Introduction
During the COVID-19 pandemic, the rampant spread of misinformation on social media has become a major concern for health authorities and healthcare providers. Government health agencies and social media platforms are stepping up their efforts to fight this problem. In the United States, lawmakers have proposed legislation to hold social media companies responsible for spreading false vaccine information (McCabe, 2021). Simultaneously, major social media platforms are taking voluntary measures to counter false claims about COVID-19 vaccines, such as removing misinformation and banning accounts of known anti-vaccine activists.
However, concerns have arisen about social media platforms’ ability to effectively moderate the constant flow of misinformation within users’ decentralized networks (Koo et al., 2021). Researchers increasingly emphasize the importance of involving regular social media users in combating misinformation (Barnidge & Rojas, 2014; Koo et al., 2021). Two approaches to countering misinformation by social media users have garnered significant attention.
The first approach involves supporting government and social media platform censorship. As it touches upon constitutionally protected freedom of speech, opinions on this regulatory approach vary among individuals from diverse demographics and socioeconomic statuses (Lazer et al., 2018). The second approach involves taking corrective actions against misinformation on social media, such as reporting to the platform and leaving comments to refute false information (Luo & Cheng, 2021; Wang & Kim, 2020; Sun et al., 2022). These corrective actions by social media users are effective to combat misinformation because they are visible to others within their interpersonal connections on social networking sites. This increased visibility enhances the likelihood of raising awareness through engagement behaviors like liking and sharing, influencing others to take similar corrective actions.
Grounded in the influence of presumed influence (IPI) theory (Gunther and Storey, 2003), this study aims to investigate how individuals, influenced by their perceived others’ exposure to social media misinformation, differ in their behavioral intentions to support censorship and engage in corrective actions against health misinformation. Despite extensive research on IPI, only a limited number of studies have explored the impact of IPI on intentions to counter the spread of health misinformation. Moreover, some of the existing studies have primarily focused on examining IPI by considering only its theoretical components, with limited exploration of its effects based on individual characteristics. This study places particular emphasis on examining the potential moderating role of key demographic variables, such as gender, age, and race/ethnicity, in predicting perceived influence on others and engagement in restrictive and rectifying behaviors. We believe these findings will offer valuable insights for policy development and implementation by both governmental bodies and social media platforms.
Theoretical Background
The Influence of Presumed Influence
As Tewksbury et al. (2004, p. 140) state, “how one responds to a message depends largely on what the message is thought to do to others.” The Influence of Presumed Influence (IPI) (Gunther & Storey, 2003) is a mass communication theory that explains the indirect effects of persuasive media messages on people’s attitudes and behaviors. Expanding on the theory of the third-person effect (TPE), IPI provides a theoretical basis for the effect of individuals’ presumed others’ exposure to persuasive media messages on their own attitudes and behaviors through their presumed influence of such messages on others. The IPI theory assumes that media content can have a dual impact on individuals’ perceptions and behaviors. This impact of persuasive messages occurs through the messages’ direct shaping of one’s presumption about others’ exposure to the message based on their own exposure and the indirect influence on one’s own behavior as they align their behaviors with the presumed influence on others (Cho et al., 2021).
Perceived Exposure and Presumed Influence of Misinformation on Others
IPI posits that perceptions of others’ exposure play a crucial role in shaping individuals’ views regarding media’ s impact on others. This relationship is rooted in the persuasive press inference hypothesis and the persuasive reach presumption (Gunther, 1998). These hypotheses state that individuals form opinions about public views by interpreting media content and believe others consume the same media (Cho et al., 2021). Simply put, when individuals engage with media content and formulate impressions thereof, they frequently make the assumption that this content mirrors the exposure of the broader public and holds significant reach. Consequently, they develop certain expectations concerning others’ exposure to the content and the extent to which others are influenced by it (Gunther & Storey, 2003). The presumed exposure by others functions as an intermediary factor between an individual’s own exposure and their presumed influence on others (Cho et al., 2021; Shen & Huggins, 2013).
