weekends and holidays, possibly because they turn to the Internet for weather information (especially following the development of smartphones) when they do not have sufficient time for TV or other traditional media on work days. Meanwhile, people staying at home are more likely to use traditional media to obtain information. We likewise used the TV and radio golden time to test the influence of warning hour on public attention to social media, and the results show the same timing effect as that in traditional media. Specifically, in traditional media, attention is highest when people wake up and prepare for rest. Social media indicates a unique public attention peak. Our sample suggests that Weibo users are most active late at night and in the early morning. Therefore, for those hazards that will happen the next day, such as rainstorms, or those that will happen in the near future, such as droughts, warning publishers can take advantage of the timing to post warnings during peak hours.

Syntactic content is another important framing element of a warning. The Internet’s affordability and media richness differentiate it from traditional media. In the social media environment, people can respond casually or informally to officially worded statements. We considered this factor and investigated how the public responds to these warning messages in different framing styles. A significant media richness effect was found: warning messages with pictures and other attachments elicit attention, suggesting that publishers should fully use social media richness to articulate warning messages that attract users. Formality of language was found to have little impact on public attention. Controversy was observed regarding the tone and style that government publishers should adopt to issue messages. Some commenters insist on using formal and official language, whereas others insist that official publishers should adjust the manner of speaking to adapt to the targeted readers. However, we found that the language of a natural hazard warning does not affect its public attention. That is, people care more about the content of messages, such as self-relatedness, than its form (Bonsón et al. 2015; Claeys and Cauberghe 2014). For years, people relied on weather prediction to plan their activities and paid constant attention to weather information. Natural hazard warnings are produced by authorities, and their usefulness, accuracy, and timeliness have been the major concern instead of the writing style. Thus, for a warning message, semantic content is more important than formal or informal language.

Social media enables complexity and variety in a piece of information, and interaction is of great value in message delivery (Ros-Diego and Castelló-Martínez 2011). Two-way communication is one dominant feature of social media tools, and it is based on constant focus from people in subtle relationships. To build a tight and dual relationship with receivers, warning publishers are supposed to maintain a certain activity level; this assumption is supported in this study. People prefer to focus on active publishers that provide the latest information and interaction in real time. Regarding the receiver, many studies have investigated how personal factors influence risk perception, warning awareness, and protection behaviors. The present study considered the economic situation at the district group level and found that warnings in developed areas elicit high public attention. The economic development level can influence public attention to warnings on social media for certain reasons. The first may be advanced technical support for convenient social media participation, including hardware and rapid Internet access. Second, financial situation is positively related to education level, which also influences an individual’s ability and habit of using social media. These factors can make a person prefer social media over other media to retrieve warning information. This finding suggests that an efficient social media warning system in a region can be facilitated by technical infrastructure construction and economic prosperity.

In addition to the abovementioned practical implications, this study also contributes to literature and theory. Previous studies that use framing theory to examine hazards usually focused on the information flow, such as number of news reports and dynamics published by media or government (Houston et al. 2012), while ignoring the content of each message. Given that content matters considerably in terms of attracting audience attention and perception, this study adds to the applications of framing theory in disaster management. Moreover, the PADM proposes “warning messages” as an important predictor of public attention. We detailed it by analyzing the message content along two dimensions: hazard information and framing of the content. We combined these two theories in this manner to reveal the underlying mechanism by which the “warning messages” themselves affect public attention and thus contribute to both theories.

This study also enriches assumptions and evidence of disaster management on social media, which are of high theoretical value. Furthermore, it examines the characteristics of user behaviors regarding using social media tools to conduct crisis communication. Thus, our findings can be interpreted to shed light on crisis communication research. Finally, we contribute to e-government research by revealing how the government communicates with the public via social media, especially in emergencies. Given that warning is one of the most important tasks in governments’ natural hazard management, this study has significant theoretical value in this regard.

