How do users adopt health information from social media? The narrative paradigm perspective

Abstract

Introduction:

A recent trend in health information seeking and sharing is the use of social media. Although there are several benefits to the use of social media for health communication, the quality of health information exchanged on social media is troubling due to its informal, unregulated mechanisms for information collection, sharing and promotion. Therefore, it is important to understand how users adopt health information from social media.

Method:

Considering the user-generated and storytelling nature of social media messages, this research employed the narrative paradigm perspective to explain the social media health information adoption phenomenon. Specifically, narrative coherence (NC) and narrative fidelity (NF) were hypothesised to have positive effects on the intention to adopt (IA). Additionally, socio-economic status (SES) was viewed as a proxy variable to cognitive capability and was hypothesised to moderate the effects of NC and NF. A scenario-based survey was conducted to test the proposed research model.

Results:

We obtained a total of 257 valid questionnaires. The results indicated that NF (p < 0.001) had a positive effect on the IA social media health information. The NC (p < 0.01) had no impact on the low SES users but a positive impact on the high SES users. Further, the effect of NF (p < 0.01) on the IA was higher for high SES users than low SES users.

Conclusions:

NC and NF are two major driving forces in social media health information adoption, and the effect of both narrative paradigm variables depends on the SES users.

Implications

Results of this study show how the narrative paradigm, with a focus on the storytelling method of communication rather than logical scientific argument, can not only explain the uptake of health messages from social media, but also provide guidance as to how to create health messages on social media that more effectively target end users.

Keywords

information seeking behaviour social media narrative medicine consumer health information health care seeking behaviour health information management socioeconomic status

Introduction

With increasing availability of online medical information sources, a greater number of people are using the Internet to find health-related information (Xiao et al., 2014). Although search engines and health websites remain important information sources, a recent trend in health information seeking and sharing is the use of social media (De Choudhury et al., 2014). Social media exerts tremendous influence over the way people around the world – people of all ages – obtain and share information. More than 60% of Internet-connected individuals in the United States now participate in social media platforms every day, with Europe not far behind (Barry et al., 2011). Among these users, 23% followed their friends’ personal health experiences or updates on the site, 17% used social networking sites to remember or memorialise other people who suffered from a certain health condition and only 15% had not obtained any health information on social media sites (Fox and Jones, 2011).

Social media is also popular in China, which is actually the world’s largest social media market (Statista). Tencent’s Qzone, the most preferred social media brand in China, is a combination of Facebook and Tumblr, where users share photographs and leave messages and blogs (Statista). The number of social media users has also increased rapidly. Between January 2016 and January 2017, active social media users in China increased by 20%, with over half the population owning a social media account (Linkfluence). On average, 570 million users, including 93% in China’s first-tier cities, log on to the social media site WeChat every day (Zhang et al., 2017). According to a recent survey in China in 2017, almost all respondents had seen (98.35%) or read (97.68%) health information on WeChat (Zhang et al., 2017). Nearly one-third of respondents frequently received and read health information through WeChat, and WeChat was selected (63.26%) as the most common method for obtaining health information (Zhang et al., 2017).

Although there are several benefits to the use of social media for health communication, the quality of health information exchanged on social media is troubling (Moorhead et al., 2013). A recent survey in China found that only 14.43% of respondents believed that health information on social media was reliable, one of the major concerns being the lack of a guarantee of professionalism (Zhang et al., 2017). A systematic review also pointed to the lack of quality and reliability of health information available on social media as a limitation (Moorhead et al., 2013). The major reason for this poor quality information is the informal, unregulated mechanisms for information collection and sharing; any user – unidentified or without clinical credentials – can upload misleading content (Moorhead et al., 2013). Quality limitations for social media health information include lack of alignment of content with clinical practice guidelines, inaccuracy, misinformation and deceptive advertising (Weitzman et al., 2011). Although similar issues exist with traditional Internet websites, these issues are heightened by the interactive and user-generated nature of social media, which allows lay users to upload information regardless of quality (Adams, 2010). Therefore, a major challenge for users in the Web 2.0 age is how to identify the quality of social media information and decide whether or not to adopt the information. In this study, information adoption was defined as accepting the information as truth and abiding by that information in the future (Filieri and McLeay, 2014).

Previous studies have addressed the important question of user persuasion in information adoption. The most commonly used theoretical framework has been the elaboration likelihood model (ELM) of persuasion (Petty and Cacioppo, 1986). The ELM is a dual process theory describing attitude change, which proposes two major routes to persuasion: the central route and the peripheral route (Petty and Cacioppo, 1984). Under the central route, persuasion will likely result from a person’s careful and thoughtful consideration of the true merits of the information presented in support of an advocacy. Conversely, under the peripheral route, persuasion results from a person’s association with positive or negative cues in the stimulus or making a simple inference about the merits of the advocated position. Previous studies have investigated user persuasion of online health information using the ELM framework (Eastin, 2001; Mun et al., 2013). Eastin (2001) investigated the credibility of online health information. He found that both knowledge of the content and expertise of the source had positive effects on perceptions of online health information. Mun et al. (2013) explored antecedents of initial trust in Web-based health information. They found that the quality of the argument, the perceived expertise of the source and the users’ perceptions of information quality and risk played important roles in determining an individual’s decision to trust health information online.

