Abstract
The 2016 US presidential election was highly contentious, as both candidates, Hillary Clinton and Donald Trump, received strong polarizing support and opposition with controversial campaigns, name-calling, and violence at campaign rallies. This may have contributed to an opinion climate where citizens were reluctant to express support for a candidate. This study tests the spiral of silence theory in the context of this election. We examine the interplay among opinion congruency, fear of isolation, and willingness to express support for a candidate. Data from an online survey show that opinion congruency for Clinton in society at large, and for Trump on Facebook, had indirect associations with willingness to express support for a candidate face-to-face, on Facebook, and in anonymous online settings through fear of isolation.
Noelle-Neumann’s (1974) spiral of silence (SOS) theory posits that individuals’ willingness to express opinions stems from their perceptions of an opinion climate. Numerous studies have tested the theory since its inception (Matthes, Knoll, & von Sikorski, 2018; Scheufele & Moy, 2000). Recent research has assessed whether the theory applies to online contexts (Chan, 2018; Fox & Warber, 2015; Gearhart & Zhang, 2015; Kim, Kim, & Oh, 2014; Liu & Fahmy, 2011; Porten-Cheé & Eilders, 2015; Stoycheff, 2016; Yun & Park, 2011), and evidence suggests that it does (Matthes et al., 2018).
Despite the rich body of work in this domain, research examining the SOS in online environments remains limited. For example, as political polarization increases in the United States, social media sites such as Facebook and Twitter afford people greater ability to select with whom they interact online, which may create a filter bubble that limits what information they are exposed to (Pariser, 2012). One question that arises is, “Will such an online environment foster the sense that one’s opinions are shared by others?” Also, fear of isolation (FOI) is a key attribute of the SOS, yet its role in linking opinion congruency with one’s willingness to express opinions has not been entirely clear in online contexts. Thus, a corollary question is, “Are people with less FOI more willing to speak out if they think their opinion is shared by the majority of people they interact with online?”
The purpose of our study is to address these questions. Specifically, we examine the interplay among opinion congruency, FOI, and willingness to express support for a candidate, with opinion congruency and willingness to express support differentiated between face-to-face and online settings. We test the SOS theory in the context of the 2016 US presidential election, one of the most contentious in the US modern presidential era (Flegenheimer & Barbaro, 2016; Kurtzleben, 2016), and one in which we assume the SOS likely played out. The premise of our study is that opinion congruency for a candidate fosters one’s willingness to express support for the candidate because such perceptions lower fear of social isolation. Data from a web survey conducted before the election were analyzed to test theoretically derived hypotheses. Findings extend understanding of why some individuals express their political preferences while others do not.
The SOS Theory
The SOS theory explains variations in opinion expression in public forums (Noelle-Neumann, 1974). SOS posits that some individuals are willing to publicly express their opinions because they perceive that their views are shared by a majority of a social group they are part of, or society at large. Such perceptions of public opinion, or opinion congruency, lead them to be expressive. In contrast, some individuals choose to remain silent because they think their views are in the minority.
The theory has an important democratic implication. The assumption that people are only willing to express opinions that they think are widely shared among others, and that they will suppress expression of unpopular opinions, implies that citizens may not be able to make the most informed decision in the democratic process (Mendelberg, 2002). An absence of dissenting and even objectionable opinions undermines the healthy marketplace of ideas, strips citizens of the opportunities to learn different viewpoints, and as a result, reduces their ability to engage in democratic governance.
According to the SOS, FOI is the key mechanism explaining why some choose to suppress expressing unpopular opinions. In Noelle-Neumann’s term, FOI refers to fears of rejection by others (Noelle-Neumann & Peterson, 2004). The social nature of human existence leads people to be afraid of isolation. Isolation might take the form of negative evaluations by others, loss of status, and exclusion from peers. If people think that a majority does not share their opinions, they choose to remain silent for fear of such negative social outcomes (Noelle-Neumann, 1974).
Although Noelle-Neumann did not provide a clear definition of FOI, Neuwirth, Frederick, and Mayo (2007) drew on her writings to define the concept as “a psychological variable representing a negative emotional state associated with the prospect of voicing one’s opinion about a given topic” (p. 452). FOI, when conceptualized in this manner, is not a personal trait that is relatively static. Rather, it is transitory in nature or unique to a given situation (Neuwirth et al., 2007). This conceptualization differs from FOI as a stable personality trait whereby individuals with FOI tend to have a general fear of being isolated (Kim, 2012).
Previous research has supported the role of FOI in the SOS process (Ho & McLeod, 2008; Kim, 2012; Neuwirth, 2000; Scheufele, 1999; Scheufele, Shanahan, & Lee, 2001). Kim (2012) uncovered that FOI was negatively related to three indicators of opinion expression (speaking on TV, engaging in an online forum, and joining a conversation). Using a cross-national sample of individuals living in South Korea, Chile, Mexico, United States, United Kingdom, Germany, France, and Russia, Matthes et al. (2012) demonstrated that FOI was positively associated with willingness to self-censor, a known negative predictor of opinion expression.
To avoid social isolation, people constantly monitor social environments and gauge the opinion climate. While early research on the SOS focused on societal majority and mass media as sources of opinion climates (Noelle-Neumann, 1974), later work has shown that people also assess opinion climates via reference groups (Liu & Fahmy, 2011; Noelle-Neumann, 1993). To evaluate their own personal attributes, people often use reference groups, such as friends and colleagues, as a point of comparison. For example, Moy, Domke, and Stamm (2001) found that perceived opinion congruency with friends and family was related to willingness to speak out but perceived majority opinion in society was not. Therefore, research points to the importance of assessing the role of reference groups versus majority opinion in predicting one’s willingness to express opinion (Oshagan, 1996; Salmon & Neuwirth, 1990).
