Abstract
Despite the growing popularity of digital narratives, research on digital storytelling and its spread through social media interactions remains limited. Inspired by the social contagion theory, we introduce the concept of narrative contagion—where a story shared by a person or organization prompts others to share their stories—and investigate its process and outcome in online cancer communities. Utilizing a large dataset of 849 Facebook posts, 47,291 comments, and 14,466 replies, our artificial intelligence (AI)-based computational analysis provides evidence for narrative contagion among individual users in organization-hosted cancer communities on Facebook. It also reveals different organizational message characteristics, such as emotional arousal, post topic, and request for storytelling, that affect user storytelling and emotional support in comments and replies. Our results contribute to a deeper understanding of digital storytelling and its collective dynamics in online communities.
In the rapidly evolving landscape of new media, storytelling enables individuals and organizations to exchange information, construct meaningful connections, and encourage social participation (Lundby, 2008; Vicari, 2021). Traditionally, digital storytelling refers to a workshop-based practice where individuals learn to use digital media to produce short audiovisual stories, often centered around their own lives (Hartley and McWilliam, 2009). It has been widely used in health promotion and education, which allows people to share stories to reflect on their illness journey and build partnerships (Briant et al., 2016; Lambert, 2010 [2007]). With the increasing popularity of social networking sites, many digital stories are told and shared on social media (Johnson, 2018). In particular, social media offers various forms of digital storytelling that users can utilize to share their stories. For example, on Facebook, users can share their feelings, thoughts, and life events through status updates; they can create photo albums to tell a visual story; they can join a Facebook group and engage in deep conversations; or they can create reels—short, fun videos with audio and augmented reality (AR) effects—to share with their friends. These diverse forms enable users to express themselves and connect with others in multifaceted ways. A growing number of studies have been conducted to understand their creation, interpretation, and sharing (Clark et al., 2015; Sampson, 2012; Zhao et al., 2022), particularly in the health context (Hale et al., 2018; Ma et al., 2023; Rodgers and Stemmle, 2020).
Two notable gaps exist within this line of research: First, while existing research on narrative communication has examined the spread of narratives through the lens of individual psychology such as the need to release emotions (Harber and Cohen, 2005) or digital literacy (Clark et al., 2015), there lacks research theorizing the spread of digital narratives in online communities, encompassing both individuals and organizations (Clark et al., 2015; Zhao et al., 2018). Nabi and Green (2015) pointed to the insufficient scholarly attention to the social sharing of emotionally evocative stories. Thus, it is imperative to delve deeper into digital storytelling as both a personal expression and networked experience. Second, existing work on digital media concentrates on post-level narratives and user engagement indicated by metrics such as likes, shares, and comments (e.g., Zhao et al., 2022), while overlooking the variety and functions of narratives within comments and replies. A theoretically and practically important question is the ripple effect of digital storytelling, namely, whether and how digital narratives in original posts promote the sharing of more stories in the comments and further in the replies to those comments. Addressing this query can contribute to a refined and ecologically valid comprehension of digital storytelling and sharing on social networking sites.
Social contagion theory (Burt, 1987, 1992) offers a theoretical framework to understand the collective dynamics of narrative spread through social connections. Contemporary digital media research emphasizes that social contagion is co-constructed through interactions within an online network (Sampson, 2012), shaped by network structure, user interactions, and platform affordances (Nahon and Hemsley, 2013). Inspired by the literature on social and emotional contagion (Barsade, 2002; Barsade et al., 2018; Levy and Nail, 1993), we theorize narrative contagion as the process in which a story shared by a person or organization prompts others to share their stories. Based on the literature on social media, narrative communication, and social contagion, this study investigates the process and outcome of narrative contagion in social media interactions within organization-hosted cancer communities on Facebook. Specifically, we employed large-scale data comprising 849 Facebook posts, 47,291 comments, and 14,466 replies, utilizing artificial intelligence (AI)-powered computational analysis to investigate the impact of post narrative status (i.e., narrative vs non-narrative), in conjunction with other message characteristics, on narrativity and emotional support in user responses.
Narrative communication and social media
Narratives are “a representation of connected events and characters that has an identifiable structure, is bounded in space and time, and contains implicit or explicit messages about the topic being addressed” (Kreuter et al., 2007: 222). They are a popular form of communication in cancer support groups on social media (Basch and MacLean, 2019; Chou et al., 2011). This is because narratives can reduce resistance to cancer prevention behaviors, render complex information comprehensible, provide connections and support, and address the emotional and existential dilemmas of cancer (Kreuter et al., 2007). Indeed, many non-profit organizations, such as Susan G. Komen, often share community members’ stories to build connections, offer support, and raise funds (Ma et al., 2023; Mitchell and Clark, 2021).