Although the ELM framework has been a popular method used to explain user information adoption, many important clues in the ELM framework, such as source credibility (peripheral route), are missing or hard to judge in the social media message context, especially when messages have been forwarded on by users (e.g. through retweets on Twitter) (Lin et al., 2016). This is mainly due to the peer user-generated and storytelling nature of social media health messages, which area different from articles published on websites whose authors are usually experts or journalists, and the website’s reputation can be easily determined (Adams, 2010). Most contributors of social media health information are actually lay users, who have no medical training (Moorhead et al., 2013). Thus, it is very difficult to judge the credibility of information from these sources, making the ELM framework difficult to apply in the social media context. Therefore, in this study, we employed the narrative paradigm (Fisher, 1984) as the theoretical basis to explain user adoption of information from social media.

Narrative theory

The narrative paradigm, a theory proposed by the 20th-century communication scholar Walter Fisher, posits that all meaningful communication is a form of storytelling or reporting of events (Fisher, 1987). It promotes the belief that human beings are storytellers and listeners and are more persuaded by a good story than by a good argument. The premise is that everything we do is, and can be laid out, as a story. Fisher (1987) argued that we cannot, in fact, do anything without it attaining some kind of narrative structure. The main tenets of the narrative paradigm are (i) humans are essentially storytellers; (ii) decisions that humans make are based on “good reasons” rather than proof; (iii) what we do and how we think is swayed by history, biography, culture, and character; (iv) our rationality is determined by our sense of narrative coherence (NC; the coherency of the narrative) and narrative fidelity (NF; whether the story rings true in line with what we already know to be true) and (v) we are continually choosing stories that we keep company with, and these stories are constantly changing (Fisher, 1987).

In short, narratives are a selective reality, and we choose what we want to believe, which is influenced by external factors (Fisher, 1987). Narratives are intrinsically persuasive because they describe a particular experience rather than general truths. Narratives have no need to justify the accuracy of their claims; the story itself demonstrates the claim (Dahlstrom, 2014). Results from previous studies have suggested that audiences are more willing to accept normative evaluations from narratives than from more logical-scientific arguments (Green and Brock, 2000; Slater and Rouner, 2002). When narrative and statistical information were both present within a single message, perceptions were skewed towards the experiences of the specific cases regardless of whether the overall evaluations aligned or not (Gibson and Zillmann, 1994). Many researchers have shown that narrative paradigms can be used to sway beliefs about health topics such as vaccines (Brodie et al., 2001) and HIV/AIDS (Vaughan et al., 2000).

The current study

In this study, we were interested in the effects of NC and NF on the user’s intention to adopt (IA) social media health information. Therefore, the first research question (RQ) of this study was:

RQ1: Do NC and NF affect a user’s IA social media health information?

The information adoption mechanism may differ for different types of users. That is, the effect of NC and NF on the IA the information may not be universal. In this study, we were interested in how users with a high or low socio-economic status (SES) might employ different strategies to process NC and NF when they were about to make an information adoption decision. Therefore, the second RQ of this study was:

RQ2: How will the user’s SES impact the user’s processing strategy regarding NC and NF?

In other words, would the impact of NC and NF on social media health information adoption vary for high and low SES users?

To answer these questions, we developed a research model based on the narrative paradigm. Two important variables derived from narrative theory, coherence and fidelity, were hypothesised to have a positive impact on the IA social media health information. In addition, SES was hypothesised to positively moderate (strengthen) the impact of coherence and fidelity on the IA. A scenario-based survey was conducted to test the proposed model.

Research model and hypothesis development

The research model is shown in Figure 1. NC and NF were hypothesised to have significant effects on IA social media health information. SES was used as the moderator. Control variables included age, gender, marital status, Internet experience, online health experience, dispositional trust (DT), perceived susceptibility (PS) and perceived severity (PSE).

Figure 1.

Research model.

Fisher’s (1984) narrative paradigm described how all communication forms are narrative, meaning we communicate to tell stories or report on events. Those receiving these messages judge the validity of the messages based on their own belief systems, regardless of whether they fit within these boundaries or paradigms. Fisher argued that individuals are able to distinguish what makes a story legitimate by using what he referred to as narrative rationality. Rationality consists of two factors, coherence and fidelity, which contribute to judgments about good reasons (Griffin et al., 2009). Coherence is the degree to which a story makes sense structurally, while fidelity is concerned with whether the story is true.

Coherence concerns the question of whether or not a story coheres or “hangs together” and whether or not the story is free of contradictions (Fisher, 1985). There are many examples of NC. For example, Brinson and Brown (1997) described it as the internal consistencies of the story that include prefacing, recounting and closing sequences. Brinson and Brown believed that this structure could prove vital in the effectiveness of the viewer or listener’s belief in the narrative. Another example of NC was given by Dowell (2003), who applied narrative theory to a rhetorical analysis of three presidential crisis speeches, where NC was determined by three questions: (i) how well does the speech “hang together” – does the story flow logically and is it internally consistent?; (ii) how does the story compare and contrast to other communication – how does the speech compare to the two other presidential crisis speeches reviewed? and (iii) how reliable are the characters, both the speaker and characters, in the speech? Are the characters believable – do the characters’ actions contradict other actions or change in unusual ways? what are their motives?