7. Limitations and further research

This study may include certain limitations. The sample messages collected were all from official accounts, which were merely a small portion of the total warning information on Weibo. Thus, future research can expand the data sources and warning-related messages published by mass media, and ordinary users can also be included. In addition, in different areas, the extent of governments’ social media usage varies, which may cause biases in the research. An in-depth survey of government social media use in different areas is recommended when choosing samples in future research.

Acknowledgments

This research was funded by the National Key R&D Program of China (2016YFC0802500) and the National Natural Science Foundation of China (71522013 and 71702180).

REFERENCES

  • Acar, A., and Y. Muraki, 2011: Twitter for crisis communication: Lessons learned from Japan’s tsunami disaster. Int. J. Web Based Commun., 7, 392402, https://doi.org/10.1504/IJWBC.2011.041206.

    • Search Google Scholar
    • Export Citation
  • Ash, K. D., R. L. Schumann, and G. C. Bowser, 2014: Tornado warning trade-offs: Evaluating choices for visually communicating risk. Wea. Climate Soc., 6, 104118, https://doi.org/10.1175/WCAS-D-13-00021.1.

    • Search Google Scholar
    • Export Citation
  • Berren, M. R., A. Beigel, and S. Ghertner, 1980: A typology for the classification of disasters. Commun. Mental Health J., 16, 103111, https://doi.org/10.1007/BF00778582.

    • Search Google Scholar
    • Export Citation
  • Bonsón, E., S. Royo, and M. Ratkai, 2015: Citizens’ engagement on local governments’ Facebook sites. An empirical analysis: The impact of different media and content types in western Europe. Gov. Inf. Quart., 32, 5262, https://doi.org/10.1016/j.giq.2014.11.001.

    • Search Google Scholar
    • Export Citation
  • Brantner, C., K. Lobinger, and I. Wetzstein, 2011: Effects of visual framing on emotional responses and evaluations of news stories about the Gaza Conflict 2009. J. Mass Commun. Quart., 88, 523540, https://doi.org/10.1177/107769901108800304.

    • Search Google Scholar
    • Export Citation
  • Brotzge, J., and W. Donner, 2013: The tornado warning process: A review of current research, challenges, and opportunities. Bull. Amer. Meteor. Soc., 94, 17151733, https://doi.org/10.1175/BAMS-D-12-00147.1.

    • Search Google Scholar
    • Export Citation
  • Bubeck, P., W. J. Botzen, and J. C. Aerts, 2012: A review of risk perceptions and other factors that influence flood mitigation behavior. Risk Anal., 32, 14811495, https://doi.org/10.1111/j.1539-6924.2011.01783.x.

    • Search Google Scholar
    • Export Citation
  • Cameron, M. A., R. Power, B. Robinson, and J. Yin, 2012: Emergency situation awareness from Twitter for crisis management. Proc. 21st Int. Conf. on World Wide Web, Lyon, France, International World Wide Web Conference Committee, 695–698, https://doi.org/10.1145/2187980.2188183.

  • Carley, K. M., M. Malik, P. M. Landwehr, J. Pfeffer, and M. Kowalchuck, 2016: Crowd sourcing disaster management: The complex nature of Twitter usage in Padang Indonesia. Saf. Sci., 90, 4861, https://doi.org/10.1016/j.ssci.2016.04.002.

    • Search Google Scholar
    • Export Citation
  • Casteel, M. A., and J. R. Downing, 2013: How individuals process NWS weather warning messages on their cell phones. Wea. Climate Soc., 5, 254265, https://doi.org/10.1175/WCAS-D-12-00031.1.

    • Search Google Scholar
    • Export Citation
  • Chatfield, A. T., and U. Brajawidagda, 2013: Twitter early tsunami warning system: A case study in Indonesia’s natural disaster management. Proc. 46th Hawaii Int. Conf. on System Sciences, Wailea, HI, IEEE, 2050–2060, https://doi.org/10.1109/HICSS.2013.579.