There has been growing scholarly attention to narratives conveyed through social media, as opposed to the traditional context (Hale et al., 2018; Rodgers and Stemmle, 2020; Zhao et al., 2022). Emerging research on social media cancer narratives has mostly focused on their characteristics and relationships with user engagement (Ma et al., 2023). For example, Rodgers and Stemmle (2020) identified that stories about cancers of the breast and prostate are often told in chronological order and the first-person perspective. Moreover, a content analysis of YouTube cancer vlogs found that sharing personal experiences, such as the journey through diagnosis and treatment, correlated with receiving empathic support in comments (Hale et al., 2018). Ma et al. (2023) recently introduced a framework for analyzing cancer narratives on Facebook. Their findings showed that breast cancer narratives that were longer, less emotionally intensive, narrated from the perspective of a breast cancer patient, included information related to gender identity, depicted the act of offering social support, explicitly asked for involvement, and incorporated vivid visual tended to elicit higher levels of engagement.
Existing studies suggest that cancer stories, especially those told with certain features, can effectively engage users on social media, as indicated by the number of likes, shares, and comments. However, key questions remain—Do narrative posts encourage further storytelling and emotional support in user responses? What message factors affect storytelling among social media users? Theoretically, answering these questions is vital for understanding the spread of narratives in social media interactions. Practically, it can help organizations harness the power of storytelling to foster a sense of community and mobilize support. We discuss the relationships between narrative posts, user storytelling, and emotional support in online cancer communities in the following section.
Storytelling and emotional support in online communities
Social media users are motivated to share their experiences and offer support in online communities, likely because these digital spaces are platforms for building and expressing social identities. According to the social identity theory (Tajfel and Turner, 1979), people develop identities associated with their memberships in different groups, termed social identities. In health contexts, different types of social identities may be relevant, such as identities based on large social categories (e.g., race, gender), health behaviors (e.g., runner, smoker), and health status (e.g., cancer survivors; Harwood and Sparks, 2003). Social identities have affective implications: People develop emotional ties with other group members (Mackie et al., 2009). Social media users who follow breast cancer organizations may develop identification and a sense of community with the online support group through their interaction with the organizations and other users (Zhao et al., 2015).
Digital narratives may promote a sense of belonging for social media users by reducing their uncertainty, building trust, and highlighting a shared vision. Followers of breast cancer organizations on social media may be closely related to breast cancer survivors or could be cancer survivors themselves. The cancer journey is filled with uncertainties about the disease, treatment choices, and everyday life coping, which bring about stress and anxiety. Narrative posts share stories about cancer survivors or cancer research, providing information about the disease in a way that is more engaging and easier to comprehend compared with non-narrative posts (Dahlstrom, 2014; Kreuter et al., 2007). They connect with followers, reassuring them that others share their experiences or the experiences of their loved ones. In situations of high uncertainty, people tend to identify with groups that provide clarity (Reid and Hogg, 2005). Therefore, the ability of digital narratives to reduce uncertainty in a relatable way is likely to enhance followers’ identification with the online community. Moreover, emotionally charged, empathetic digital narratives also communicate the organization’s concern for the struggles faced by its followers, establishing trust. The posts emphasize the shared vision of the organization and the online community members to combat the disease. The combination of trust and the shared vision further strengthens the sense of connection and belonging of community members (Zhao et al., 2015).
When social media users feel connected to the online community, they are motivated to contribute to the community by offering support to other members. In their review of the role of social identity in health, Haslam et al. (2009) argued that people are willing to help or support strangers based on a shared social identity. Similarly, Guan and So (2016) suggested that social identity is associated with the perception of different types of social support, particularly emotional support. In online patient communities specifically, identification with the community predicts people’s empathy toward other members (Zhao et al., 2013). Therefore, as digital narratives help social media followers build a sense of identity and belonging with the online community, the followers are likely to reciprocate by showing empathy and providing emotional support to other community members.
Social contagion theory
Social contagion is the process by which a person or group influences the emotions, attitudes, or behavior of another through social networks (Burt, 1987, 1992). 1 Social contagion is a type of social influence, where the recipient does not perceive an intentional influence attempt from the sender (Levy and Nail, 1993). According to Burt (1992), social contagion serves as a key mechanism of diffusion and innovation to explain how early adopters of new ideas influence others slower in adoption. Among various forms of social contagion, emotional contagion has been the central focus of scholarly attention, with numerous studies supporting its prevalence in interpersonal interactions (Elfenbein, 2014; Hatfield et al., 1993) and social media (Ferrara and Yang, 2015; Wang and Lee, 2021).
