Abstract
The novel affordances and unique features on social media have transformed the way people assess public opinion. Drawing on the spiral of silence (SOS) theory, this study examines the roles that user-generated comments (UGCs) and aggregated user representations (AURs), represented by reaction emojis, play in shaping perceptions of the opinion climate. It also investigates how features on the platform trigger perceptions of source credibility to influence willingness to speak out. Results from a 3 (opinion cues: UGCs-only vs AURs-only vs UGCs and AURs) × 2 (opinion climate congruency: congruent vs incongruent) × 2 (source credibility: high vs low) between-subjects experimental design revealed that audiences perceived reaction emojis to reflect public opinion. Source credibility was found to affect willingness to speak out and moderate opinion climate congruency. The findings extend the SOS theory by highlighting the significant role that novel affordances have on SOS components operating online. Implications of the findings were discussed.
Keywords
The dominant role that the mass media play in shaping perceptions of public opinion has been widely acknowledged in media research. Audiences are constantly bombarded with consonant messages from the ubiquitous mass media, which creates the perception that media messages are reflective of the broader opinion landscape. Drawing from the spiral of silence (SOS) theory, the mass media establish the dominant position by influencing audiences’ perceptions of the opinion climate (Noelle-Neumann, 1974). This demonstrates the vital role that the mass media play in the public opinion formation process.
With the proliferation of social media sites, there is a renewed interest to examine how public opinion manifests in the online realm. The shift in focus from traditional media channels to new media channels can be attributed to the prominent role that the latter plays in presenting the opinion climate in the contemporary media environment. Notably, Facebook has solidified its status as the leading and most influential social media site, with a reported 2 billion active users monthly (Fiegerman, 2017). Given Facebook’s extensive reach and widespread influence, it is vital to examine how public opinion manifests on the platform.
The development of opinion cues such as comments and aggregated user representations (AURs) has spurred research into the effects of these cues on perceptions of the online opinion climate (Lee and Jang, 2010; Neubaum and Krämer, 2017). Although comments were found to serve as a stronger indicator of the opinion climate than AURs on Facebook (Neubaum and Krämer, 2017), these findings could be ascribed to the inequivalent manipulation of comments and AURs, in part, due to the limitations of the platform’s features. Comments were manipulated in terms of positive and negative comments, while AURs were manipulated in terms of a high and a low number of “likes” (Neubaum and Krämer, 2017). Arguably, the low number of “likes” falls short of portraying the unfavorable opinion climate depicted by the negative comments. Facebook’s recent introduction of reaction emojis (i.e. “Love,” “Haha,” “Wow,” “Sad,” and “Angry”), which are graphic symbols that emulate facial expressions allow users to express a wider range of emotions through AURs. This study examines the effect of AURs represented by reaction emojis to determine the role of AURs, and to provide a more equivalent comparison of how comments and AURs influence perceptions of the opinion climate.
Social media platforms also possess unique platform characteristics and features that elicit different mechanisms to influence audiences’ perception of the opinion climate. For instance, individuals frequently rely on source credibility to heuristically evaluate online information (Wirth et al., 2007). Features such as Facebook groups could serve as a heuristic cue for the evaluation of the content, and the eventual acceptance or rejection of the opinion climate. This study bridges source credibility research with the SOS theory to generate insights for how source evaluation influences audiences’ perceptions of the opinion climate and willingness to speak out.
Testing the original assumptions of the SOS theory in the online environment is also crucial as the validity of SOS effects in diverse environments has been contested. Some scholars argue that the original assumptions of the SOS theory, which were conceptualized during the mass media era, may not hold in the new media environment (Ho and McLeod, 2008). For instance, they note differences in information selection patterns and perception processes that may produce differences in perceptions of the opinion climate (Schulz and Roessler, 2012). While some people seek out congruent information to avoid cognitive dissonance, others seek out incongruent information to become more informed of opposing arguments on social media. The contexts in which SOS effects were observed are also vastly different. The SOS theory was formulated during the Nazi Germany era, where people engaged in extreme self-censorship and faced far more adverse consequences for expressing incongruent opinions (Weiss, 2009). In contrast, people commonly express conflicting viewpoints on social media, and the threat of punishment typically manifests in an indirect manner. Yet, SOS effects were found to manifest similarly online as it did offline (Hampton et al., 2014). These mixed positions regarding the original assumptions of the SOS theory underscore the value of investigating its effects in the new media environment.