Several studies (e.g., Eveland et al., 1999; Hong, 2023; Paek et al., 2011; Shen and Huggins, 2013) have examined the correlation between perceived exposure by others and presumed influence on others within both mass media and social media contexts. Eveland et al. (1999) demonstrated that the perceived likelihood of exposure to socially undesirable content, such as violent and misogynic rap music, strongly predicted presumed influence on others. They suggested that this causal relationship might be linked to the naive, powerful media effect model that other people will be strongly influenced by negative media content. In testing the potential question order effects in IPI studies, Shen and Huggins’s (2013) study indicated that when self-related questions are asked before other-related questions in the questionnaire, a positive correlation can be observed between perceived exposure of others and the presumed influence on them. They confirmed that IPI worked in a causal chain: self-exposure → other-exposure → presumed influence on others → behavior.
This relationship between self-exposure and other-exposure receives mixed findings in the realm of social media. A study conducted by Cho et al. (2021) concerning E-cigarette commercials on YouTube indicated a strong positive association between presumed exposure by others and the perceived influence on others. However, Lim and Golan (2011) did not find support for their hypothesis that participants would perceive a greater influence of the political parody video on others when the video was assumed to have a higher likelihood of exposure. This divergence might be attributed to their treatment of the perceived likelihood of exposure as a dichotomous variable and their comparison of two exposure conditions—high versus low exposure. Based on the IPI’s theoretical premise and related research, we propose the following hypothesis about the relationship between perceived others’ exposure and presumed influence on others.
Behavioral Outcomes: Support for Censorship and Corrective Actions
Another notable difference between IPI and TPE is its potential to expand beyond the traditional behavioral outcomes of the theory. While TPE has typically focused on two types of behavioral responses—restrictive or corrective actions (Sun et al., 2008), IPI has the capacity to broaden the scope of behavioral outcomes to include other persuasive effects. For example, Gunther and Storey (2003) found that a radio-based health campaign in Nepal indirectly influenced listeners’ attitudes and interactions with health workers through their presumed influence on others. Similarly, Hong (2023) examined the indirect impact of PrEP (pre-exposure prophylaxis) exposure on intentions to seek HIV risk information and promote PrEP education, mediated by presumed influence on others and presumed exposure by others. However, within the context of negative media messages, previous IPI research has typically focused on investigating restrictive and corrective actions.
Previous IPI research consistently shows a positive correlation between perceived media influence on others and support for censorship (Baek et al., 2019; Wang & Kim, 2020; Cohen & Weimann, 2008; Dohle et al., 2017; Tal-Or et al., 2010; Gunther, 1995; Hoffner et al., 1999; Riedl et al., 2022). For instance, Riedl et al. (2022) surveyed U.S. internet users and found that the perceived effect of social media content on others strongly predicted support for content moderation. Chung and Moon (2016) also found that presumed influence on others is a more robust predictor of attitudes toward media censorship than the self-other perceptual gap. Chung’s (2023) study, based on four national surveys in the U.S., UK, South Korea, and Mexico, found a positive association between the presumed influence of misinformation on others and support for content moderation, with the exception of the Mexican sample. Likewise, Luo and Cheng (2021), in their research involving American and Chinese populations, found that presumed influence of COVID-19 misinformation on others significantly increased public endorsement of social media censorship.
Based on this theoretical background, we propose the following hypothesis:
Recognizing the limitation of restrictive action in implementation, researchers of TPE and IPI have paid their special attention to the potential of corrective action. In particular, Rojas (2010) proposed the corrective action hypothesis, drawing insight from the anecdotal insight in Davison’s seminal work— “some counteraction would have to be taken” (Davison, 1983, p. 2). The corrective action hypothesis is particularly gaining importance in the age of social media, where the audience of a negative persuasive message is no longer a passive recipient and actively reacting to the message to counterbalance the potential negative impact that the message may have on other people and society. The corrective action hypothesis suggests an alternative avenue wherein individuals engage in reactive actions to voice their viewpoints and counterbalance the perceived media effects (Barnidge & Rojas, 2014; Lim & Golan, 2011; Rojas, 2010; Sun et al., 2022; Wang & Kim, 2020).