  • Chavez, C., M. A. Repas, and T. Stefaniak, 2010: A new way to communicate with residents: Local government use of social media to prepare for emergencies. ICMA Rep., 16 pp., http://www.unapcict.org/ecohub/a-new-way-to-communicate-with-residents-local-government-use-of-social-media-to-prepare-for-emergencies/at_download/attachment1.

  • Chew, C., and G. Eysenbach, 2010: Pandemics in the age of Twitter: Content analysis of tweets during the 2009 H1N1 outbreak. PLOS ONE, 5, e14118, https://doi.org/10.1371/journal.pone.0014118.

    • Search Google Scholar
    • Export Citation
  • Chong, D., and J. N. Druckman, 2007: Framing theory. Annu. Rev. Polit. Sci., 10, 103126, https://doi.org/10.1146/annurev.polisci.10.072805.103054.

    • Search Google Scholar
    • Export Citation
  • Chou, C. H., F. M. Zahedi, and H. Zhao, 2014: Ontology-based evaluation of natural disaster management websites: A multistakeholder perspective. Manage. Inf. Syst. Quart., 38, 9971016, https://doi.org/10.25300/MISQ/2014/38.4.03.

    • Search Google Scholar
    • Export Citation
  • Claeys, A. S., and V. Cauberghe, 2014: What makes crisis response strategies work? The impact of crisis involvement and message framing. J. Bus. Res., 67, 182189, https://doi.org/10.1016/j.jbusres.2012.10.005.

    • Search Google Scholar
    • Export Citation
  • Comstock, R. D., and S. Mallonee, 2005: Comparing reactions to two severe tornadoes in one Oklahoma community. Disasters, 29, 277287, https://doi.org/10.1111/j.0361-3666.2005.00291.x.

    • Search Google Scholar
    • Export Citation
  • Demuth, J. L., R. E. Morss, J. K. Lazo, and D. C. Hilderbrand, 2013: Improving effectiveness of weather risk communication on the NWS point-and-click web page. Wea. Forecasting, 28, 711726, https://doi.org/10.1175/WAF-D-12-00118.1.

    • Search Google Scholar
    • Export Citation
  • Entman, R. M., 2007: Framing bias: Media in the distribution of power. J. Commun., 57, 163173, https://doi.org/10.1111/j.1460-2466.2006.00336.x.

    • Search Google Scholar
    • Export Citation
  • Gladwin, H., J. K. Lazo, B. H. Morrow, W. G. Peacock, and H. E. Willoughby, 2009: Social science research needs for the hurricane forecast and warning system. Bull. Amer. Meteor. Soc., 90, 2530, https://doi.org/10.1175/2008BAMS2606.1.

    • Search Google Scholar
    • Export Citation
  • Grunig, J. D., and T. Hunt, 1984: Managing Public Relations. Holt, Rinehart and Winston, 550 pp.

  • Guan, W., H. Gao, M. Yang, Y. Li, H. Ma, W. Qian, Z. Cao, and X. Yang, 2014: Analyzing user behavior of the micro-blogging website Sina Weibo during hot social events. Physica A, 395, 340351, https://doi.org/10.1016/j.physa.2013.09.059.

    • Search Google Scholar
    • Export Citation
  • Hoekstra, S., K. Klockow, R. Riley, J. Brotzge, H. Brooks, and S. Erickson, 2011: A preliminary look at the social perspective of warn-on-forecast: Preferred tornado warning lead time and the general public’s perceptions of weather risks. Wea. Climate Soc., 3, 128140, https://doi.org/10.1175/2011WCAS1076.1.

    • Search Google Scholar
    • Export Citation
  • Honan, M., 2011: Watch the Virginia earthquake spread across Twitter. Gizmodo, 25 August, http://gizmodo.com/5834048/watch-the-virginia-earthquake-spread-across-twitter.

  • Höppner, C., R. Whittle, M. Bründl, and M. Buchecker, 2012: Linking social capacities and risk communication in Europe: A gap between theory and practice? Nat. Hazards, 64, 17531778, https://doi.org/10.1007/s11069-012-0356-5.