Sampson (2012) highlights how social media amplifies and extends the impact of emotional contagion. He posits that emotional contagion is co-produced through interactions within a network, influenced by network connectivity and the emotional intensity of connections. This network-based relational perspective moves beyond traditional biological metaphors that simplify social contagion as spreading linearly like a virus. Nahon and Hemsley (2013) further address user interactions (e.g., shares) and network structure (e.g., key actors as gatekeepers) as dynamic forces shaping social contagion. From a user experience perspective, Hemsley and Kelly’s (2019) interviews show that both user endorsement and site algorithms affect the spread of content elements on social sites.
In the context of online cancer communities, storytelling plays a prominent role (Høybye et al., 2005) and is connected to emotional contagion (Wei et al., 2023). However, the spread of digital narratives through social media interactions remains unexplored. Nabi and Green (2015) noted this gap, emphasizing the lack of scholarly attention to the social sharing of emotionally evocative stories. They also discussed secondary social sharing, where narratives shared with one group are relayed to others, aligning with the emotional broadcaster theory of emotional disclosure (Harber and Cohen, 2005). While this line of work addresses the spread of narratives rooted in individual psychological processes, it does not account for the relational dynamics and message-specific factors affecting story-sharing on social networking sites.
Narrative contagion
Social contagion theory offers a theoretical framework to understand the collective dynamic of narrative sharing and spread in online environments. We define narrative contagion as the process in which a story shared by a person or organization prompts others to share their stories. Distinct from narrative persuasion, narrative contagion occurs when the recipient does not perceive an intentional influence attempt from the sender (Levy and Nail, 1993). Narrative contagion also differs from narrative diffusion, which primarily focuses on the spread of an identical story (Harber and Cohen, 2005). Instead, narrative contagion emphasizes the spread of narrativity, which encompasses a collection of narrative themes, among users within an online community. For instance, if a user, prompted by an organization’s survivor story, shares her mother’s similar experience, this involves narrative contagion because there is a transfer of the narrative theme of survivorship. Narrative contagion suggests that one’s storytelling can lead to a chain reaction of further commenting and storytelling among other users through social networks in online communities.
Drawing from the literature on social contagion (Barsade, 2002, Barsade et al., 2018; Levy and Nail, 1993), narrative contagion on social media can be attributed to two core mechanisms: the emotional and cognitive mechanisms. First, social media users infer emotional cues from multimodal elements in a sender’s message, such as negative words, emojis, and happy faces, and respond with corresponding emotions. The tendency of emotional inference might be more pronounced among those connected to an online cancer community, given their shared sense of belonging and identification (Zhao et al., 2015). Hancock et al. (2008) found that participants experimentally primed with negative emotions used words signaling those emotions more frequently and exhibited slower responses while communicating with a partner over an instant messaging platform. As textual and visual cues in online interactions can afford social media users to discern emotions, users mimic these emotional expressions and provide feedback by sharing their own emotionally charged stories.
In addition, social media users can compare their experiences and moods with the characters or events in the stories and then respond based on what seems appropriate for the situation (Barsade et al., 2018). In the context of online cancer communities, users as in-group members perceive similarity or identification of one another, leading to affective convergence (Elfenbein, 2014) likely exhibited through telling their own stories or providing emotional support. Similarly, according to the social learning theory (Bandura and Walters, 1977), individuals engage in social contagion through observation and imitation of other users’ behaviors, because observing others perform a certain behavior reduces the perceived constraint of the behavior. Taken together, members in an online community who perceive similarity with each other are expected to experience narrative contagion, driven by the cognitive mechanisms of social comparison and learning.
Summary of hypotheses and research questions
This study investigates the process and outcome of narrative contagion in social media user responses, specifically user comments and replies in breast cancer communities hosted by non-profit organizations on Facebook. Based on the discussion of narrative contagion, when an organization tells an emotionally evocative story, users are expected to respond by sharing their own stories, motivated by emotional resonance and a process of social comparison and learning. The story can also evoke a sense of connection and identification among users in an online cancer community, prompting a feeling of empathy and a desire to offer emotional support among users. Thus, we propose the first hypothesis on the impact of post narrative status on user responses:
H1: Post narrative status is positively associated with (a) narrativity and (b) emotional support in user responses on social media.
We also examine the roles of various message characteristics in affecting narrative contagion on social media. The literature on emotional contagion suggests that emotional arousal can promote emotional contagion (Barsade, 2002; Elfenbein, 2014). It’s likely that the heightened emotional intensity not only better captures the audience’s attention but also conveys a stronger emotional signal requesting audience storytelling. Meanwhile, a higher level of emotional arousal could indicate a more pressing need for various forms of emotional support. Thus, emotional arousal in posts is expected to increase the level of narrativity and emotional support in user response.
H2: Emotional arousal is positively associated with (a) narrativity and (b) emotional support in user responses on social media.