Any content communicated is effective only if it makes sense to the listener, and coherence is the degree of sense making of a narrative. The effectiveness in delivering a story is influenced by factors such as the structure of the narrative, resemblance between stories and credibility of the characters. The relationship between NC and the persuasion effect has been demonstrated (e.g. Brinson and Brown, 1997; Dowell, 2003). Brinson and Brown (1997) evaluated the persuasive appeals of nine public service announcements (PSAs) that advocated either condom use for sexually active young adults or sexual inactivity. They found that the advertisements were more likely to be effective when they contained high NC (probability). Dowell (2003) also reported that NC (e.g. characterological coherence, material coherence and structural coherence) was particularly useful for understanding why political speeches were successful. Therefore, we hypothesised a positive relationship between NC and social media information adoption:

H1: NC will be positively associated with the IA social media health information.

Fidelity is concerned with whether or not a story is true. NF suggests that if a story matches our own beliefs and experiences, it will be accepted. Fidelity determines how the story fits into the background of the world as a person has known it. Questions to ask when considering the fidelity of the text are whether the story presented is faithful to the real-life experiences of the target audience and whether the story is true. Narrators can also contribute to the NF of the story. Wang and Arpan (2008) found the credibility of the message increased if the audience shared the ethnic identity of the spokesperson. Lumpkins (2012) incorporated NF with other factors such as technical qualities, narrators of the story, the values communicated and competing narratives to better understand breast cancer prevention among African American women. Dowell (2003) demonstrated NC in the analysis of three presidential crisis speeches by asking five questions: (i) what are the story’s values?; (ii) are the values appropriate for the story’s moral or the characters’ actions from the audience’s perspective?; (iii) do the values have positive consequences in the lives of people from the audience’s perspective?; (iv) are the story’s values in agreement with the audience? and (v) are the values part of an ideal script for social behavior from the audience’s perspective?

Fidelity defines the credibility or reliability of the narrated story, the ability to persuade the listener determines whether the person accepts the story or not. The relationship between NF and the persuasion effect has been demonstrated (Lumpkins, 2012; Olsen and Reynolds, 1999). Olsen and Reynolds (1999) found the strong effect that message fidelity had upon perceptions of message effectiveness and identification with the speaker. In another study by Lumpkins (2012), NF was used to design and evaluate the breast cancer risk narratives in a pilot sample of PSAs targeting underinsured and uninsured women. Therefore, we hypothesised there would be a positive relationship between NF and social media information adoption.

H2: NF is positively associated with the IA social media health information.

SES is a combined economic and sociological measure of a person’s work experience and of an individual’s or family’s economic and social position in relation to others. It is strongly associated with cognitive ability and achievement during childhood and beyond (Noble et al., 2005). Previous studies have indicated that SES is an important predictor of neurocognitive performance, particularly of language and executive function and that SES differences are found in neural processing even when performance levels are equal (Hackman and Farah, 2009). Therefore, the information-processing strategy may differ among people with different SES. A recent study suggested that individuals with different levels of SES vary in the heuristics and search patterns they rely upon to direct their searches to obtain health-related information (Perez et al., 2016). These results indicated lower socio-economic individuals were more likely to use an intuitive, rather than deliberative, approach to Internet health information seeking. Lower socio-economic participants were more likely than their higher socio-economic counterparts to narrow the scope of their search. Vaughan and Dunton (2007) also confirmed that socio-economic conditions may affect risk judgments through the utilisation or processing of relevant health information. Their findings showed that when compared to laborers who believed themselves to be less dependent on their employment situation, participants who felt more economically dependent utilised scientific evidence less when judging the risks presented by environmental chemicals. All this evidence indicates that SES is associated with the user’s cognitive performance or information-processing strategy. For users with a low SES, the information-processing strategy regarding NC and NF may be superficial and casual; however, such a strategy may be deep and definite for high SES users. Hence, the effect of NC and NF on the IA social media health information may be stronger for high SES users. Therefore, we proposed the following hypotheses:

H3: The higher the SES, the higher the positive effect of NC will be on the IA social media health information.

H4: The higher the SES, the higher the positive effect of NF will be on the IA social media health information.

Method

The survey

A scenario-based survey was conducted to test the proposed hypotheses. This is a useful method to investigate the behavior of people in certain contexts, especially where it is costly or impossible to reproduce the scenario in an experimental situation (e.g. in context of ethical dilemmas) (Jafarkarimi et al., 2016). A survey is also well suited for testing human behavior-related hypotheses and an appropriate method for this study. Ethics approval was obtained from the East China University of Science and Technology Human Research Ethics Committee.

Participants and recruitment

Data for this study were collected through an online questionnaire (see Appendix 1). Participants were recruited voluntarily from a public health forum in China Tianyayiyuan (http://bbs.tianya.cn/list-100-1.shtml) because they were typical social media users who should have had some experiences with social media health messages. Only subjects who had read at least one health-related text message sent from other “buddies” (friends) on WeChat in the last month were eligible for the survey. Recruitment advertisement messages were posted by the research assistant several times on the forum within a 1-week period. To better motivate participants, each user who completed the survey was rewarded with a mobile phone recharge card worth about $5.