Spiral of silence theory
The SOS theory states that individuals possess an inherent fear of isolation that motivates their behaviors (Noelle-Neumann, 1974). Those who perceive their position to be aligned with the majority would be emboldened to speak out, while those who perceive their position to contradict the majority would censor themselves. To assert the threat of isolation, the issue must be controversial or morally loaded (Noelle-Neumann, 1974). This study will utilize nuclear energy adoption as a study context given the public discourse and contention surrounding the issue over the last few decades (Ho and Kristiansen, 2019).
The issue of nuclear energy has received increased media spotlight in Singapore due to plans in the Southeast Asian region to adopt nuclear energy as an alternative energy source (Soh, 2016). While plans for nuclear energy adoption in Singapore remain ambiguous, strategies to enhance nuclear-related expertise and capabilities are currently underway (Soh, 2016). Moreover, policymakers are open to the possibility of adopting nuclear energy to achieve energy security and reduce carbon emissions (Stolarchuk, 2019).
Public opinion is among the foremost considerations for policymakers when formulating new policies and developing expertise and capabilities related to controversial science and technology such as nuclear energy. Past experiences have emphasized the role that public support (Ho et al., 2019b, 2019c) plays in the adoption and implementation of controversial science and technology. Due to the salience and personal relevance of this issue, many Singaporeans have taken to online platforms to express their opinions (Ho et al., 2019a; Soezean, 2017). As such, online and social media platforms could potentially shape public opinion toward nuclear energy in the long run.
Spiral of silence online
Social media sites have transformed the way people perceive the opinion climate. Instead of relying on top-down information dissemination from the mass media, people are adopting a bottom-up approach of seeking online user feedback to assess the distribution of opinions. Online user feedback is usually embodied in the following two forms: comments and AURs.
On Facebook, comments refer to the remarks left below a post. Fueled by the interactive nature of social media, comments have gained traction as a means for netizens to express their opinions (Domingo et al., 2008). Notably, comments in response to controversial issues are often biased in favor of one side of the argument (Walter et al., 2018). They frequently manifest in the form of anecdotes or testimonials, serving as exemplars that offer concrete illustrations for otherwise abstract and complex issues such as nuclear energy (Zillmann, 1999). These singular exemplars are subsequently generalized as representing a wider phenomenon and used to form judgments (Zillmann, 1999). Indeed, exemplars were found to influence people’s perceptions of the opinion climate and the future opinion distribution (Perry and Gonzenbach, 1997). Thus, it can be reasonably inferred that comments function as exemplars that shape perceptions of the opinion climate. Drawing from the SOS theory, individuals will likely compare the extent to which the comments are congruent with their personal opinions to gauge whether their opinions align with the majority position. Accordingly, those exposed to congruent comments will perceive greater support for their view than those exposed to incongruent comments.
AURs serve as another form of online user feedback that allow people to infer the opinion climate. On Facebook, AURs are represented by an aggregated number of “likes” below each post that indicates the general sentiment toward it (Stinson, 2016). Audiences have been found to generalize such numeric aggregations as being representative of public opinion (Zerback et al., 2015). Hence, it can be reasonably assumed that AURs function as numeric cues that shape perceptions of the opinion climate. Based on the SOS theory, individuals are likely to assess AURs in terms of their congruency with their own position. Subsequently, those exposed to congruent AURs will perceive the opinion climate to be more supportive of their view than those exposed to incongruent AURs.
Collective influence of comments and AURs
The cumulative power of comments and AURs are likely to have an amplified effect on perceptions of the opinion climate in comparison to only one opinion cue or the other. Although the independent roles that comments and AURs play in influencing attitudes and behavior has been acknowledged, AURs may be limited in their ability to express the polarity of information (Chevalier and Mayzlin, 2006). For instance, Facebook users can express positive sentiments through “likes,” but are not able to express negative sentiments (i.e. the lack of a “dislike” button). Comments can supplement the effects of AURs to amplify their influence on perceptions of the opinion climate. When the comments and AURs are consistent with each other, the combined impact of the cues will be magnified (Baek et al., 2012). Therefore, it is reasonable to infer that the collective influence of comments and AURs will influence perceptions of the opinion climate. Subsequently, audiences who perceive the comments and AURs to reflect the opinion climate will compare them against their own opinion toward the issue to determine the degree congruence between their view and the majority view, to gauge the extent of support for their position.
Relative influence of comments and AURs
The two kinds of cues often appear on social media in conjunction with one another. Therefore, the degree to which they complement or compete with one another can be assessed. Research provides mixed arguments about the role of comments and AURs on perceptions of the opinion climate. Comments are arguably more vivid and memorable. However, AURs provide a more representative reflection of the opinion distribution since a wider fraction of the audience participate in opinion expression through AURs than comments due to the ease of participation (Pang et al., 2016).