Recent studies also show that the explicit request for certain actions such as donation can promote user engagement on social media (Chae, 2021; Ma et al., 2023). Following this rationale, the explicit request for storytelling in social media posts should increase narrativity in user responses. As such, we propose the following hypothesis:
H3: Request for storytelling is positively associated with narrativity in user responses on social media.
Another message characteristic potentially affecting narrative contagion is post topic. Recent studies show that message topic plays an important role in affecting information diffusion and engagement (Yang et al., 2018; Zhao and Chen, 2022). For example, Wang and Lee (2021) found that cancer-related tweets emphasizing social relationships had a broader reach. However, it is not known what role post topic plays in narrative contagion in the context of online cancer communities. We thus ask a research question (RQ) on how post topic affects narrativity and emotional support in user responses:
RQ1: How is post topic associated with (a) narrativity and (b) emotional support in user responses?
Furthermore, we examine three aspects of narrative structure, including plot complexity, initial plot, and flashback, and their relationship with narrativity and emotional support. Plot complexity is operationalized as the number of events in a story, which can indicate how intricate the story is. The initial plot refers to the first event that appears in the story, which can set the tone or context for the story. A flashback refers to presenting a scene that occurred earlier in the story, which can be used to provide relevant background information. In sum, plot complexity, initial plot, and flashback may alter the perceived salience of a story and thus affect narrativity and emotional support in user responses. Given the exploratory nature of our study regarding narrative structure, we propose an RQ on how three aspects of narrative structure, including plot complexity, initial plot, and flashback, affect narrativity and emotional support in user responses.
RQ2: Among narrative posts, how does narrative structure (plot complexity, initial plot, and flashback) affect (a) narrativity and (b) emotional support in user responses?
Last, we examine how post narrative status influences narrative contagion at two distinct levels of user responses (i.e., comments and replies to comments) and analyze whether narrative contagion is amplified or dampened across these levels. As discussed, narrative contagion can be driven by emotional and cognitive processes (Barsade, 2002, Barsade et al., 2018; Levy and Nail, 1993). The presence of narratives in both posts and comments may convey stronger emotional implications and provide more information for social comparison and learning, leading to more storytelling in the replies. As such, we expect reply-level narrativity to be affected by both post- and comment-level narrativity.
H4: Narrativity at the reply level is positively associated with (a) post narrative status and (b) narrativity at the comment level.
Method
We compiled a list of non-profit organizations dedicated to breast cancer by conducting a thorough review of online publications (e.g., Canadian Cancer Survivor Network, n.d.; Huizen, 2019). We scrutinized the Facebook pages of organizations with a general breast cancer focus, as opposed to specific subtypes. From this, we pinpointed the top five organizations with the most Facebook followers: Susan G. Komen for the Cure, Breast Cancer Now, National Breast Cancer Foundation (US), National Breast Cancer Foundation (Australia), and A Future Without Breast Cancer—Canadian Cancer Society. We downloaded a total of 8580 Facebook posts from these identified organizations from January 1, 2016, to February 20, 2021, using CrowdTangle, a social media analytical software owned by Facebook. We then focused on the top 10% of these posts in terms of total interactions (sum of emotional reactions, likes, comments, and shares) (N = 858) and retrieved the corresponding responses, including both comments and replies (N = 107,458), using Facebook Scraper, an open-source tool accessible on GitHub. This allowed us to prioritize top-engaging posts, ensuring a robust dataset of comments and replies for analysis.
To achieve a balanced dataset of user responses, we sampled 100 comments and 50 replies from each post. If a post had fewer than 100 comments or 50 replies, we retained all available responses. After removing missing data, our final dataset includes 849 Facebook posts, 47,291 comments, and 14,466 replies.
Manual content analysis
We conducted a manual content analysis to determine the narrative status of the posts. Following the definition of narratives, a post was coded as a narrative if its text includes at least one character who experienced some events (Bilandzic and Busselle, 2013). Two expert coders annotated the narrative status of all posts. The overall agreement rate was above 0.9. The disagreements were resolved by discussion, and the consensus results were used for further analyses (i.e., the highest standard of intercoder reliability [ICR]; Riff et al., 2019).
If a post was identified as a narrative, we further content analyzed its narrative structure by identifying a sequence of events in the text. Following the cancer control continuum (National Cancer Institute, 2020), we coded five narrative events: (a) prevention and risk factors, (b) detection, screening, or diagnosis, (c) treatment process (e.g., getting the IV chemo) and effects (e.g., bald head), (d) treatment milestones and completion (e.g., ringing the chemo bell), and (e) survivorship, including recovery, recurrence, death, and philanthropic activities. ICR was achieved among three coders within four rounds of training (n = 60). For various narrative events, Krippendorff’s alphas varied from .77 to 1.00.