Measures

SES was measured by self-reported income. SES is a theoretically based construct with three dimensions: income, education and occupation. Of these dimensions, income has been the most commonly used measure because of its significant association with health outcomes (Williams et al., 1997); education and occupation are strongly associated with income (Griliches and Mason, 1972). Therefore, income was used as the proxy variable to measure SES. Respondents were asked to rate their income from 1 to 6 with 1 the lowest and 6 the highest. An income <5000 Yuan (levels 1–3) was defined as low income/SES, and an income >5000 Yuan (levels 4–6) was defined as high income/SES.

Measures for other constructs (NC, NF, DT, PS, PSE and IA) were based on previous studies with adjustments to fit this specific research context (see Box 1). All items were measured on a seven-point Likert-type scale with anchors from 1 = Strongly disagree to 7 = Strongly agree.

Box 1.

Constructs and items.

Constructs	Items		Literature
NC	NC1	The story is internally consistent.	Adapted from Stutts and Barker (1999)
	NC2	The story has no missing elements or arguments.
	NC3	The behavior of the characters in the story is consistent.
NF	NF1	The story looks true.	Adapted from (Stutts and Barker (1999)
	NF2	The story sounds reasonable.
	NF3	The story employs valid arguments.
	NF4	The story reflects the same values as mine.
	NF5	The idea reflected in the story is correct.
DT	DT1	In general, I trust people who I meet.	Adapted from Moin et al. (2015)
	DT2	I think that people are generally honest in their dealings.
	DT3	I usually trust people until they give me a reason not to trust them.
	DT4	In general, I trust what I read in the newspapers.
	DT5	In general, I trust the news provided by TV channels.
PS	PS1	I have an increased risk of falling ill with the disease described in the story.	Adapted from Nexøe et al. (1999)
	PS2	I am concerned about the risk of falling seriously ill with the disease described in the story.
	PS3	I get sick with the disease described in the story more easily than other people my age.
PSE	PSE1	The disease described in the story may lead to serious health problems.	Adapted from Nexøe et al. (1999)
PSE	PSE2	If I had the disease described in the story, I would not be able to manage daily activities.	Adapted from Nexøe et al. (1999)
IA	IA1	I believe the information in the story.	Adapted from Filieri and McLeay (2014)
	IA2	I intend to closely follow the suggestions given by the story in the future.
	IA3	I will recommend the suggestions given by the story to other people.

NC: narrative coherence; NF: narrative fidelity; DT: dispositional trust; PS: perceived susceptibility; PSE: perceived severity; IA: intention to adopt.

Procedure

Users were asked to imagine a scenario and then answer questions based on that scenario. Each subject was asked to recall a recent health-related text message she/he read that was sent from other buddies on WeChat. The message format was limited to text format because text is the most common format for health-related messages on WeChat; videos or pictures may have different effects compared with text. The message source was limited to his/her buddies rather than established public healthcare organisations because the messages from authoritative sources may be more convincing than non-authoritative sources. Based on the perceptions of that message, each subject was asked to answer some questions regarding NC, NF, IA, among others, using a seven-point Likert-type scale. Demographic information (age, gender, marital status and SES) were also collected. This scenario-based approach, which asked subjects to recall a recently read message and then fill out the questionnaire based on that message, has also been used in other studies, such as Wang et al. (2017), who examined the prediction of instant information sharing on microblogs, where respondents were asked to recall the last time they instantly shared information and answer relevant questions.

Data analysis

We first analysed the reliability and two types of validity for the measurement model (i.e. convergent validity and discriminant validity). Then, we assessed the structural model and tested the hypotheses. Analyses of both the measurement model and the structural model were provided by SmartPLS 2.0, which estimated the parameters in outer and inner models and appropriate when the purpose of the research is exploratory (Chin, 1998).

Results

Demographic characteristics

We obtained a total of 257 valid questionnaires after removing 12 incomplete cases. The response rate was approximately 89.7%. Participants belonged to different age groups and came from different walks of life. Demographic data of participants are presented in Table 1. This sample is likely to be representative of social media users (e.g. young adults are more likely than older adults to use social media, and income is not significantly associated with social media use) (Duggan and Brenner, 2013).

Table 1.

Respondent demographics.

Variables	Values	Frequency	Percentage (%)
Gender	Male	127	49.42
Gender	Female	130	50.58
Age	<25	24	9.34
	25–30	91	35.41
	30–35	74	28.79
	≥35	68	26.46
Income (Chinese Yuan)	<1000	4	1.56
	1000–3000	14	5.45
	3001–5000	70	27.24
	5001–8000	83	32.30
	8000–15,000	68	26.46
	> 15,000	18	7.00
Marital status	Yes	193	75.10
Marital status	No	64	24.90

Measurement model

Reliability is the consistency of a set of measurements, revealing a strong mutual interrelation between two outcomes measured by similar methods of the same construct (Campbell and Fiske, 1959). To assess the reliability of the constructs, we verified the composite reliability (CR), average variance extracted (AVE) and Cronbach’s α (Chin, 1998; Fornell and Larcker, 1981; Wetzels et al., 2009). These measures have been widely used to assess the construct reliability for a survey research (Guo et al., 2013; Li et al., 2018). As shown in Table 2, the CR exceeds 0.86, which is well above the suggested cut-off value of 0.70; AVE is equal to or greater than 0.711, which exceeds the accepted threshold of 0.50 and Cronbach’s α is above the suggested cut-off value of 0.7, indicating good construct reliability (Chin, 1998; Fornell and Larcker, 1981; Wetzels et al., 2009).