Despite this, empirical studies that simultaneously evaluated both types of opinion cues found the effect of comments to supersede that of AURs in influencing perceptions of the opinion climate (Lee and Jang, 2010; Neubaum and Krämer, 2017). These findings can be attributed to the inequivalent manipulation of comments and AURs in earlier studies. Lee and Jang (2010) compared comments against numerical approval ratings on news websites. Unsurprisingly, the impersonal and pallid nature of the latter was less effective in influencing judgments than the personal and vivid nature of the former (Zillmann, 1999). Neubaum and Krämer (2017) replicated this study on Facebook, manipulating comments in terms of positive- and negative-valence comments and AURs in terms of a high and a low number of “likes” (Neubaum and Krämer, 2017). The use of the “like” symbol to represent positive AURs allows for a more accurate portrayal of the favorable sentiments in positive comments as compared to a numerical rating. However, the low number of “likes” arguably falls short of being able to accurately portray the unfavorable sentiments of negative comments. Such manipulation could be attributed to technological constraints of Facebook in the past. The operational problems in earlier studies may be mitigated by Facebook’s recent addition of reaction emojis—“Love,” “Haha,” “Wow,” “Sad,” and “Angry”—that allow for a wider range of expression through which people can infer the opinion climate.
Classifying emotionality of reaction emojis
As communication diffuses to mediated platforms, new modes of expressions such as emojis were introduced to improve communication in a cues filtered-out environment. Emojis are graphical symbols depicting facial expressions, actions, or gestures that allow users to express emotions (Derks et al., 2008) and enhance understanding of the message (Dresner and Herring, 2010). The utility value of emojis in message construction on social media has been acknowledged (Issac, 2015) and widely adopted as a feature of everyday mediated communication.
The prevalence of emojis in communication has spurred research interest in sentiment analysis. Novak et al. (2015) constructed an emoji sentiment lexicon that analyzed emojis in terms of their valence and the intensity of their valence. Juxtaposing the Facebook reaction emojis against the Emoji Sentiment Ranking, “Love,” “Like,” “Haha,” and “Wow” were categorized as positive-valence emotions in descending intensity, while “Sad” was relatively neutral in valence, and “Angry” was negative-valence (Refer to Supplemental Appendix A). This categorization of reaction emojis into positive- and negative-valence emotions suggests that AURs represented by reaction emojis will allow for a more equivalent comparison with positive- and negative-valence comments.
Willingness to speak out
Noelle-Neumann (1974) conceptualized willingness to speak out as an individual’s inclination to either speak out or remain silent on a controversial issue as a function of the opinion climate. Hayes (2007) criticized this conceptualization as being one-dimensional and providing an erroneous depiction of social discourse, given the range of actions that can be adopted in place of remaining silent. People strategically engage in opinion avoidance strategies to vocalize their opinion without providing an explicit stance on the issue (Hayes, 2007). The conceptualization of willingness to speak out in terms of opinion expression and opinion avoidance corresponds to the distinction made between speaking out and speaking up (McDevitt et al., 2003). While speaking out requires one to take an explicit stand, speaking up allows one to voice one’s views without taking a stand. To accurately portray the intricacies of genuine social discourse, it is better to include both opinion expression and opinion avoidance in the conceptualization of willingness to speak out.
With the expansion of and diversity of functionalities aimed at facilitating user interactivity and contribution, users can express themselves both verbally and non-verbally online. Opinion expression is typically operationalized in terms of people’s willingness to express themselves verbally through comments, or non-verbally through “likes” (Gearhart and Zhang, 2015). Hence, this study will operationalize opinion expression verbally through comments, and non-verbally through the “like,” “love,” “sad,” and “angry” reaction emojis.
As abovementioned, individuals’ perceived congruency of their opinions with the opinion climate predicts willingness to speak out. This relationship has received extensive empirical support (Gearhart and Zhang, 2015; Ho and McLeod, 2008; Scheufele et al., 2001). Therefore, it is reasonable to expect opinion climate congruency, reflected through comments and AURs, to predict willingness to speak out.
Source credibility
Source credibility is a factor that has the potential to influence willingness to speak out, yet it has been left largely unexamined. Tsfati (2003) argues that the lack of trust elicits media skepticism, which influences perceptions of the opinion climate. Notably, credibility was suggested as a key element of trust and used as a measure of media skepticism (Tsfati, 2003). The results show that news media effect was moderated by trust, which suggests that source credibility could potentially influence the SOS effects.