AI-powered computational analysis
We performed zero-shot classification through prompt engineering, leveraging OpenAI’s GPT-3 and GPT-4, conversation-based large language models (LLMs) trained on immense and diverse data (Zong and Krishnamachari, 2022). By allowing subject-matter experts to directly develop automated measures using intuitive natural language prompts, this method reduces reliance on complex, “black-box” machine learning algorithms, which are often difficult to interpret. This AI-based approach enhances transparency in research processes and broadens accessibility for diverse stakeholders. There has been initial support for the effectiveness of LLMs in coding social constructs, such as stances and emotions (Ziems et al., 2024).
In the paradigm of prompted zero-shot classification, the performance of LLMs can exhibit variability in response to prompts (Perez et al., 2021). To ensure valid and reliable automated measurement of narrative-related constructs, we adhered to best practices in prompt engineering as outlined by Ziems et al. (2024), such as giving instructions after the context to clarify the expected output. We proceeded to refine our prompts iteratively through two rounds of discussions, guided by narrative theory and expert coding results from a small sample (n = 50).
Upon identifying the prompts with the highest prediction accuracy (for details, see Table A1 in the Appendix), we conducted further validation using two distinct datasets drawn from random sampling: 10% posts (n = 86) and 150 comments/replies. These datasets were independently annotated by two expert coders for the calculation of ICRs between the automated tool and human coders. Human coders also used the prompts to guide their coding decisions. Regarding posts, Krippendorff’s alpha was .80 for emotional arousal and .87 for request for storytelling. Regarding comments/replies, Krippendorff’s alpha was .80 for narrativity and .69 for emotional support. After validation, we scaled up and used the finalized prompts for prediction.
Topic identification
We employed word embeddings and K-means clustering for topic identification. In particular, we harnessed the capacities of OpenAI’s text-embedding model (Text-embedding-ada-002), a powerful model that outperforms many other models in language-related tasks (Greene et al., 2022). OpenAI’s text-embedding model offers a richer and more nuanced representation of language by considering the context and semantic meanings surrounding words. As such, this approach yields a more accurate and contextually relevant solution for topic identification compared with traditional topic modeling techniques. Specifically, a text embedding is a high-dimensional numeric vector that represents the meaning of a text. If two pieces of text share similar meanings or topics, their numeric representations will be closer to each other. For instance, “dogs” and “cats” would have similar embeddings because they both relate to pets, whereas “dogs” and “computers” would be more distant. This is a more nuanced approach than simply counting how often words appear, allowing us to delve deeper into understanding the meaning of words.
After obtaining the text embedding of each post, K-means clustering was used to group these semantically similar texts, aiding in topic discovery. Based on the elbow method and silhouette method, we determined the optimal number of topics to be 4 (for details, see Figures A1 and A2 in Appendix). We then manually inspected the results based on 3, 4, or 5 topics and concluded that 4 topics were optimal, grounded in the principles of interpretability and parsimony. Figure A3 visualizes the four topics in 2D space.
Measures
Dependent variables
Narrativity
Following the definition of narratives (Bilandzic and Busselle, 2013), the level of narrativity in user responses was measured based on the extent to which the text tells a story about someone’s experience or action on a 3-point scale, where 1 indicates low (no experience or action), 2 indicates moderate (some experience or action), and 3 indicates high narrativity (detailed experience and action). To compute the narrativity in responses to each post, these response-level narrativity scores were averaged. Note that user responses include both comments and replies. As such, we computed three sets of narrativity scores in responses overall (M = 1.56, SD = 0.40), in comments (M = 1.63, SD = 0.47), and in replies (M = 1.38, SD = 0.37).
Emotional support
The presence of emotional support in user responses was measured by whether the text shows one’s emotional support, such as encouraging words, prayers, blessing, concern, sympathy, or gratitude, for people (e.g., cancer survivors) or issues (e.g., initiatives). Emotional support at the response level was coded in a binary way (1 = present, 0 = not present). To compute emotional support in responses to each post, we computed the frequency of responses providing emotional support. We also computed three sets of emotional support scores in responses overall (72.77%), in comments (70.25%), and in replies (72.74%).
Independent variables
Narrative status
Narrative status was measured by whether a post belonged to a narrative. 49.00% of posts were categorized as narrative, while the remaining 51.00% were categorized as non-narrative.
Request for storytelling
The presence of a storytelling request was measured based on whether the post asked users to tell or share their own stories, experiences, or thoughts. Note that request for other actions, including providing emotional support, congratulations, donations, or following the organization, was not counted. The request was present in 23.27% of posts.
