Health Risk Information Engagement and Amplification on Social Media

Social Amplification of Health Risk Information

The SARF identifies four information mechanisms that contribute to the amplification of risk perceptions: volume of information about a particular risk, ambiguity of information and apparent disagreement among experts, dramatization of facts and possible consequences, and symbolic connotations communicated in the process of information sharing (Kasperson et al., 1988). Specific to the communication about an emerging pandemic, a recent study has shown that covering H1N1 influenza, mass media attention translated to an increased coverage of H1N1, and the dramatization of news occurred through increased sharing of threat over precautionary news (Klemm, Das, & Hartmann, 2016). Another study has looked at the news about Ebola and shown that the ambiguity of communicated signals, which resulted in positive appraisal of information and trust in governmental actions as well as negative appraisal and concerns about the spread of Ebola, present challenges for the dissemination of information about emerging pandemics (Johnson, 2016).

SARF also identifies audiences as communication stations that contribute to information amplification (Pidgeon, Kasperson, & Slovic, 2003). Amplification refers to the process during which audience members disseminate health risk information through their social networks, thus increasing information reach and engaging others (Pidgeon et al., 2003). Consequently, the tenets of the SARF allow explaining information behavior on social media. Seeking and passively consuming information can be viewed as the first step to the engagement with an issue (Napoli, 2011). However, in the context of health information dissemination, laypersons may not be able or willing to analyze fully the risk information that they encounter (McComas, 2006) and rely on shortcuts to assess the level of risk (Fischhoff & Kadvany, 2011). Information consumers who express interest in a particular news, for example, by liking a post on Facebook, create social cues that signal the value of information to others (Gittelman et al., 2015; Sun, Rau, & Ma, 2014). By actively commenting, information consumers increase the relevance of the topic for other members of their social networks, thus acting as dissemination agents for the original message and amplifying message effects.

Health Information and Audience Engagement

Collectively, social media are a powerful communication channel that can be used to disseminate information to large audiences (Ho, 2014). In general, social media can be defined as an environment that facilitates the creation and exchange of user-generated content (Rains, Brunner, & Oman, 2014). A growing body of research shows that communication through social media is a welcome trend among health consumers (Clayman, Manganello, Viswanath, Hesse, & Arora, 2010; Fisher & Clayton, 2012; Thackeray, Crookston, & West, 2013) and a practice widely adopted by public health organizations (Kim & Liu, 2012).

Social media both carry distinct benefits and present unique challenges for health communication and education. Specifically, they provide a platform for quick information dissemination about public health risks but may cause unintended consequences when audiences actively share their interpretations of information and contribute to the amplification of risk perceptions (Merchant, Elmer, & Lurie, 2011). These challenges may be especially dire for government organizations (Larson & Heymann, 2010). Yet the potential of social media as a dissemination channel has been recognized by public health agencies (Thackeray, Neiger, Smith, & Wagenen, 2012), and while most U.S. state–level public health agencies have small Facebook audiences (Thackeray et al., 2012), the CDC page had 573,530 followers as of March 31, 2016 (Facebook, 2016).

Social media listening can aid in the understanding of the mechanisms of information amplification, and Facebook feeds represent the evidence of information behavior for those who choose to communicate actively and play a role in information amplification. To examine the differences in social media information behaviors in response to posts about an emerging pandemic and general health promotion topics, this study asked the following research questions:

Situational Context

In December 2013, a 2-year-old boy in Guinea died of an unknown disease. Soon after, his mother, sister, and grandmother died. Only in March 2014, almost 4 months later, the spread of a deadly fever was recognized, and it took almost 10 months to identify the boy as Patient Zero in what has become the largest Ebola virus outbreak ever recorded (Baize et al., 2014; Gostin, Lucey, & Phelan, 2014). Although the cases of Ebola were primarily contained to West Africa, the outbreak attracted significant attention from the U.S. public (Love et al., 2014). Attention to Ebola increased with the withdrawal of several U.S. missionary medical workers in August 2014 and skyrocketed after the first domestic case of Ebola diagnosis in September 2014 (Upadhyay, Sittig, & Singh, 2014). As the information coverage of Ebola increased, the public’s anxiety and fear related to it rose as well (Chan, 2014). Addressing the public’s need for information, the CDC provided regular updates and news releases regarding Ebola while continuing to cover health promotion topics.