Understanding the influence of source credibility is also necessary given the abundance of information and the lack of gatekeepers in this new media age that has shifted the responsibility of evaluating information to individuals. Yet, people often behave as cognitive misers who rely on a few salient features to serve as heuristic cues instead of scrutinizing all the message elements (Lang, 2000). Indeed, source credibility has been found to serve as a key heuristic cue for the evaluation of online information (Metzger et al., 2010). Thus, it is essential to investigate the role that source credibility plays in influencing people’s information evaluation and their willingness to speak out.
Source credibility consists of the following two key dimensions: expertise and trustworthiness. Expertise refers to the assessment of a communicator as being able to make correct assertions, while trustworthiness refers to the extent to which individuals believe that the communicator can be relied upon to tell the truth (Hovland et al., 1953). Source credibility can be easily triggered by simply stating the name of the source on a webpage (Sundar, 2008). Likewise, it can be expected that source credibility can be activated on Facebook using features such as the profile name and profile picture of a Facebook group page.
Earlier studies suggest that source credibility can potentially influence willingness to speak out. On Twitter, source credibility was found to predict people’s intentions to retweet (Boehmer and Tandoc, 2015) and actual retweeting behavior (Ha and Ahn, 2011). Although research related to source credibility and speaking out is scarce, these earlier studies on Twitter provide a basis to infer that a source perceived to possess high credibility can evoke greater willingness to speak out than a source perceived to possess low credibility on alternative social media platforms such as Facebook.
Source credibility could also potentially moderate the relationship between opinion climate congruency and willingness to speak out. To elaborate, source credibility would likely have a greater impact on willingness to speak out in a congruent than in an incongruent opinion climate because people tend to disregard information that conflicts with their pre-existing attitudes (Nickerson, 1998). To avoid cognitive dissonance, people are more likely to ignore incongruent information, which mutes the effect of source credibility on willingness to speak out.
On the contrary, when presented a congruent opinion climate, people will be motivated to pay attention. When individuals perceive themselves to share the same sentiments with a highly credible source, they become assured of their attitudes, which encourages speaking out. However, those who perceive themselves to share similar sentiments as a non-credible source may develop apprehension about their attitudes, which dampens their willingness to speak out.
Method
Study design and participants
This study employed a 3 (opinion cues: comments-only vs AURs-only vs comments-and-AURs) × 2 (opinion climate congruency: congruent vs incongruent) × 2 (source credibility 1 : high vs low) between subjects factorial design. This experimental study (embedded in the form of an online survey) was hosted on the Qualtrics Survey Software platform and engaged Research Now SSI’s self-selected panel for data collection.
A sample of 360 2 current Facebook users were recruited to ensure that participants were sufficiently familiar with the platform and its affordances to facilitate informed responses. The sample is also made up of members of the public across Singapore who are aware of the term nuclear energy. A total of 350 responses were analyzed after removing invalid data. Participants comprised 90.6% Singaporeans and 9.4% Permanent Residents, aged from 21–73 (M = 34.9, SD = 11.3). Participants consisted of 48.9% females and 51.1% males, and 81.4% Chinese, 10% Malays, 6.6% Indians, 0.9% Eurasians, and 1.1% Others. Participants were remunerated with Qualtrics points that could be exchanged for cash vouchers.
Procedure
Prior to stimulus exposure, participants reported their Facebook usage patterns, attitude toward nuclear energy and demographic variables. They were then randomly assigned to 1 of the 12 experimental conditions. At the end, they answered a post-test questionnaire containing the dependent variables and manipulation check questions.
Experimental stimulus. 3
A Facebook mock-up page was designed to replicate an actual Facebook page. Intra-medium elements such as the Facebook post, profile names, and profile pictures of commenters were kept constant. Participants viewed a neutral tone article entitled “Should Singapore adopt nuclear energy?” that stated one advantage and one disadvantage of nuclear energy.
Comments
Four positive- or negative-valence comments were displayed below the Facebook post. The comments were gathered from actual responses to nuclear-related topics across various social media posts to enhance ecological validity and were shortlisted based on the results from the pilot test (Refer to Supplemental Appendix C). All the comments contained 30 words to ensure that there were no biases resulting from the length of the comments.
Aggregated user representations
Positive- or negative-valence AURs were displayed below the Facebook post. Based on the results from the pilot test, positive-valence AURs were reflected by a high number of “like” and “love” reaction emojis, while negative-valence AURs were reflected by a high number of “sad” and “angry” reaction emojis. 4 The number of reaction emojis in all conditions containing AURs were determined using a random number generator 5 and kept constant across all conditions.
Source credibility
Source credibility was elicited through the name and profile picture of a Facebook group page. Based on the results from the pilot test, the high source credibility conditions reflected a Facebook group by the World Health Organization, while the low source credibility conditions reflected a Facebook discussion forum by laypeople.