Emotional arousal
Emotional arousal measures the intensity or activation level of emotions in the post on a binary scale from 0 (low arousal) to 1 (high arousal). Notably, 69.02% of the posts had high emotional arousal, whereas the remaining 30.98% had low emotional arousal.
Post topic
The posts encompassed four topics. The first topic focused on breast cancer research, including discussions on treatment and prevention, accounting for 22.97% of the content. For instance, posts highlighted positive facts, such as the increase in 5-year breast cancer survival rates. Second, 27.33% of the posts centered on motivating breast cancer survivors and paying tribute to their journeys, with content like, “We are thankful for the hope and inspiration found in the hearts of survivors everywhere.” Third, participation in cancer awareness and fundraising events made up 18.73% of the content, exemplified by references to initiatives like Pink Ribbon Breakfasts. Finally, personal experiences with breast cancer and the provision of resources constituted 30.98% of the content. Examples include “Finding a lump, having an abnormal mammogram, or hearing the words ‘you have cancer’ are unforgettable moments.”
Narrative structure
Narrative complexity was measured by the total number of narrative events in all narrative posts (M = 1.57, SD = 0.73). In cases where narrative posts contained multiple events (n = 173), we also recorded the narrative event occurring in the initial position of the post. Three dummy variables (1 = yes, 0 = no) were generated: detection as the initial event (63.01%), treatment (including process and outcome) as the initial event (24.86%), and survivorship as the initial event (10.98%). Flashback (24.28%) was recorded using a binary approach, indicating whether a narrative event that chronologically occurred after another event was narrated before the latter event in the post. This includes instances where a story initiated with recovery and subsequently discussed treatment.
Control variables
Control variables include likes received at posting, organization, message vividness, and text length (Zhao et al., 2022). We measured the total number of likes received by an account upon posting a message (M = 670,601, SD = 732,888), indicating the current level of popularity of the account. Organization was measured as dummy variables (1 = yes, 0 = no) for Susan G. Komen for the Cure (24.91%), Breast Cancer Now (21.99%), U.S. National Breast Cancer Foundation (16.96%), and Australian National Breast Cancer Foundation (22.34%). Text length was measured by the total number of words in the text of a post (M = 70.87, SD = 58.34). Message vividness was measured by the type of multimedia: external link (22.34%), photo (68.54%), and video (9.12%). Two dummy variables for photo and video were generated.
Results
We conducted regression models on two outcome variables, narrativity and emotional support in user responses including both comments and replies. These models included the same sets of control variables and predictors. Table 1 reports the estimated parameters of the models.
Regressions predicting narrativity and emotional support in responses.
p < .05; **p < .01; ***p < .001.
Regarding the effects of post narrative status on narrativity (H1a) and emotional support (H1b) in user responses (H1a), the results show that narrative posts (vs non-narrative posts) post had a lower level of narrativity in user responses with all the control variables considered, b = –0.13, SE = 0.04, p < .001. Narrative posts had a higher percentage of responses containing emotional support compared with non-narrative posts, b = 0.08, SE = 0.02, p < .001. H1b was supported, but H1a was not.
Emotional arousal in the post had a positive association with the percentage of responses containing emotional support, b = 0.06, SE = 0.02, p < .01, but not the level of narrativity in user responses. This supports H2b but not H2a. In addition, there was a positive effect of the request for storytelling on the level of narrativity in user responses, b = 0.24, SE = 0.03, p < .001, supporting H3.
Regarding the effects of post topics (RQ1), we found that the posts containing the topic of awareness and fundraising events were associated with a lower level of narrativity (b = –0.38, SE = 0.04, p < .001) and the percentage of emotional support in responses (b = –0.17, SE = 0.02, p < .001) in user responses, compared with other topics. The posts about the topic of personal experiences were associated with a higher level of narrativity (b = 0.08, SE = 0.04, p = .051) but not the percentage of emotional support in responses. The posts about the topic of motivating survivors and honoring journeys were positively associated with the percentage of emotional support (b = 0.06, SE = 0.02, p = .029) but negatively associated with the level of narrativity (b = –0.18, SE = 0.02, p < .001) in responses.
To test the effects of post- and comment-level narrativity on reply-level narrativity (H4), we constructed a path model with post-level narrativity as an exogenous variable, comment-level narrativity as an intermediary variable, and reply-level narrativity as an endogenous variable. Figure 1 shows that both post narrative status (b = 0.24, SE = 0.04, p < .001) and comment-level narrativity (b = 0.18, SE = 0.04, p < .001) had a positive direct effect on reply-level narrativity. This supports both H4a and H4b. A further examination of the total, direct, and indirect effects showed that there was a significant positive total effect of post narrative status on reply-level narrativity (b = 0.17, SE = 0.04, p < .001), with a slight negative indirect effect through comment-level narrativity (ab = –0.06, SE = 0.02, p < .001).
