Getting the Word Out, Rain or Shine The Impact of Message Features and Hazard Context on Message Passing Online

• Published in: Weather, climate, and society Vol. 11; no. 4; pp. 763 - 776
• Main Authors: Sutton, Jeannette, Renshaw, Scott L., Vos, Sarah C., Olson, Michele K., Prestley, Robert, Gibson, C. Ben, Butts, Carter T.
• Format: Journal Article
• Language: English
• Published: Boston American Meteorological Society 01-10-2019
• Subjects: Communication; Digital media; Imagery; Mental task performance; Message passing; Meteorological services; Microstructure; Nonprofit organizations; Regression analysis; Regression models; Social media; Social networks; Social organization; Stakeholders; Visual system; Weather; Weather conditions; Weather effects; Weather forecasting
• ISSN: 1948-8327; 1948-8335
• DOI: 10.1175/WCAS-D-19-0021.1

Networked social media provide governmental organizations, such as the National Weather Service (NWS), the opportunity to communicate directly with stakeholders over long periods of time as a form of online engagement. Typologies of engagement include aspects of message content that provide information, contribute to community building, and inspire action and aspects of message microstructural features that facilitate interaction and dialogue, such as directed messages, hashtags, and URLs. Currently, little is known regarding the effect of message strategies on behavioral outcomes, and whether those effects vary under different weather conditions. In this paper we examine how message practices used on Twitter by the NWS are related to message engagement under routine and nonroutine weather conditions. Our analysis employs a census of tweets sent by 12 NWS Weather Forecast Offices in spring 2016 and uses a combination of manual and automated coding to identify engagement content and microstructure features present in each message. We identify factors that increase and decrease message retransmission (retweets) within this corpus under varying threat conditions, using a mixed-effects negative binomial regression model. We find that inclusion of actionable message content, information about historical weather facts, attached visual imagery (such as a map or infograph), and named event hashtags increases message passing during both threat and nonthreat periods. In contrast, messages that include forecast and nowcast content and messages that are sent in reply to other users have a lower passing rate. Findings suggest that common message features do alter message passing, potentially informing message design practices aimed at increasing the reach of messages sent under threat conditions.