Table 2.

Reliability and convergent validity.

Construct	CR	AVE	Cronbach’s α	Item	Loading	T-statistics
NC	0.889	0.728	0.813	NC1	0.87	49.64
				NC2	0.84	35.78
				NC3	0.85	38.35
NF	0.932	0.734	0.909	NF1	0.81	29.01
				NF2	0.86	38.57
				NF3	0.86	40.12
				NF4	0.88	58.26
				NF5	0.87	57.83
DT	0.861	0.553	0.800	DT1	0.76	11.76
				DT2	0.76	13.52
				DT3	0.76	12.69
				DT4	0.72	10.70
				DT5	0.72	11.65
PS	0.930	0.817	0.900	PS1	0.92	72.54
				PS2	0.91	66.05
				PS3	0.88	43.99
PSE	0.877	0.783	0.740	PSE1	0.95	32.07
PSE	0.877	0.783	0.740	PSE2	0.82	8.76
IA	0.951	0.867	0.923	IA1	0.93	81.70
				IA2	0.94	108.74
				IA3	0.93	102.31

NC: narrative coherence; NF: narrative fidelity; DT: dispositional trust; PS: perceived susceptibility; PSE: perceived severity; CR: composite reliability; AVE: average variance extracted; IA: intention to adopt.

Convergent validity relates to the degree to which a scale measuring the same construct provides the same results (O’Leary-Kelly and Vokurka, 1998). Item loadings were examined to assess the convergent validity of the constructs, and a value of less than 0.7 is considered insufficient (Anderson and Gerbing, 1988; Komiak and Benbasat, 2006; McKnight et al., 2002). Item loadings in our research ranged from 0.715 to 0.945, which were higher than 0.70 and in support of convergent validity.

Discriminant validity is the extent to which the measurement differs between two different constructs (McKnight et al., 2002; O’Leary-Kelly and Vokurka, 1998). Previous studies tested discriminant validity by comparing the square root of the AVEs and correlations of this variable with any other model’s constructs (Awad and Krishnan, 2006; Chin, 1998; Wetzels et al., 2009). In Table 3, we determined that all of the square roots of the AVEs were higher than the correlations; therefore, discriminant validity was acceptable. Thus, our measurement model was verified to be reliable.

Table 3.

Discriminant validity.^a

	Coherence	Fidelity	Trust	Severity	Susceptibility	Adopt
Coherence	0.853
Fidelity	0.808	0.857
Trust	0.241	0.315	0.744
Severity	0.237	0.266	0.141	0.885
Susceptibility	0.134	0.302	0.097	0.295	0.904
Adopt	0.692	0.854	0.288	0.246	0.348	0.931

AVE: average variance extracted.

^a Diagonal data refer to the square roots of the AVEs.

Structural model

To examine the impact of NC, NF and SES, we compared the partial least squares (PLS) regression results of four models (as shown in Table 4). Similar to other survey studies using PLS regression, the four models covered control variables (model 1), main effects (model 2) and interaction effects (models 3 and 4) (Li et al., 2018).

Table 4.

PLS results.^a

Variables	Model 1	Model 2	Model 3	Model 4
Age	−0.009 (0.06)	−0.026 (0.03)	−0.016 (0.03)	−0.013 (0.04)
Age
Gender	−0.076 (0.06)	−0.021 (0.03)	−0.023 (0.03)	−0.016 (0.03)
Gender
Marriage	0.075 (0.07)	0.023 (0.04)	0.025 (0.04)	0.025 (0.04)
Marriage
Internet experience	0.015 (0.06)	−0.036 (0.04)	−0.038 (0.03)	−0.031 (0.03)
Internet experience
Online health experience	0.064 (0.06)	0.023 (0.04)	0.031 (0.03)	0.027 (0.03)
Online health experience
DT	0.230^b (0.06)	0.019 (0.03)	0.021 (0.03)	0.016 (0.03)
DT
PS	0.285^b (0.06)	0.104^c (0.04)	0.111^b (0.04)	0.112^c (0.04)
PS
PSE	0.112 (0.07)	−0.008 (0.04)	−0.015 (0.04)	−0.021 (0.04)
PSE
NC		0.032 (0.07)	−0.293^d (0.16)	0.048 (0.07)
NC
NF		0.793^b (0.07)	0.794^b (0.07)	0.445^c (0.20)
NF
Socio-economic status			−0.365^c (0.17)	−0.365^d (0.19)
Socio-economic status
Coherence × Status			0.485^c (0.22)
Coherence × Status
Fidelity × Status				0.490^c (0.24)
Fidelity × Status
Observations	257	257	257	257
R ²	0.223	0.741	0.748	0.749

PLS: partial least squares; NC: narrative coherence; NF: narrative fidelity; DT: dispositional trust; PS: perceived susceptibility; PSE: perceived severity.

^a Standard errors are in parenthesis.

^b p < 0.001.