Data Source

The data were collected from the CDC Facebook channel through a Microsoft Excel add-on, Power Query, which allows downloading posts, comments, share counts, and limited user metadata. Data collection occurred in late November 2014, and all posts from March 25, 2014 (the date of the first World Health Organization report concerning Ebola in West Africa), to October 31, 2014 (end of data collection), were included in the analysis. A delay between the last post date and the data collection date allowed capturing comments to the later CDC posts.

In total, 836 posts and 35,973 comments to the original posts were downloaded from the CDC Facebook page. Using text search formulae, the posts were categorized into two groups, emerging pandemic (Ebola) and health promotion. Presence of the word “Ebola” was used to identify the posts for the former category. A word frequency analysis also showed that words “outbreak” (N = 67), “epidemic” (N = 15), and “pandemic” (N = 1) were mentioned in 83 posts, of which 27 related to a Salmonella outbreak. These 27 posts and associated comments were excluded from the analysis to minimize data contamination. Additionally, following methodological guidelines for content analysis (Krippendorff, 2012), a random sample of 50 health promotion posts was reviewed to establish the reliability of automated coding. All reviewed posts were not related to Ebola or other emerging pandemics and no manual changes were necessary.

The final data set included 652 (80.59%) posts on health promotion and 157 posts about Ebola. Table 1 shows additional information about associated comments and active users.

OPEN IN VIEWER

Table 1.

Data Set Characteristics Including Number and Percentage of Posts, Comments, Comment Likes, and Unique Users.

Topic	Posts, n (%)	Comments, n (%)	Comment, likes, n (%)	Users, n (%)
Health promotion	652 (80.59)	12,771 (35.75)	30,933 (33.97)	3,211 (25.58)
Ebola	157 (19.41)	22,948 (64.25)	60,116 (66.03)	8,406 (64.35)
Both topics				1,445 (11.06)
Total	809	35,719	91,049	13,062

A Facebook-assigned unique user ID was selected as the unit of analysis. Average number of likes per comment was operationalized as the measure of information amplification, and total comments per user was operationalized as the measure of information engagement. Table 2 provides information about other information behavior variables used in this study.

OPEN IN VIEWER

Table 2.

Operationalization of Variables.

Variable label	Variable coding description
User ID	Facebook-assigned unique user ID. This field was used as a unit of analysis.
Post ID	Facebook-assigned post ID was recoded and used to calculate the number of unique posts and comments per post ID.
Post day	Facebook-assigned original post date was recoded. March 25, 2014, was assigned a code of Day 1 with all subsequent days labeled sequentially in numerical order.
Ebola coverage	All collected posts were coded as 0 if no word “Ebola” was found by a text search formula, or 1 if the word “Ebola” appeared in the post. A random sample of 50 posts coded as non-Ebola was reviewed to ensure the quality of automated coding. All reviewed posts were not related to Ebola and no manual changes were necessary.
Total comments	The total number of Ebola and non-Ebola comments to the CDC Facebook page was calculated for each unique user based on the Ebola coverage variable.
Comment lag	Facebook-provided time-stamps for posts and comments, rounded to the fill hour, were used to calculate the lag time between a post and a comment. The average lag time in hours was calculated for each unique user and used on the data analysis.
Earliest comment day	The earliest day of a comment, using March 25 as Day 1
Latest comment day	The latest (most recent) day of a comment, using March 25 as Day 1 and October 31, 2014, as Day 221
Comment length	LEN() formula in Excel was used to calculate total comment length, which included spaces and all user-provided special characters and emoticons. The average comment length was calculated for each unique user and used on the data analysis.
Comment likes	Average number of likes per user comment was calculated to assess communication amplification contributed by CDC audience members.

Note. CDC = Centers for Disease Control and Prevention.