Measures
Attitude toward nuclear energy
Attitude toward nuclear energy was assessed on a 6-point Likert-type scale (1 = strongly disagree, 6 = strongly agree) for the statement “I support the adoption of nuclear energy in Singapore.” Higher scores indicate greater support for nuclear energy adoption (M = 3.18, SD = 1.45).
Willingness to speak out
Willingness to speak out was measured on a 7-point Likert-type scale (1 = very unlikely, 7 = very likely) by the following items: “I am willing to (a) argue for my position, (b) provide a neutral comment on the post, (c) talk about the opinion of someone else (d) pretend to agree with the majority (e) click the ‘like’ reaction emoji, (f) click the ‘love’ reaction emoji, (g) click the ‘sad’ reaction emoji, and (h) click the ‘angry’ reaction emoji” (M = 3.43, SD = 1.73, Cronbach’s α = .73).
Perceptions of the opinion climate
Perceptions of the opinion climate was measured using 3 items on a 7-point Likert-type scale (1 = strongly disagree, 7 = strongly agree). The items used include (a) “People in this Facebook group are supportive of nuclear energy,” (b) “Singaporeans are supportive of nuclear energy,” and (c) “Singaporeans will be supportive of nuclear energy over the next 10 years” (M = 3.53, SD = .50, Cronbach’s α = .82).
Manipulation checks
Opinion cues
The amount of attention paid to the comments and AURs was assessed by the question “Which of the following did you view earlier?” Possible answers included user-generated comments (UGCs), reaction emojis, both, or I did not notice. Those who selected the option “I did not notice” or answered incorrectly (i.e. those who could not identify whether they were exposed to an experimental stimulus reflecting comments-only, AURs-only, or comments-and-AURs based on the experimental stimulus they were randomly assigned to) were excluded from the analysis.
Source credibility
Source credibility was measured using 6 items on a 7-point semantic-differential scale “Overall, I find the Facebook post to be: (a) Trustworthy–Untrustworthy, (b) Unbiased–Biased, (c) Reliable–Unreliable, (d) Expert–Non-expert, (e) Qualified–Unqualified, and (f) Experienced–Inexperienced.” Incorrect responses (i.e. those who could not identify whether they were exposed to a credible or non-credible source based on the experimental stimulus they were randomly assigned to) were excluded from the analysis.
Participant allocation
After answering the pre-test questionnaire, participants were allocated to a 1 of 12 conditions. To examine the impact of opinion cues and opinion climate congruency on perceptions of the opinion climate, participants were further divided into two groups based on whether they were exposed to a pro- or anti-nuclear energy stimulus for data analysis. In response to the question “I support the adoption of nuclear energy in Singapore,” those who indicated “4 = mildly agree,” “5 = agree,” or “6 = strongly agree” were taken to be supportive of nuclear energy. Based on this, participants were allocated to either the congruent opinion climate condition, or the incongruent opinion climate condition. Those who indicated “1 = strongly disagree,” “2 = disagree,” or “3 = mildly disagree” were taken to be unsupportive of nuclear energy adoption. Likewise, these participants allocated to either the congruent opinion climate condition, or the incongruent opinion climate condition.
Four groups emerged based on this method of random participant assignment. Group 1 consisted of pro-nuclear energy participants that were allocated to a congruent condition (pro-nuclear opinion climate). Group 2 consisted of anti-nuclear energy participants that were allocated to a congruent condition (anti-nuclear opinion climate). Group 3 consisted of pro-nuclear energy participants that were allocated to an incongruent condition (anti-nuclear opinion climate). Group 4 consisted of anti-nuclear energy participants that were allocated to an incongruent condition (pro-nuclear opinion climate).
Results
All data analyses were conducted using SPSS V21. Two-way between subjects ANOVA was used to test H1a, H1b, H1c, and RQ1 (see Table 1). The results for the two-way ANOVA indicated that there was no significant effect for opinion cues (F (2, 338) = 2.68, n.s.). However, there was a significant main effect for opinion climate congruency (F (3, 338) = 110.5, p < .001, partial η2 = .50; see Figure 1). There was also a significant interaction effect between opinion cues and opinion climate congruency (F (6, 338) = 8.18, p < .001, partial η2 = .13; see Figure 2).
Two-way ANOVA for opinion cues and opinion climate congruency on perceptions of the opinion climate.
ANOVA: analysis of variance.
R squared = .55 (Adjusted R square = .53).
Main effect of opinion climate congruency on perceptions of the opinion climate.
Interaction effect of opinion cues and opinion climate congruency on perceptions of the opinion climate.