^c p < 0.01.

^d p < 0.05.

In model 1, only the control variables were included. Results showed that eight factors (i.e. age, gender, marriage, Internet experience, online health experience, DT, PS and PSE) explained 22.3% of the variance of the dependent variable, and DT and PS were found to be significant.

In model 2, NC and NF were further included. Results indicated that the impact of NF on IA was significant (β = 0.793, p < 0.001), lending support to hypothesis 2; however, the effect of NC was found to be non-significant (β = 0.032, p > 0.05). Therefore, hypothesis 1 was not supported. The inclusion of trust increased the R ² value from 0.223 to 0.741, indicating that the proportion of variance explained by IA can be greatly increased when NC and NF are taken into account.

In model 3, the interaction effect of NC and SES was included. Results showed that the interaction effect was significant (β = 0.485, p < 0.01), lending support to hypothesis 3. The inclusion of the interaction effect increased the R ² value to 0.748, thereby increasing the explained proportion of variance for IA. This result suggested that NC and status had a significant interaction effect on the IA the information. This interaction effect is illustrated in Figure 2. When the SES was low, NC had no significant effect on the IA; however, when the status was high, NC had a positive effect on the IA the information.

Figure 2.

Interaction effect between NC and socio-economic status. NC: narrative coherence.

In model 4, the interaction effect of NF and SES was included. Results showed that the interaction effect was significant (β = 0.490, p < 0.01), lending support to hypothesis 4. The inclusion of the interaction effect increased the R ² value to 0.749, thereby increasing the explained proportion of variance for IA. This result suggested that NF and status had a significant interaction effect on the IA the information. This interaction effect is also illustrated in Figure 3. As status increased, NF had a more positive effect on the IA the information.

Figure 3.

Interaction effect between NF and socio-economic status. NF: narrative fidelity.

Discussion

There are several major findings from this study. First, NF had a positive effect on the IA information; the higher the NF, the higher the intention of an individual to adopt social media health information. Second, the effect of NC on the IA social media health information depended on the SES of the user. NC had no effect on the IA for low SES users. In other words, low SES users did not consider the NC when deciding whether to adopt health information from social media. By contrast, NC had a positive effect on the IA for high SES users. This means that NC was an important factor to consider for high SES users when deciding whether to adopt health information from social media. A possible explanation for this finding is that the NC is more difficult or requires more cognitive effort to judge than NF. Judging NC requires the architectural or logical analysis of the whole story. However, judging NF only requires comparing the story with what the user has known, without logical reasoning or an internal consistency check. Since SES has been positively associated with the user’s cognitive performance, judging NC may be a harder task for low SES users. Some low SES users may not have distinguished high fidelity stories from low fidelity stories. Therefore, NF had no effect for the low SES users. Third, the effect of NF on the IA social media health information was higher for high SES users. This suggests that although NF may have been a universal factor to consider for both high status and low status users, it was more predictive of the adoption behavior of high status users than low status users in this study.

Theoretical contribution

This study makes a number of theoretical contributions to the literature. It is the first empirical study to investigate the important and emerging phenomenon of social media health information adoption. Although social media is a very popular medium, and much of the health information that is circulated through social media is incorrect (Moorhead et al., 2013), few studies have focused on how public users (lay people) identify and adopt this information. Previous studies investigating the adoption of online health information have mainly focused on information from search engines or formal websites (Diviani et al., 2015).

Second, this study employed a narrative paradigm perspective to explain the trust and adoption of social media health information. Previous studies on health information adoption have mainly been based on the well-established ELM or health belief model (HBM) frameworks. In this study, we considered the user-generated and storytelling nature of social media health messages based on a narrative paradigm perspective rather than an ELM or HBM perspective.

Third, this study has extended the traditional narrative paradigm theory by incorporating SES as a moderating variable. Our results have indicated that narrative paradigm theory was more predictive for high status users than low status users. In particular, we found that NC had no effect on low status users but had a positive effect on high status users.

Implications

This study has several practical implications. First, it has demonstrated that the adoption of health messages from social media could be well explained by the narrative paradigm rather than the traditional rational world paradigm. The rational world paradigm, which is rooted in the sciences, holds that humans are essentially rational beings and goes on to explain the reasoning behind this assumption. The narrative paradigm presents the alternative humanistic view, which goes one step further and suggests that humans are essentially storytellers. The results of this study suggest that creating a good story in health communication is important, maybe even more so than providing logical, scientific arguments or statistical figures.

Second, results from this study have provided some guidance as to how to create and influence health messages on social media. For example, study findings highlighted the importance and universal effect of NF; that health message narrations about the same topic from different sources should be consistent or users will perceive contradictions that may hinder their adoption intentions. In addition, if messages were to come from multiples sources and users could see it several times, the message would be more convincing. Results of this study also suggest that health messages on social media should be designed separately for high SES users and low SES users. For example, we may send a lengthy but highly coherent message to high SES users and a shorter more simple message to low SES users.

Limitations

Participants were asked to recall a recent health-related message that they had read on social media. However, we did not control for how recently the messages had been read, which posed a limitation to this research and to other similar studies (Wang et al., 2017). A message read in the past 24 hours may be recalled differently than a message read a month ago. We also did not fully control for the message source of the recalled message. Some message sources (e.g. good friends) may be more convincing than others. This is another common limitation that has seldom been addressed in similar studies (Wang et al., 2017). Future studies could use defined health messages delivered on social media rather than messages recalled by participants to increase validity of results.