Analytic Procedures

A t test was run to answer Research Question 1 and compare mean differences in audience behavior for emerging pandemic and health promotion posts. Next, a cluster analysis was performed to answer Research Question 2 and identify audience segments of active CDC Facebook users who followed Ebola coverage. Earlier studies provided methodological suggestions for the identification of the number of clusters (Matthes & Kohring, 2008; Miller, Andsager, & Riechert, 1998; Strekalova, 2014). The hierarchical agglomerative analysis with squared Euclidean distance method was used to assess dissimilarity and distance among cluster items and identify the optimal number of clusters for the characteristics of user information behaviors. ¹ Using the elbow rule to examine at which point the addition of a cluster resulted in a lesser addition to variance explained, three groups were identified as the optimal number of clusters. Next, a two-step cluster analysis was performed to assign the users to groups and confirmed that three clusters was the optimal number for a good model fit (average silhouette = 0.7). Once users were assigned to cluster groups, an analysis of variance was conducted to determine differences across base variables.

Theoretical Implications

Although the framework was developed almost two decades before the introduction of social media as an information-sharing channel, this study contributed to the growing body of literature that showed that SARF provides a robust grounding for the analysis of audience engagement and communication amplification behaviors online (Chew & Eysenbach, 2010; Klemm et al., 2016; You & Ju, 2015). Social media have been identified as versatile communication channels that combine the features of a rich information source that affords interactive communication and feedback. Earlier studies of social media and discussions about the measurement of the information flow identified several important areas where measurement is necessary and effective: volume of communication, engagement and feedback from the audience, and the role of the audience as secondary sources of information. SARF identifies audiences as possible amplification stations, and changes in information behavior on social media can serve as indicators of information amplification. The results of this exploratory study allow formulating some operationalization and propositions related to the audience information behavior.

First, a comment lag can be used as an indicator of audience attention to and involvement with an issue. Specifically, this behavior suggests that online information consumers search for the past posts and exercise information value judgment when commenting. Subsequently, active audience members, serving as social amplification stations, are likely to search for relevant information rather than screen readily available and most recent information.

Second, the larger number of likes relevant to health promotion posts suggests that commenters indeed serve as amplification stations for select information topics. This finding, therefore, provides empirical evidence to SARF’s assumption about social amplification stations applied to social media environment. Health promotion communication is a broad area of research and practice, and future studies could use social amplification as a framework for the analysis of dissemination activities related various health domains. This study also found that the shortest comment lag was associated with more comment likes from other users. In other words, social amplification of recent information is likely to have a larger ripple effect compared with past information.

Third, earlier research on the link between information needs and information evaluation showed that the increase in people’s perceptions of risk, fear, and anxiety may lead to a higher need for information (Huurne & Gutteling, 2008). SARF also posits that information users may selectively attend to the information distributed by their social circles. This study showed that more users exhibited active information engagement in response to Ebola posts. Therefore, situations of high-affect information perceptions will increase the number of social media users actively engaging with information through active commenting and passive liking and further amplifying communication on the topic.

Practical Implications

Communication during emerging pandemics presents outreach and education opportunities for public health agencies. The twofold nature of online communication presents a challenge for public health organizations. On the one hand, these organizations need to raise awareness about particular topics of public interest and importance. In the case of Ebola, they may need to caution travelers or encourage donations for disease prevention efforts. At the same time, by posting more information about an emerging pandemic, these organizations may create a false sense of anxiety, which will amplify harm perceptions. The communication efforts could have two competing goals, to reassure people and to alert them at the same time. However, to be effective, health communication messages should also account for possible differences in the audiences and target the information. Previous research looked at the role of demographic factors that could serve as the base for communication targeting. However, information behavior factors can provide additional, psychographic information to explain how audiences engage with and use information.

Findings of this study have several practical implications. First, information posted on social media channels has a long-term, retrospective effect, but the majority of audience responds to most immediate posts. For topics of high importance, public health and health education organizations could plan reposting core messages to increase audience information exposure. Second, audience behaviors vary depending on an information topic. Therefore, targeted message and communication strategies could increase the effectiveness of social media communication efforts. For example, messages can include calls for action (e.g., post sharing) or invite comments. Third, the audiences that actively engage through social media are not homogenous and can form separate, nonoverlapping groups depending on the topic. As this study showed, only a tenth of the CDC Facebook followers were interested in both Ebola and non-Ebola information, which means that regular health promotion efforts should continue taking place during a health crisis.

Overall, this study demonstrated that information behavior and audience engagement is topic dependent. In addition, this study showed that online health information audiences are not homogenous and could have varying effects as communication amplification stations. Strategic message and communication decisions could help organizations engaged in online information dissemination achieve higher effectiveness in outreach and stakeholder engagement efforts.