The corrective action hypothesis is underpinned by three core theoretical assumptions. Firstly, it posits that mass media are perceived as wielding a disproportionately strong influence over other people, as well as public opinion. Secondly, it suggests that certain individuals are predisposed to take corrective measures when they encounter media messages that they perceive as being biased or hostile towards their in group members. Thirdly, it asserts that those who perceive the media as exhibiting bias or hostility are driven by a strong intention to counterbalance the media’s influence through active engagement in both online and offline expressive behaviors (Barnidge & Rojas, 2014; Rojas, 2010; Wintterlin et al., 2021). Previous studies have endeavored to investigate a range of corrective measures against perceived media biases. These measures include activities such as engaging in political talk (Barnidge & Rojas, 2014), participating in social media activism (Lim & Golan, 2011), and conducting content moderation (Wang & Kim, 2020; Wintterlin et al., 2021).
The corrective action hypothesis has broadened the range of potential behavioral outcomes (Rojas, 2010; Barnidge & Rojas, 2014). A few studies have tested the corrective action hypothesis by applying it to biased, misleading, or hostile social media effects (Golan & Lim, 2016; Wang & Kim, 2020; Wintterlin et al., 2021). However, the empirical support for the corrective action hypothesis has not been consistently established across these studies. Barnidge and Rojas (2014) found that the presumed influence of biased media led people to engage more frequently in political talk in order to address perceived biases those perceived biases. Golan and Lim (2016) have examined the people’s intention to counterbalance the potential social influence of ISIS recruitment messages by engaging in social media activism. Wang and Kim (2020) examined people’s intention to request the platform’s moderation of uncivil comments. Additionally, Wintterlin and his colleagues (2021) examined social media users’ intention to correct misinformation or fight harmful content by using Facebook’s features, such as reporting to the platform or responding directly with comments or dislikes. However, they did not find supporting evidence for the corrective action hypothesis. Naab et al.’s (2021) research investigated intentions to take corrective actions against rude comments on a news story about refugees posted on a mock Facebook page of Germany’s Spiegel and failed to link the TPP of uncivil comments to corrective actions. Caution must be taken when interpreting the results, as the study was based on a single experiment without replication and lacks external validity due to the reaction to a single rude comment. Sun et al. (2022) found that individuals who perceive a high threat of misinformation influence on others experience emotions like anticipated guilt, which motivated them to correct the misinformation. We posit the following hypothesis based on previous studies that reported the direct effect of influence on others on corrective actions.
H3. Presumed influence on others positively correlates with the intention to engage in corrective actions.
As mentioned earlier, a couple of studies did not find statistically meaningful results for the corrective action hypothesis. Lim (2017), however, argued that the correlation tables of those studies showed positive correlations among TPP, censorship, and corrective actions. He further argued that the fact that the significant relations were no longer significant in the multivariate analysis implies full mediation through censorship. As suggested, he found the result that supported the mediation hypothesis in which TPP of misleading online advertising of cosmetic surgery led to corrective actions through support for regulation. Cheng and Luo reported the results that showed the significant indirect effect of TPP of misinformation about COVID-19 on corrective actions through support for government regulation. However, a closer look at their measure of corrective actions indicates that their measure of corrective actions is more related to attitudes toward media literacy than corrective actions implied by Davison (1983) and Rojas (Rojas, 2010). Based on the theoretical assumption and supporting evidence from previous research (Lim, 2017), we predict as follows:
Role of Demographics in Presumed Influence on Others
Examining individuals’ demographic characteristics, such as age, gender, and ethnicity, is essential for understanding potential differences in the impacts of IPI across various groups. Different demographic segments can exhibit differing levels of susceptibility to specific forms of media influence or misinformation. By recognizing these distinctions, policymakers and public health professionals can more precisely tailor their interventions and allocate resources to effectively address the distinct needs and vulnerabilities of each group. Moreover, exploring the role of demographics in IPI could assist public health professionals in strategically allocating resources to combat misinformation. Additionally, social media platforms could enhance the effectiveness of their policies and incentives in addressing misinformation by considering the demographic composition of their user base.