The relationship between post-level, comment-level, and reply-level narrativity.
Finally, regarding the effects of narrative structure of narrative posts (n = 420), we found that plot complexity was negatively associated with the level of narrativity (b = –0.06, SE = 0.02, p < .01) and the percentage of emotional support (b = –0.05, SE = 0.02, p = .021) in user responses. Among those narrative posts with more than one narrative event (n = 173), there was no effect of flashback. These posts starting with the narrative events of detection (b = 0.12, SE = 0.04, p < .01) and treatment (b = 0.06, SE = 0.03, p = .070) had a higher percentage of emotional support in user responses. Yet, the initial event did not affect the level of narrativity in responses.
Discussion
Overall, our results based on a computational analysis of a large Facebook dataset provide initial support for narrative contagion in organization-hosted cancer communities on Facebook and reveal various message characteristics in affecting storytelling and emotional support in user comments and replies to organizational posts. Our results contribute to a deeper understanding of digital narrative effects and the collective dynamics of storytelling in online health communities.
Our results suggest that storytelling in user responses was related to post narrative status, request for storytelling, and post topic. Surprisingly, post narrative status had a negative effect on user storytelling, inconsistent with the literature on social and emotional contagion (Barsade, 2002). Although we have controlled different organizations in the analysis, existing organization-public relationships might confound the relationship between narrative posts and user storytelling. While narratives could enhance perceived organizational authenticity, they may only prompt user storytelling when there are well-established organization-public relationships. For instance, Susan G. Komen often shares emotional and empowering personal stories of survivors, yet its relationship with the public has been complex, influenced by its recognized efforts in research and advocacy, as well as controversies such as pinkwashing. Regional and cultural differences might also account for the observation. We found that organizations in Australia and Canada use significantly fewer emotionally arousing stories than those in the United States and United Kingdom. This is likely due to stricter regulations on charitable fundraising (Phillips, 2020) and cultural preferences for factual over emotional appeals, as exemplified by the Australian National Breast Cancer Foundation. In addition, the explicit request for storytelling in the post turned out to be the most important factor affecting user storytelling. This supports the literature on the importance of making explicit requests to elicit target actions on social media (Chae, 2021). The topic of personal cancer experiences elicited user storytelling, whereas the topics of fundraising and survivor motivation reduced storytelling. This highlights the importance of personification for organization storytelling in boosting social media engagement.
The evidence of narrative contagion became apparent when we scrutinized user responses, differentiating between organizational-individual interactions in comments and individual interactions in replies. Both organizational narrative posts and user storytelling in the comments directly increased user storytelling in the replies. And organizational narrative posts had a stronger positive association with storytelling in replies when user storytelling was also present in comments. From a social identity viewpoint, narrative posts reduce uncertainty and build connections with the followers, which help followers construct social identity associated with the online community (Reid and Hogg, 2005; Zhao et al., 2015). Observing storytelling from others in the comments may further strengthen the identification, which prompts user engagement in the replies through mimicking the behavior of other community members, such as sharing stories and supporting those with a shared identity (Haslam et al., 2009). This supports the narrative contagion perspective: Within organization-hosted online communities, narrative contagion might be more associated with individual users’ engagement in storytelling, as peer storytelling signals stronger emotional cues and more information for social comparison and learning (Bandura and Walters, 1977). These results, in conjunction with the negative association between organizational narrative posts and user storytelling in comments, suggest that organizational narrative posts alone might not stimulate narrative contagion among individuals.
We also found that emotional support in user responses was affected by post narrative status, emotional arousal, and post topic. Posts containing cancer narratives, high emotional arousal, and the topic of survivor motivation increased the amount of emotional support in user response. Users are more likely to offer emotional support on social media when they identify with an online community that reduces uncertainty, fosters trust, and emphasizes a shared vision (Haslam et al., 2009; Zhao et al., 2015). Through sharing stories of cancer survivors or research, narrative posts reduce the users’ uncertainties about the cancer journey in an engaging and relatable way (Dahlstrom, 2014; Kreuter et al., 2007). Organization narrative posts also demonstrate that the organizations care about their followers’ experience with breast cancer and share the vision of combating the disease. As a result, it came as no surprise that narrative posts related to more emotional support compared with their non-narrative counterparts. In addition, emotionally charged posts, as well as posts that motivated cancer survivors and honored their cancer journey, may prompt social media users to display empathy and provide emotional support by increasing their desire to assimilate to the emotional state of the posts (i.e., affective convergence; Elfenbein, 2014).