H1a predicted that opinion climate congruency of comments-only would affect perceptions of opinion climate supportiveness. Pro-nuclear energy participants exposed to a congruent condition (M = 5.18, SD = .85) perceived the opinion climate to be more supportive of nuclear energy than participants exposed to an incongruent condition (M = 4.32, SD = .91). Anti-nuclear energy participants exposed to a congruent condition (M = 1.93, SD = .98) perceived the opinion climate to be more unsupportive of nuclear energy than participants exposed to an incongruent condition (M = 2.98, SD = 1.20). Thus, H1a was supported.
H1b posited that opinion climate congruency of AURs-only would affect perceptions of opinion climate supportiveness. Pro-nuclear energy participants exposed to a congruent condition (M = 4.75, SD = 1.01) perceived the opinion climate to be more supportive of nuclear energy than participants exposed to an incongruent condition (M = 3.29, SD = 1.03). Among anti-nuclear energy participants, those exposed to a congruent condition (M = 2.76, SD = 1.15) perceived the opinion climate to be more unsupportive of nuclear energy than participants exposed to an incongruent condition (M = 3.72, SD = 1.28). Thus, H1b was supported.
H1c postulated that opinion climate congruency of comments-and-AURs would affect perceptions of opinion climate supportiveness. Pro-nuclear energy participants exposed to a congruent condition (M = 5.25, SD = .81) perceived the opinion climate to be more supportive of nuclear energy than participants exposed to an incongruent condition (M = 4.05, SD = 1.10). Anti-nuclear energy participants exposed to a congruent condition (M = 1.72, SD = .84) perceived the opinion climate to be more unsupportive of nuclear energy than participants exposed to an incongruent comments-and-AURs condition (M = 2.75 SD = 1.10). Thus, H1c was supported.
RQ1 sought to explore the relative influence of comments and AURs on perceptions of the opinion climate. Among Group 1 participants, there were no significant differences in perceptions of the opinion climate among participants exposed to the comments-only condition (M = 5.18, SD = .85), AURs-only condition (M = 4.75, SD = 1.01), or comments-and-AURs condition (M = 5.25, SD = .81, n.s.).
Among Group 2 participants, those exposed to the comments-only condition (M = 1.93, SD = .98, p < .001) and comments-and-AURs condition (M = 1.72, SD = .84, p < .001) perceived the opinion climate to be more unsupportive of nuclear energy than those exposed to the AURs-only condition (M = 2.76, SD = 1.15). However, there were no significant differences in perceptions of the opinion climate among participants in the comments-only condition (M = 1.93, SD = .98) and comments-and-AURs condition (M = 1.72, SD = .84, n.s.).
Among Group 3 participants, those exposed to the comments-only condition (M = 2.89, SD = 1.20, p < .05) and comments-and-AURs condition (M = 2.75, SD = 1.10, p < .05) perceived the opinion climate to be more unsupportive of nuclear energy than those exposed to the AURs-only condition (M = 3.73, SD = 1.28). However, there were no significant differences in perceptions of the opinion climate among participants in the comments-only condition (M = 2.89, SD = 1.20) and comments-and-AURs condition (M = 2.75, SD = 1.10, n.s.).
Among Group 4 participants, those exposed to the comments-only condition (M = 4.32, SD = .91, p < .01) and comments-and-AURs condition (M = 4.05, SD = 1.10, p < .001) perceived the opinion climate to be more supportive of nuclear energy than those exposed to the AURs-only condition (M = 3.29, SD = 1.03). However, there were no significant differences in perceptions of the opinion climate between participants in the comments-only condition (M = 4.32, SD = .91) and comments-and-AURs condition (M = 4.05, SD = 1.10, n.s.).
We conducted a second two-way ANOVA to address H2, H3, and H4 (see Table 2). The results for the two-way ANOVA indicated that there was no significant effect for opinion climate congruency (F (1, 346) = .72, n.s.). However, there was a significant main effect for source credibility (F (1, 346) = 11.08, p < .01, partial η2 = .03; see Figure 3). There was also a significant interaction effect between source credibility and opinion climate congruency (F (1, 346) = 4.02, p < .05, partial η2 = .01; see Figure 4).
Two-way ANOVA for opinion climate congruency and source credibility on willingness to speak out.
ANOVA: Analysis of variance.
R squared = .04 (Adjusted R square = .03).
Main effect of source credibility on willingness to speak out.
Interaction effect of source credibility and opinion climate congruency on willingness to speak out.
H2 predicted that participants would be more willing to speak out about nuclear energy when exposed to a congruent than an incongruent condition. However, there was no significant effect found between opinion climate congruency and willingness to speak out (F(1, 346) = .72, n.s.). Thus, H2 was not supported.