While the impact of demographics on persuasive media effects has been acknowledged by scholars, these factors have mainly served as control variables within analyses (e.g., Tewksbury et al., 2004; Wang & Kim, 2020; Sun, 2022, Wei et al., 2017). Only a few scholars have attempted to examine these potential differences resulting from gender, age, and ethnicity (Loomba et al, 2021; Luo & Cheng, 2022; Lo & Wei, 2002, Jang & Kim, 2018; David et al., 2008; Riedl et al., 2022). For example, Lo and Wei (2002) examined the differences in the perceptual gap related to internal pornography between males and females. Their findings indicated that female students exhibited stronger intentions to support restricting pornography on the Internet. Regarding age, Riedl et al., (2022) found that age was positively related to support for social media content review, suggesting that older individuals tend to endorse restrictive measures on the media content. Luo and Cheng (2022) found a distinct pattern wherein age exhibited a positive correlation with censorship support while displaying a negative association with corrective actions. This outcome suggests that young people tend to oppose government intervention or censorship of COVID-19 misinformation. Instead, they seem to lean towards engaging in self-directed measures to address misinformation issues.
Based on the previous studies, we propose the following two research questions.
Method
Survey respondents
Survey respondents were randomly selected from the U.S. national panel of Qualtrics, a leading survey management platform. Each respondent received $5 for completing the survey. IRB approval was obtained from a large private university in the United States. The survey was conducted in March 2021, about a year after the start of the COVID-19 pandemic, when J&J’s COVID-19 vaccine had just been rolled out, and some U.S. adults had received an initial dose of the Pfizer or Moderna vaccines. The final sample included 1,030 respondents: 60.4% were White, 18.7% Hispanic or Latino, 12.6% African American, and 6.3% Asian American. Female respondents made up 49.7%, and male respondents accounted for 49.2%. The mean age was 33.22 (SD = 10.47). Participants were divided into three age groups by generation: 27.5% were 18-24 years old (Generation Z), 46.0% were 25-40 years old (Millennials), and 26.5% were 41 years and older (Generation X and Boomers). Table 1 summarizes the demographic information. of the participants.
Measures
Before measuring the key variables of interest, we provided respondents with nine statements regarding COVID-19 vaccine-related misinformation and asked them to indicate the extent to which they believe each claim reflects what they truly believe on a 7-point Likert scale, anchored by “Very untrue of what I believe” (1) and “Very true of what I believe” (7). Samples of the claims include: COVID-19 vaccines cause autism; COVID-19 vaccines change people’s DNA; COVID-19 vaccines are designed to control a population for non-public health purposes. These statements contained false and misleading information about the COVID-19 vaccine, which was sourced from authoritative public health websites such as the WHO. Importantly, we refrained from informing respondents that these statements were actually misinformation. This approach was taken to prevent potential measurement errors, including social-desirability bias and the influence of leading questions. Table 2 presents all measures, means, SDs, and reliability for each measure used in the study. The composite score for each measure was calculated by averaging the total score of all items for that specific measure.
Perceived others’ exposure (OTHER-EXP). Respondents were asked, “Please estimate how often you think other people, in general, have encountered the previous claims regarding the COVID-19 vaccines from each of the following sources: Twitter, Facebook, YouTube, Digital Content Linked via other social media.” For each source, participants were asked to estimate the frequency of others’ exposure to the claims they read on a 7-point scale (1: Never, 7: Very often) (M = 4.48, SD = 1.67, α = .89).
Presumed influence on others (OTHER-INF). To measure respondents’ presumed influence of COVID-19 vaccine-related on others, we adapted five items from Lim et al.’s (2020) research. The respondents indicated their agreement with the four statements on a 7-point Likert scale (1: strongly disagree, 7: strongly agree) (M = 4.78, SD = 1.45, α = .89).
Support for Censorship (CENSOR). Respondents were asked to indicate how much they support social media platforms or lawmakers to regulate the claims they read regarding the COVID-19 vaccines. We adapted six items from Guo and Johnson’s (2020) study. All items were measured on a 7-point Likert scale (1: strongly disagree, 7: strongly agree) (M = 3.46, SD = 1.06, α = .91).