    • Search Google Scholar
    • Export Citation
  • Houston, J. B., B. Pfefferbaum, and C. E. Rosenholtz, 2012: Disaster news: Framing and frame changing in coverage of major U.S. natural disasters, 2000–2010. J. Mass Commun. Quart., 89, 606623, https://doi.org/10.1177/1077699012456022.

    • Search Google Scholar
    • Export Citation
  • Hu, X., X. Wu, P. Yin, and X. Zheng, 2017: An investigation into consumers’ continued social shopping intentions. Proc. Pacific-Asia Conf. on Information Systems, Kedah, Malaysia, Association for Information Systems, 63, https://aisel.aisnet.org/pacis2017/63/.

  • Hyvärinen, O., and E. Saltikoff, 2010: Social media as a source of meteorological observations. Mon. Wea. Rev., 138, 31753184, https://doi.org/10.1175/2010MWR3270.1.

    • Search Google Scholar
    • Export Citation
  • Koenker, R., and G. Bassett Jr., 1982: Robust tests for heteroscedasticity based on regression quantiles. Econometrica, 50, 4361, https://doi.org/10.2307/1912528.

    • Search Google Scholar
    • Export Citation
  • Lee, K. L., R. J. Meyer, and E. T. Bradlow, 2009: Analyzing risk response dynamics on the web: The case of Hurricane Katrina. Risk Anal., 29, 17791792, https://doi.org/10.1111/j.1539-6924.2009.01304.x.

    • Search Google Scholar
    • Export Citation
  • Levin, D. Z., and R. Cross, 2004: The strength of weak ties you can trust: The mediating role of trust in effective knowledge transfer. Manage. Sci., 50, 14771490, https://doi.org/10.1287/mnsc.1030.0136.

    • Search Google Scholar
    • Export Citation
  • Levin, I. P., and G. J. Gaeth, 1988: How consumers are affected by the framing of attribute information before and after consuming the product. J. Consum. Res., 15, 374378, https://doi.org/10.1086/209174.

    • Search Google Scholar
    • Export Citation
  • Li, L., M. F. Goodchild, and B. Xu, 2013: Spatial, temporal, and socioeconomic patterns in the use of Twitter and Flickr. Cartogr. Geogr. Inf. Sci., 40, 6177, https://doi.org/10.1080/15230406.2013.777139.

    • Search Google Scholar
    • Export Citation
  • Lin, X., K. A. Lachlan, and P. R. Spence, 2016: Exploring extreme events on social media: A comparison of user reposting/retweeting behaviors on Twitter and Weibo. Comput. Human Behav., 65, 576581, https://doi.org/10.1016/j.chb.2016.04.032.

    • Search Google Scholar
    • Export Citation
  • Lindell, M. K., and R. W. Perry, 2012: The protective action decision model: Theoretical modifications and additional evidence. Risk Anal., 32, 616632, https://doi.org/10.1111/j.1539-6924.2011.01647.x.

    • Search Google Scholar
    • Export Citation
  • Liu, B. F., J. D. Fraustino, and Y. Jin, 2016: Social media use during disasters: How information form and source influence intended behavioral responses. Commun. Res., 43, 626646, https://doi.org/10.1177/0093650214565917.

    • Search Google Scholar
    • Export Citation
  • Liu, S. H., H. L. Liao, and J. A. Pratt, 2009: Impact of media richness and flow on e-learning technology acceptance. Comput. Educ., 52, 599607, https://doi.org/10.1016/j.compedu.2008.11.002.

    • Search Google Scholar
    • Export Citation
  • Ma, X. J., W. X. Wang, Y. C. Lai, and Z. Zheng, 2010: Information explosion on complex networks and control. Eur. Phys. J., 76B, 179183, https://doi.org/10.1140/epjb/e2010-00208-1.