Due to the non-significant association, we conducted a post hoc analysis that examined perceptions of the opinion climate as a mediator between opinion climate congruency and willingness to speak out (see Figure 5). We utilized Model 4 of the PROCESS macro (Hayes, 2013) to analyze the mediation effects. The mediation test was based on 10,000 bootstrap samples and a 95% bias-corrected confidence intervals. A significant relationship was found between opinion climate congruency and perceptions of the opinion climate, b = .29, t(348) = 4.16, p < .001. There was also a significant relationship between perceptions of the opinion climate and willingness to speak out, b = .20, t(347) = 5.38, p < .001. The results demonstrated a significant total effect of perceptions of the opinion climate and opinion climate congruency on willingness to speak out. In this model, perceptions of the opinion climate was significantly associated with willingness to speak out, b = .20, t(347) = 5.38, p < .001. On the contrary, opinion climate congruency was not significantly associated with willingness to speak out, b = .02, t(347) = .41, p > .05.
Effect of opinion climate congruency on willingness to speak out, as mediated by perceptions of the opinion climate.
H3 posited that participants would be more willing to speak out about nuclear energy when exposed to a high credibility source than a low credibility source. Participants who perceived the source to possess high credibility (M = 3.60, SD = .08) were more willing to speak out than participants who perceived the source to possess low credibility (M = 3.24, SD = .08). Thus, H3 was supported.
H4 postulated that source credibility will moderate opinion climate congruency to influence willingness to speak out. The findings revealed that when participants were faced with an incongruent opinion climate, there was a negligible difference in their willingness to speak out regardless of whether they were in the high credibility (M = 3.39, SD = 1.01) or low credibility condition (M = 3.53, SD = .96). However, when faced with a congruent opinion climate, participants in the high credibility condition (M = 3.66, SD = .98) were significantly more willing to speak out than those in the low credibility condition (M = 3.08, SD = 1.06). Thus, H4 was supported.
Discussion
This study investigated some of the SOS theory’s original assumptions in the social media environment. It extended upon and mitigated the shortcomings of earlier research regarding opinion cues on Facebook, and examined the effect of source credibility in conjunction with the SOS theory. The findings demonstrated that using reaction emojis provided a more accurate depiction of the opinion climate than the number of “likes.” Source credibility was also found to influence willingness to speak out, and moderate opinion climate congruency to influence willingness to speak out.
Influence of opinion cues on perceptions of the opinion climate
Overall, across the comments-only, AURs-only, as well as comments-and-AURs conditions, opinion cues were found to influence participants’ perceptions of the opinion climate regarding nuclear energy. This demonstrates that users are made cognizant of other people’s attitudes toward an issue through opinion cues and generalize them to be representative of public opinion. These findings also corroborate results from earlier research (Lee and Jang, 2010; Neubaum and Krämer, 2017), and reinforces the role that comments play in influencing perceptions of the opinion climate.
Notably, the findings suggest that AURs with the inclusion of reaction emojis affect perceptions of the opinion climate. This contrasted the negligible effects of AURs found in earlier studies (Lee and Jang, 2010; Neubaum and Krämer, 2017). These findings suggest that reaction emojis can mitigate the absence of emotionality and polarity of AURs to more accurately reflect the opinion climate, and highlight to need to examine AURs represented by reaction emojis in future SOS research.
The significant findings for comments-and-AURs also suggest that people perceive the opinion climate through different opinion cues, and emphasize the importance of examining both types of opinion cues in the investigation of perceptions of the opinion climate. Yet, it is important to note that while both comments and AURs affected perceptions of the opinion climate, participants’ perceptions of the opinion climate were more strongly influenced by comments-only and comments-and-AURs, than AURs-only. These findings are in line with earlier arguments that the emotional nature and vividness of comments make them easier to grasp than the impersonal and pallid base-rate descriptions of AURs (Zillmann, 1999). This suggests that comments may be superior to AURs in influencing perceptions of the opinion climate. The lack of significant differences between the comments-only and comments-and-AURs condition also suggests that they function in a similar manner. Alternatively, in the face of both types of opinion cues, audiences may pay more attention to comments than AURs.