Corrective actions (CORRECT). Respondents were asked to report their willingness to engage in corrective actions regarding COVID-19 vaccine-related misinformation on social media. We adopted a five-item 7-point scale (1: unwilling at all 7: very willing) from Lim et al.’s (2020) research (M = 4.40, SD = 1.71, α = .93).
Results
Testing H1 to H4
The current study predicted that perceived others’ exposure would positively predict presumed influence on others (H1) and that presumed influence on others would positively predict support for censorship (H2) and corrective action (H3). We also proposed that the effect of presumed influence on others on corrective action would be mediated by support for censorship (H4) (Figure 1). To test these hypotheses, we performed a sequential mediation analysis using Mplus 8. Table 3 summarizes the results of the mediation analysis.
Answering RQ1 and RQ2
We proposed RQ1 and RQ2 to investigate potential variations in the impact of IPI across different demographic groups. To answer them, the gender variable was recoded into a dichotomous classification: male and female. Due to limited numbers, other gender identifications were treated as missing data. Similarly, the ethnicity groups were recoded as a dichotomous variable: White and non-White. Age was recoded into three groups: Generation Z (18-24 years old), Millennials (25-40 years old), and Generation X and older (41 years old and above). A multiple-group path analysis was employed to examine whether differences in the structural parameters across groups were statistically significant. Cross-group invariance was assessed through a comparison of two nested models using the Chi-square test: the baseline model wherein no constraints were specified and the constrained model where each path was constrained to be invariant for the comparison paths between the two groups. Table 4 displays the results of the nested model comparison.
The results indicated that several relationships were different by gender. First, the positive effect of perceived exposure by others on presumed influence was significantly stronger (p <.001) for males (β = .629, p <.001) than for females (β = .422, p <.001). Second, the differences in the positive effects of presumed influence on support for censorship for male (β = .564, p <.001) and female (β = .335, p <.001) was also significant (p <.001). Third, the positive effects of presumed influence on corrective action were significantly higher (p <.001) for males (β = .355, p <.001) than females (β = .137, p <.001).
Regarding the ethnicity groups, we focused on the comparison between White respondents and Non-White respondents. Our findings indicated that in comparison to Non-White respondents (β = .409, p <.001), the effect of presumed influence on censorship among White respondents (β = .478, p <.001) was significantly stronger (p =.020).
In examining three age groups-Gen Zers (18-24), Millennials (25-40), and Xers and older (41 and above), we found significant differences in the effects of exposure on presumed influence (p < 0.001). Specifically, the influence of perceived exposure on presumed influence was more pronounced for Millennials (β = .599, p <.001) than Gen Zers (β = .495, p <.001) and Xers (β = .432, p <.001).
Discussion
The Grounded in a theoretical premise of IPI, we proposed three hypotheses that assume the pathways of OTHER-EXP → OTHER-INF → BEHAVIOR (i.e., CENSOR, CORRECT). Our findings support the positive correlations among these variables suggested by Shen and Huggins (2013). We observed the large effect of OTHER-EXP on OTHER-INF and a positive direct effect of OTHER-INF on both CENSOR and CORRECT. The positive impact of OTHER-INF on CENSOR is well-compatible with the results of other studies (Cohen & Weimann, 2008; Rojas et al., 1996). In the context of misinformation-related censorship, our research can be summarized as follows: The more people perceive the influence of misinformation on others, the stronger their intention to support the platform’s banning of misinformation and to support legislation to stop its spread.
The current research also found a statistically significant indirect effect of OTHER-EXP on CORRECT through OTHER-INF and CENSOR. This result further confirms the mediating role of support for censorship in corrective action that has been found in previous studies (Cheng & Luo, 2021; Lim, 2017). Identifying the conditions that affect people’s corrective actions is a key to developing an effective communication strategy to encourage their voluntary engagement. The idea of inducing corrective actions through presumed influence on others is often implemented in persuasive messages that employ descriptive norms in campaigns for health prevention (Hong & Kim, 2020) and environmental conservation (Cialdini, 2003).