    • Search Google Scholar
    • Export Citation
  • Meyer, R. J., E. J. Baker, K. F. Broad, J. Czajkowski, and B. Orlove, 2014: The dynamics of hurricane risk perception: Real-time evidence from the 2012 Atlantic hurricane season. Bull. Amer. Meteor. Soc., 95, 13891404, https://doi.org/10.1175/BAMS-D-12-00218.1.

    • Search Google Scholar
    • Export Citation
  • Miao, Q., and D. Popp, 2014: Necessity as the mother of invention: Innovative responses to natural disasters. J. Environ. Econ. Manage., 68, 280295, https://doi.org/10.1016/j.jeem.2014.06.003.

    • Search Google Scholar
    • Export Citation
  • O’Reilly, T., 2007: What is web 2.0: Design patterns and business models for the next generation of software. Commun. Strategies, 65, 1737, https://mpra.ub.uni-muenchen.de/id/eprint/4580.

    • Search Google Scholar
    • Export Citation
  • Ripberger, J. T., 2011: Capturing curiosity: Using Internet search trends to measure public attentiveness. Policy Stud. J., 39, 239259, https://doi.org/10.1111/j.1541-0072.2011.00406.x.

    • Search Google Scholar
    • Export Citation
  • Ripberger, J. T., H. C. Jenkins-Smith, C. L. Silva, D. E. Carlson, and M. Henderson, 2014: Social media and severe weather: Do tweets provide a valid indicator of public attention to severe weather risk communication? Wea. Climate Soc., 6, 520530, https://doi.org/10.1175/WCAS-D-13-00028.1.

    • Search Google Scholar
    • Export Citation
  • Rive, G., J. Hare, J. Thomas, and K. Nankivell, 2012: Social media in an emergency: A best practice guide. Wellington Region CDEM Group Rep., 58 pp., https://www.civildefence.govt.nz/assets/Uploads/CDEM-Resilience-Fund/CDEM-Resilience-Fund-2011-03-social-media-in-an-emergency-guide.pdf.

  • Ros-Diego, V. J., and A. Castelló-Martínez, 2011: CSR communication through online social media. Rev. Lat. Comun. Soc., 67, 4767, https://doi.org/10.4185/RLCS-067-947-047-067.

    • Search Google Scholar
    • Export Citation
  • Sakaki, T., M. Okazaki, and Y. Matsuo, 2010: Earthquake shakes Twitter users: Real-time event detection by social sensors. Proc. 19th Int. Conf. on World Wide Web, Raleigh, NC, International World Wide Web Conference Committee, 851–860.

  • Scheufele, D. A., 1999: Framing as a theory of media effects. J. Commun., 49, 103122, https://doi.org/10.1111/j.1460-2466.1999.tb02784.x.

    • Search Google Scholar
    • Export Citation
  • Scheufele, D. A., and D. Tewksbury, 2007: Framing, agenda setting, and priming: The evolution of three media effects models. J. Commun., 57, 920, https://doi.org/10.1111/j.0021-9916.2007.00326.x.

    • Search Google Scholar
    • Export Citation
  • Schumacher, R. S., D. T. Lindsey, A. B. Schumacher, J. Braun, S. D. Miller, and J. L. Demuth, 2010: Multidisciplinary analysis of an unusual tornado: Meteorology, climatology, and the communication and interpretation of warnings. Wea. Forecasting, 25, 14121429, https://doi.org/10.1175/2010WAF2222396.1.

    • Search Google Scholar
    • Export Citation
  • Semetko, H. A., and P. M. Valkenburg, 2000: Framing European politics: A content analysis of press and television news. J. Commun., 50, 93109, https://doi.org/10.1111/j.1460-2466.2000.tb02843.x.

    • Search Google Scholar
    • Export Citation
  • Simmons, K. M., and D. Sutter, 2008: Tornado warnings, lead times, and tornado casualties: An empirical investigation. Wea. Forecasting, 23, 246258, https://doi.org/10.1175/2007WAF2006027.1.