Opinion climate congruency on willingness to speak out
Although opinion climate congruency did not have a direct effect on willingness to speak out, perceptions of the opinion climate were found to mediate the relationship between opinion climate congruency and willingness to speak out. Put simply, people were more willing to speak out in the face of a congruent opinion climate if they can accurately perceive the opinion climate to be in line with their position on the issue. This finding concurs with Bodor’s (2012) argument that individuals may be limited in their ability to estimate the extent of dissonance between one’s own position and the perceived broader opinion climate, making it difficult to influence speaking out behavior. Moreover, there is weak empirical support found for the relationship between opinion climate congruency on willingness to speak in extant research. Some studies found mixed results for the effect of opinion climate congruency on speaking out behavior (Salmon and Neuwirth, 1990), while others found contradictory results (McDevitt et al., 2003). This suggests that the significant findings could be attributed to measurement error, or that opinion climate congruency may only be one of the many factors influencing willingness to speak out, and contingent on other factors.
Source credibility on willingness to speak out
This study generates important insight regarding the role of source credibility in the SOS literature. Participants were found to be more willing to speak out when they perceived the source to possess high credibility than low credibility, which supports the claim that sharing the same stance as those perceived to be an expert and trustworthy encourages people to speak out. Source credibility also served as a moderator for opinion climate congruency and willingness to speak out. This is unsurprising, as people who perceive their views as being supported by a source that is highly credible would likely become more certain of their view and willing to speak out, and vice versa. On the contrary, when faced with an incongruent opinion climate, source credibility had a negligible impact on people’s willingness to speak out due to the confirmation bias (Nickerson, 1998).
Theoretical and practical implications
Theoretically, this study provides important implications for the SOS theory by showing that people adopt new modes of expression to perceive and evaluate the opinion climate. This highlights the importance of assessing both comments and AURs with the inclusion of reaction emojis in the examination of online SOS effects. In addition, the impact of comments was found to supersede that of AURs. More research is required to understand how people perceive different opinion cues, and whether there are different motivations driving people’s choice of speaking out through comments or AURs. Practically, the findings highlight to relevant stakeholders about the importance of monitoring both verbal and non-verbal forms of subversive online behavior that attempt to influence public opinion regarding issues of national significance (e.g. nuclear energy adoption).
This study also ascertained the effect of source credibility on willingness to speak out and emphasizes the need to consider source credibility in future SOS studies. The role of source credibility is particularly salient on social media which is based on subjective opinions rather than objective facts due to the lack of journalistic gatekeeping. As social media use continues to gain prominence, it is probable that people may become more reliant on the online opinion climate to determine public opinion than the offline opinion climate. Researchers should identify other factors aside from source credibility that could potentially influence online silencing effects.
Limitations and future research
The generalizability of the findings is limited due to the self-selected nature of the participants. However, quotas were enforced to ensure that the study population closely reflected Singapore’s demographics. As this study is only interested in examining the Facebook population, conducting the study through an experiment (embedded in a survey) would provide greater ecological validity as it allows participants to view and interact with the experimental stimulus as they would in a natural setting. Another shortcoming is the use of a single-item measure for attitudes toward nuclear energy. While there is empirical support for using single-item measures to gauge attitudes (Gardner et al., 1998), future studies should strive to include a minimum of three items.
This study also manipulated opinion climate in terms of either a positive-only or a negative-only opinion climate. However, social media often reflect mixed opinion climates. It is necessary to understand how this may influence SOS effects. Mixed opinion climate conditions (e.g. positive-valence comments and negative-valence AURs or vice versa) will also allow for meaningful comparisons among the comments-only, AURs-only, and comments-and-AURs conditions to ascertain the relative impact of different opinion cues on perceptions of the opinion climate.
This study was conducted based on the opinion cues available on Facebook. However, AURs manifest in various forms across different social media platforms, which may have differential effects on the SOS. For instance, AURs on Instagram and Twitter are represented by the heart symbol, which limits the expression of negative-valence emotions. Future studies should examine whether the intended effects pan out in a similar manner across various social media platforms.
Finally, it is important to note that although comments and AURs serve as cues for people to detect the broader opinion climate, they may not require the same level of cognitive processing from the participants. To elaborate, it likely requires less effort for participants to observe the number of AURs as compared to reading comments. Thus, future studies should delve deeper into examining the role of cognitive effort in influencing perceptions of the opinion climate.
Supplemental material
Supplementary_file - Perceiving online public opinion: The impact of Facebook opinion cues, opinion climate congruency, and source credibility on speaking out
Supplementary_file for Perceiving online public opinion: The impact of Facebook opinion cues, opinion climate congruency, and source credibility on speaking out by Alisius D Leong and Shirley S Ho in New Media & Society
Footnotes
Acknowledgements
All the authors have agreed to the submission and the article is not being considered for publication by any other print or electronic journal.
Declaration of conflicting interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work was supported by the Wee Kim Wee School of Communication and Information, Nanyang Technological University (M4082243.060).
Supplemental material
Supplemental material for this article is available online.
Notes
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.