We also observed age-related effect on the relationship between OTHER-EXP and OTHER-INF. Specifically, the perception of other’s exposure to COVID-19 vaccine misinformation had a stronger impact on Millennials’ (25-40) presumed influence of misinformation on others than Gen Zers (18-24). This finding aligned with the outcomes observed by Martínez-Costa et al. (2023), which showed individuals’ tendency to underestimate the ability of others who are not in their same age group in identifying disinformation. In our study, respondents within the middle-aged category exhibited a higher level of bias. However, the presumed influence on others did not demonstrate age-related differential effects on individuals’ intention to support censorship and adopt corrective measures.
Regarding different racial/ethnic groups, our study found that the perceived influence of misinformation on others had a stronger impact on the intention of White respondents to endorse the censorship of misinformation compared to people of color. However, this perceived influence did not generate significantly different effects on White respondents’ intentions to take actions against misinformation compared to people of color. Furthermore, we did not observe a varying impact of OTHER-EXP on OTHER-INF across different racial/ethnic groups. These findings suggested that individuals from diverse race/ethnicity groups did not perceive others to be more susceptible to the influence of exposing to COVID-19 vaccine misinformation. However, White respondents were more inclined to support censorship regarding misinformation due to their presumptions about the influence of such information on others.
In summary, our findings demonstrated that the effects induced by the presumed influence vary significantly across crucial demographic variables, contributing to the existing literature. This finding supported Davison’s assertion that hypothesized influence on others is a complex phenomenon influenced by diverse psychological, personal, and situational factors. Prior efforts to explain these effects stemming from individual differences have relied on theoretical frameworks such as the self-enhancement bias (see Tal-Or, 2007) and overconfidence (see Martínez-Costa et al., 2023). Individuals often hold the belief that their abilities are better than average or others, or that they are better than they really are (Martínez-Costa et al., 2023). For instance, Martínez-Costa et al. (2023) introduced and investigated the “nobody-fools-me-perception,” delineating a cognitive bias characterized by overconfidence in one’s ability to detect disinformation. Likewise, Wang and Kim (2020) found that individuals perceived themselves as more adept than others in distinguishing true and false COVID-19 related statements. Our study also implied the presence of an overconfidence bias due to the mismatch between individuals’ perceptions and their reality. We performed a post-hoc assessment of COVID-19 vaccine-related knowledge based on ten yes-no questions. The results showed that women scored higher (M = 3.4, SD = 1.92) compared to men (M = 2.96, SD = 1.65, t =3.91, df =1017). While female respondents demonstrated a greater level of knowledge about COVID-19 vaccine than male respondents, the latter group still exhibited a stronger inclination to believe that exposure to COVID-19 vaccine misinformation significantly impacted others and consequently felt more inclined to take action against misinformation. Understanding the moderating roles of these demographic factors in the IPI effect can facilitate public health agencies and social media platforms to devise appropriate measures and policies to encourage specific populations to participate in combatting misinformation.
Limitations and Suggestions for Future Research
The current research has some merits of using nationally representative samples. To ensure representativeness, a random sample was drawn from the national panelists registered in Qualtrics. Furthermore, to minimize the social-desirability bias, this study avoided the use of the term “misinformation” in the questionnaire.
Despite all the efforts, this research still has some limitations. First, the IPI model proposed in this study did not consider other behavioral intentions (e.g., vaccine uptake intentions), which are important to health communication researchers. In addition, we only attempted to see if demographics of gender, age, and race/ethnicity would make differences for the proposed IPI model. Other factors, including prior vaccination experience, vaccine-related knowledge, perceived severity and susceptibility to disease, and emotional responses, may also influence individuals’ presumed influence on others by COVID-19 vaccine-related messages. Therefore, we recommend that future research expand the current IPI model to incorporate more factors and evaluate both vaccine uptake intentions and corrective actions.
Second, this research leaves an open question regarding the presumed influence of misinformation on others, as estimated by expert groups. We argue that understanding how experts perceive this influence is crucial for exploring the censorship-related aspects of the theory and for deriving meaningful policy implications. To address this gap, we recommend that future research investigate experts’ support for the censorship of vaccine misinformation on social media.
Third, we did not consider political or ideological affiliation, but respondents may have different perceptions of influences and behaviors toward the presented false claims depending on their political orientation.
PDF Links
PubReader
ePub Link
XML Download
Full text via DOI
Download Citation
Print