    • Search Google Scholar
    • Export Citation
  • Sorensen, J. H., 2000: Hazard warning systems: Review of 20 years of progress. Nat. Hazards Rev., 1, 119125, https://doi.org/10.1061/(ASCE)1527-6988(2000)1:2(119).

    • Search Google Scholar
    • Export Citation
  • Stafford, T. F., M. R. Stafford, and L. L. Schkade, 2004: Determining uses and gratifications for the Internet. Decis. Sci., 35, 259288, https://doi.org/10.1111/j.00117315.2004.02524.x.

    • Search Google Scholar
    • Export Citation
  • Stapa, S. H., and A. H. Shaari, 2012: Understanding online communicative language features in social networking environment. GEMA, 12, 817830, http://www.ukm.my/ppbl/Gema/gemacurrentissues.html.

    • Search Google Scholar
    • Export Citation
  • Tang, Z., L. Zhang, F. Xu, and H. Vo, 2015: Examining the role of social media in California’s drought risk management in 2014. Nat. Hazards, 79, 171193, https://doi.org/10.1007/s11069-015-1835-2.

    • Search Google Scholar
    • Export Citation
  • Tumasjan, A., T. O. Sprenger, P. G. Sandner, and I. M. Welpe, 2010: Predicting elections with Twitter: What 140 characters reveal about political sentiment. Proc. Fourth Int. AAAI Conf. on Weblogs and Social Media, Washington, DC, AAAI, 178–185.

  • Tversky, A., and D. Kahneman, 1981: The framing of decisions and the psychology of choice. Science, 211, 453458, https://doi.org/10.1126/science.7455683.

    • Search Google Scholar
    • Export Citation
  • Vieweg, S., A. L. Hughes, K. Starbird, and L. Palen, 2010: Microblogging during two natural hazards events: What Twitter may contribute to situational awareness. Proc. SIGCHI Conf. on Human Factors in Computing Systems, Atlanta, GA, ACM, 10791088.

  • Weber, R. P., 1990: Basic Content Analysis. 2nd ed. Sage Publications, 96 pp.

  • Wei, J., J. Xu, and D. Zhao, 2015: Public engagement with firms on social media in China. J. Inf. Sci., 41, 624639, https://doi.org/10.1177/0165551515586712.

    • Search Google Scholar
    • Export Citation
  • Wei, J., F. Wang, and M. K. Lindell, 2016: The evolution of stakeholders’ perceptions of disaster: A model of information flow. J. Assoc. Inf. Sci. Technol., 67, 441453, https://doi.org/10.1002/asi.23386.

    • Search Google Scholar
    • Export Citation
  • Westerman, D., P. R. Spence, and B. Van Der Heide, 2014: Social media as information source: Recency of updates and credibility of information. J. Comput. Mediat. Commun., 19, 171183, https://doi.org/10.1111/jcc4.12041.

    • Search Google Scholar
    • Export Citation
  • Wogalter, M. S., 2006: Handbook of Warnings. CRC Press, 864 pp.

  • Wong, T. F., and Y. Y. Yan, 2002: Perceptions of severe weather warnings in Hong Kong. Meteor. Appl., 9, 377382, https://doi.org/10.1017/S1350482702003110.

    • Search Google Scholar
    • Export Citation
  • Zhang, F., and Coauthors, 2007: An in-person survey investigating public perceptions of and responses to Hurricane Rita forecasts along the Texas Coast. Wea. Forecasting, 22, 11771190, https://doi.org/10.1175/2007WAF2006118.1.

    • Search Google Scholar
    • Export Citation
1

Web 2.0 means that software is delivered as a continually updated service that improves the more people use it, consuming and remixing data from individual users while providing its own data and services in a form that allows remixing by others. The core of web 2.0 is the support for user-generated content (O’Reilly 2007).

2

Report on social media user behavior in 2014 by CNNIC.

3

Note that 1 is for town level, 2 for county level, 3 for city level, 4 for multiple-city level, 5 for province level, and 6 for multiple-province level.

Save