Conclusion

In this study, we investigated important and emerging RQs, such as “how do users adopt the health information from social media?” Considering the user-generated and storytelling nature of social media messages, this research employed the narrative paradigm perspective to explain the social media health information adoption phenomenon. Specifically, NC and NF were hypothesised to have a positive effect on the IA. Additionally, SES was viewed as a proxy variable for cognitive capability and hypothesised to moderate the effects of NC and NF. A scenario-based survey was conducted to test the proposed research model. Results indicated that NF had a positive effect on the IA social media health information. NC had no impact on low SES users but a positive impact on high SES users. The effect of NF on the IA was higher for high SES users than low SES users.

Footnotes

Declaration of conflicting interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work was supported by the National Natural Science Foundation of China under grants (71371005, 71471064 and 91646205).

ORCID iD

Xuan Liu, PhD

Appendix 1

References

Adams

(2010) Revisiting the online health information reliability debate in the wake of “web 2.0”: an inter-disciplinary literature and website review. International Journal of Medical Informatics 79: 391–400.

Anderson

Gerbing

(1988) Structural equation modeling in practice: a review and recommended two-step approach. Psychological Bulletin 103: 411–423.

Awad

Krishnan

(2006) The personalization privacy paradox: an empirical evaluation of information transparency and the willingness to be profiled online for personalization. MIS Quarterly 30: 13–28.

Barry

Markey

Almquist

, et al. (2011) Putting social media to work. Available at: http://www.bain.com/publications/articles/putting-social-media-to-work.aspx (Accessed 28 August 2018).

Brinson

Brown

(1997) The AIDS risk narrative in the 1994 CDC campaign. Journal of Health Communication 2: 101–112.

Brodie

Foehr

Rideout

, et al. (2001) Communicating health information through the entertainment media. Health Affairs 20: 192–199.

Campbell

Fiske

(1959) Convergent and discriminant validation by the multitrait-multimethod matrix. Psychological bulletin 56: 81.

Chin

(1998) Issues and opinion on structural equation modeling. MIS Quarterly 22: 1–14.

Dahlstrom

(2014) Using narratives and storytelling to communicate science with nonexpert audiences. Proceedings of the National Academy of Sciences 111: 13614–13620.

10.

De Choudhury

Morris

White

(2014) Seeking and sharing health information online: comparing search engines and social media. Proceedings of the 32nd annual ACM conference on human factors in computing systems (ed Schraefel

). ACM, pp. 1365–1376.

11.

Diviani

van den Putte

Giani

, et al. (2015) Low health literacy and evaluation of online health information: a systematic review of the literature. Journal of Medical Internet Research 17: e112.

12.

Dowell

(2003) Uncovering the Storytelling Power of Communication: Applying Walter Fisher’s Narrative Theory to Three Presidential Crisis Speeches. Omaha: University of Nebraska at Omaha.

13.

Duggan

Brenner

(2013) The Demographics of Social Media Users, 2012. Washington: Pew Research Center’s Internet & American Life Project.

14.

Eastin

(2001) Credibility assessments of online health information: the effects of source expertise and knowledge of content. Journal of Computer-Mediated Communication 6: 0–0.

15.

Filieri

McLeay

(2014) E-WOM and accommodation: an analysis of the factors that influence travelers’ adoption of information from online reviews. Journal of Travel Research 53: 44–57.

16.

Fisher

(1984) Narration as a human communication paradigm: the case of public moral argument. Communications Monographs 51: 1–22.

17.

Fisher

(1985) The narrative paradigm: an elaboration. Communications Monographs 52: 347–367.

18.

Fisher

(1987) Human Communication as Narration: Toward a Philosophy of Reason, Value, and Action. Columbia: University of South Carolina Press.

19.

Fornell

Larcker

(1981) Structural equation models with unobservable variables and measurement error: algebra and statistics. Journal of Marketing research 18: 382–388.

20.

Fox

Jones

(2011) The social life of health information. Available at: http://www.pewinternet.org/2011/05/12/the-social-life-of-health-information-2011/ (accessed 28 August 2018).

21.

Gibson

Zillmann

(1994) Exaggerated versus representative exemplification in news reports: perception of issues and personal consequences. Communication Research 21: 603–624.

22.

Green

Brock

(2000) The role of transportation in the persuasiveness of public narratives. Journal of Personality and Social Psychology 79: 701.

23.

Griffin

Crossman

Bordia

, et al. (2009) A First Look at Communication Theory. 7th ed. Boston: McGraw-Hill.

24.

Griliches

Mason

(1972) Education, income, and ability. Journal of Political Economy 80: S74–S103.

25.

Guo

Sun

Wang

, et al. (2013) The dark side of elderly acceptance of preventive mobile health services in China. Electronic Markets 23: 49–61.

26.

Hackman

Farah

(2009) Socioeconomic status and the developing brain. Trends in Cognitive Sciences 13: 65–73.

27.