Finally, our exploratory analysis showed that among narrative posts, plot complexity, measured by the number of events in a story, was negatively associated with the level of narrativity and emotional support in user responses. This finding is consistent with prior work that found low-complexity mystery stories were more enjoyable than medium- and high-complexity mystery stories (Knobloch-Westerwick and Keplinger, 2008). It suggests that a simpler story might be easier for readers to understand and connect with, leading to more story-sharing and evoking greater emotional support. Moreover, we found that narrative posts starting with the events of detection and treatment increased the percentage of emotional support in user responses. This finding is not surprising because such stories often share challenges faced during detection and the subsequent treatment, which can make readers more empathetic and understanding of the experience. It highlights the role that the initial event plays in setting the emotional tone for the story. Finally, we found that the positioning of narrative events (i.e., flashbacks) did not affect the level of narrativity in user responses. This finding implies that readers can connect with the story regardless of the chronological order of the events. It is, however, important to note that the stories examined in our study are relatively short and may involve few events.
Theoretical and practical implications
Our study contributes to narrative communication research, social contagion theory, and social media research. First, it advances narrative communication research by studying the ripple effect of digital storytelling in organization-hosted cancer communities online. Existing research has mostly focused on understanding the cognitive and emotional responses toward narratives, but seldom on the behavioral reactions (e.g., whether telling a story promotes further storytelling). By analyzing a large dataset of social media data using an AI-based approach, we show that organizational post characteristics, such as narrative status and storytelling request, as well as narrative structure such as plot complexity, influence user storytelling. Social media platforms, through their unique affordances, foster personalized and interactive spaces for storytelling and sharing. Our study explicates the spread of digital narratives in online communities, contributing to a refined understanding of the social sharing of emotional stories.
Second, it adds the narrative contagion effect to social contagion theory and digital media research. Although social contagion theory is often used to understand how a user’s behaviors or emotions can influence the actions or expressions of others in a network, our study suggests a similar effect: When a user shares a story, it can trigger a chain reaction of further storytelling among other users within the network. Our results differentiating individual and organizational actors also suggest the importance of understanding the varying impact of narrative contagion for different actors and the potential role of their relationships (e.g., organization–public relationships) in narrative contagion. This nuanced insight advances the network-based relational perspective of social contagion.
Our findings also offer several practical suggestions. First, organizations can explicitly request users to share their stories in the comments to foster a sense of online community and mobilize support. Second, sharing non-complex stories on social media can be an effective way to prompt users to offer emotional support to others. However, without an established relationship between the organization and the public, narrative posts may not naturally elicit storytelling in comments. Third, organizations should carefully consider the specific narrative events that they include in the posts, depending on their objectives of user engagement. For instance, posting personalized stories of cancer experiences can stimulate users’ story-sharing, whereas stories that are emotionally charged, mention survivor motivation, or start with diagnosis or treatment may encourage users to provide emotional support. Organizations should also consider regional and cultural differences when implementing this approach.
Limitations and future directions
Our study has several limitations. First, we centered our analysis on the most engaging posts from five breast cancer organizations on Facebook, allowing us to study narrative contagion within an organization’s post. Nonetheless, narrative contagion can manifest in various forms, such as when a user shares the organization’s story on their personal timeline or discusses it in a private chat. Due to our reliance on public Facebook data, our study did not capture other forms of narrative contagion. Future research can examine different forms of narrative contagion on other platforms such as Reddit. Second, our conclusion is confined to the contagion of breast cancer narratives among individual users in organization-hosted communities on Facebook. Any generalization of the findings should be made with caution. As our findings imply that narrative contagion can be conditioned by the type of actors and their relations, future research should investigate their impact on narrative contagion. Third, although our study provides initial evidence for narrative contagion, it does not establish a causal pathway from post-level storytelling through to comment-level storytelling and finally to reply-level storytelling. To establish these causal relationships, future research could employ longitudinal surveys or controlled experiments. Last, the reliability of emotional support between machine and human coding was somewhat lower than expected. This could stem from the AI-based measurement process requiring simultaneous judgments of the presence of support and its emotional nature. Given these nuanced constructs that involve dual judgments, future research might gain accuracy by employing chain-of-thought prompting techniques.
Supplemental Material
sj-pdf-1-nms-10.1177_14614448241285445 – Supplemental material for Analyzing narrative contagion through digital storytelling in social media conversations: An AI-powered computational approach
Supplemental material, sj-pdf-1-nms-10.1177_14614448241285445 for Analyzing narrative contagion through digital storytelling in social media conversations: An AI-powered computational approach by Xinyan Zhao, Zexin Ma and Rong Ma in New Media & Society
Footnotes
Funding
The author(s) received no financial support for the research, authorship, and/or publication of this article.
Supplemental material
Supplemental material for this article is available online.
Notes
Author biographies
References
Supplementary Material
Please find the following supplemental material available below.
For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.
For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.