Jafarkarimi

Sim

ATH

Saadatdoost

, et al. (2016) Designing a scenario-based questionnaire to assess behavioral intention in social networking sites’ ethical dilemmas. Blurring the Boundaries Through Digital Innovation 19: 145–159.

28.

Komiak

Benbasat

(2006) The effects of personalization and familiarity on trust and adoption of recommendation agents. MIS Quarterly 30: 941–960.

29.

Liu

, et al. (2018) Why and when do patients use e-consultation services? The trust and resource supplementary perspectives. Telemedicine and E-health 24: 77–85.

30.

Lin

Spence

Lachlan

(2016) Social media and credibility indicators: the effect of influence cues. Computers in Human Behavior 63: 264–271.

31.

Linkfluence. Chinese Social Media: A Marketing Masterclass. Available at: https://linkfluence.com/chinese-social-media-marketing/ (accessed 10 June 2018).

32.

Lumpkins

(2012) Hollywood and breast cancer prevention among African American women: An analysis of breast cancer risk narratives in the California’ Every Woman Counts Campaign’. Journal of Media and Communication Studies 4: 95–106.

33.

McKnight

Choudhury

Kacmar

(2002) Developing and validating trust measures for e-commerce: an integrative typology. Information Systems Research 13: 334–359.

34.

Moin

Devlin

McKechnie

(2015) Trust in financial services: impact of institutional trust and dispositional trust on trusting belief. Journal of Financial Services Marketing 20: 91–106.

35.

Moorhead

Hazlett

Harrison

, et al. (2013) A new dimension of health care: systematic review of the uses, benefits, and limitations of social media for health communication. Journal of Medical Internet Research 15: e85.

36.

Mun

Yoon

Davis

, et al. (2013) Untangling the antecedents of initial trust in Web-based health information: the roles of argument quality, source expertise, and user perceptions of information quality and risk. Decision Support Systems 55: 284–295.

37.

Nexøe

Kragstrup

Søgaard

(1999) Decision on influenza vaccination among the elderly: a questionnaire study based on the health belief model and the multidimensional locus of control theory. Scandinavian Journal of Primary Health Care 17: 105–110.

38.

Noble

Norman

Farah

(2005) Neurocognitive correlates of socioeconomic status in kindergarten children. Developmental Science 8: 74–87.

39.

O’Leary-Kelly

Vokurka

(1998) The empirical assessment of construct validity. Journal of Operations Management 16: 387–405.

40.

Olsen

Reynolds

(1999) Exploring the relationship between perceived narrativity and persuasiveness. Carolinas Communication Annual 15: 41–54.

41.

Perez

Kravitz

Bell

, et al. (2016) Characterizing internet health information seeking strategies by socioeconomic status: a mixed methods approach. BMC Medical Informatics and Decision Making 16: 107.

42.

Petty

Cacioppo

(1984) Source factors and the elaboration likelihood model of persuasion. In: Kinnear

(ed) NA-Advances in Consumer Research Volume 11. Provo: Association for Consumer Research, pp. 668–672.

43.

Petty

Cacioppo

(1986) Communication and Persuasion: Central and Peripheral Routes to Attitude Change. New York: Springer-Verlag.

44.

Slater

Rouner

(2002) Entertainment—education and elaboration likelihood: understanding the processing of narrative persuasion. Communication Theory 12: 173–191.

45.

Statista. Social Networks in China - Statistics & Facts. Available at: https://www.statista.com/topics/1170/social-networks-in-china/ (accessed 28 August 2018).

46.

Stutts

Barker

(1999) The use of narrative paradigm theory in assessing audience value conflict in image advertising. Management Communication Quarterly 13: 209–244.

47.

Vaughan

Dunton

(2007) Difficult socio-economic circumstances and the utilization of risk information: a study of Mexican agricultural workers in the USA. Health, Risk and Society 9: 323–341.

48.

Vaughan

Rogers

Singhal

, et al. (2000) Entertainment-education and HIV/AIDS prevention: a field experiment in Tanzania. Journal of Health Communication 5: 81–100.

49.

Wang

Zhou

Jin

, et al. (2017) The influence of affective cues on positive emotion in predicting instant information sharing on microblogs: gender as a moderator. Information Processing and Management 53: 721–734.

50.

Wang

Arpan

(2008) Effects of race and ethnic identity on audience evaluation of HIV public service announcements. The Howard Journal of Communications 19: 44–63.

51.

Weitzman

Cole

Kaci

, et al. (2011) Social but safe? Quality and safety of diabetes-related online social networks. Journal of the American Medical Informatics Association 18: 292–297.

52.

Wetzels

Odekerken-Schröder

Van Oppen

(2009) Using PLS path modeling for assessing hierarchical construct models: Guidelines and empirical illustration. MIS Quarterly 33: 177–195.

53.

Williams

Jackson

, et al. (1997) Racial differences in physical and mental health: socio-economic status, stress and discrimination. Journal of Health Psychology 2: 335–351.

54.

Xiao

Sharman

Rao

, et al. (2014) Factors influencing online health information search: an empirical analysis of a national cancer-related survey. Decision Support Systems 57: 417–427.

55.

Zhang

Wen

Liang

, et al. (2017) How the public uses social media WeChat to obtain health information in china: a survey study. BMC Medical Informatics and Decision Making 17: 66.