The SOS and the Online Opinion Climate
Social media, or online networks that facilitate interaction around the production and exchange of online content (Bruns, 2006), have become a popular avenue for opinion expression and exposure to political issues (Gottfried & Shearer, 2016). Social media present unique opportunities to extend the SOS theory. Through social media, one can form, develop, and maintain connections with people of different types beyond time and space constraints (Stoycheff, 2016). For example, one may keep in touch with friends from high school, colleagues from past jobs, and professional peers. By scanning what others say about issues on social media, people can judge what opinion seems in the majority or minority. Thus, the opinions viewed on social media might serve as a reference point.
One unique attribute of social media is enhanced selectivity. Social media afford users the ability to selectively interact with like-minded individuals and seek ideologically congruent opinions, which might contribute to forming echo chambers or filter bubbles where people attend largely to ideologically aligned content while shutting off cross-cutting opinions (Pariser, 2012; Sunstein, 2001).
However, the filter bubble thesis is subject to debate (e.g., Gentzkow & Shapiro, 2011; Manjoo, 2015). For example, Gentzkow and Shapiro (2011) found that Internet users tended to visit news sites across the political spectrum when they were actively seeking news. Other research suggests that passive consumers of news through social media tend to engage in selective ideological consumption or interaction (Flaxman, Goel, & Rao, 2016; Towner & Dulio, 2011). For example, Flaxman et al. (2016) showed that while social media users tended to have higher ideological distance, they were also exposed to cross-cutting information by chance. These studies suggest a difference between seeking ideologically confirming interactions and incidental exposure to ideologically challenging content on social media. In fact, research shows that while social media may foster exposure to ideologically cross-cutting opinions, users tend to share and click on ideologically confirming content (Bakshy, Messing, & Adamic, 2015). The key hence seems to be the active choices people make in the selection of whom to interact with and what information to read.
Different social media sites allow varied degrees of anonymity. While one’s identity is relatively known on Facebook, other social media sites such as Twitter, Reddit, and blogs allow people to use fictitious usernames and remain anonymous. This difference may influence one’s willingness to express unpopular opinions and FOI. Compared to the ephemeral nature of offline conversations, online expression can exist permanently, spread fast, and be readily copied and searched (boyd, 2010), which can increase the likelihood that others will discover the expression of unpopular views and, as a result, the risk of social sanctions (Stoycheff, 2016), particularly when one’s identity is known to others. One exception is hardcore nonconformists who do not care about social isolation and thus speak out regardless of opinion climates (e.g., Matthes, Morrison, & Schemer, 2010; Moy et al., 2001).
Prior research has generally shown that the SOS theory can be applied to opinion expression in non-anonymous online settings (Fox & Warber, 2015; Gearhart & Zhang, 2015; Stoycheff, 2016). For example, Fox and Warber (2015) investigated why some lesbian, gay, bisexual, and transgender (LGBT)+ individuals were unwilling to disclose on Facebook. The researchers showed that FOI derived from the perception that a majority of one’s Facebook network did not support LGBT+ appeared to inhibit willingness to disclose. Gearhart and Zhang (2015) examined social networking sites where users’ identity is commonly visible to other users (e.g., Facebook, LinkedIn, Google+) and reported that people who received negative reactions to political posts were less likely to respond to disagreeable posts and more likely to refrain from responding.
Research on the SOS in anonymous online settings has provided mixed evidence (Liu & Fahmy, 2011; Porten-Cheé & Eilders, 2015; Yun & Park, 2011). Yun and Park (2011) found that participants who were anonymous were no more willing to post a comment on abortion in an online forum than those who posted to the forum using personal information (e.g., names, email addresses). Porten-Cheé and Eilders (2015) found that opinion congruency was not associated with willingness to speak out in an anonymous blog setting. Liu and Fahmy (2011) found that opinion congruency was not related to willingness to express opinions on the legalization of same-sex marriage in typically anonymous online settings (e.g., chat room, blog, website).
Yet Liu and Fahmy (2011) showed that FOI was negatively related to willingness to speak out online, suggesting that even if anonymous online settings foster a sense of protection from being identified by others, possible isolation from an online community plays a similar opinion suppressing role. Liu and Fahmy (2011) speculated, however, that the effect of FOI might be weakened in the presence of even a small reference group willing to speak up on one’s behalf. This suggests that people are still susceptible to the SOS effect, but the role of FOI might not be as salient as other contexts where one’s identity is visible and known to others.
Based on SOS research, the potential effects of anonymity, and our focus on an election context, we examine three forms of willingness to express opinions—willingness to express support for a candidate face-to-face, on Facebook, and in anonymous online settings.
Goals of the Study
We propose three sets of hypotheses. First, we parse opinion congruency into two forms—societal majority and reference groups on Facebook with which people affiliate. This distinction is important, because research shows that when societal majority opinions and one’s reference groups are equally salient, people use the latter to judge an opinion climate (Moy et al., 2001; Oshagan, 1996), thereby clarifying a theoretical mechanism of the SOS. Yet, the role of reference groups on social media has not been fully clarified, particularly in comparison with offline societal majority opinions. Therefore, the first hypothesis investigates a relationship between opinion congruency and FOI:
H1. Opinion congruency (a) in society and (b) on Facebook will be negatively associated with FOI.
Second, we predict FOI to negatively influence opinion expression not only in face-to-face settings but also on social media where their affordances can increase the chance that one’s expressed opinions will be noticed by other users (Stoycheff, 2016). This possibility may reduce people’s willingness to express unpopular views for fear that others would find their expression of such opinions and that, as a consequence, they might face social isolation. Although FOI likely plays a role in anonymous online setting (Liu & Fahmy, 2011), its influence may not be as strong as other settings where one’s identity is known to others. Thus, we formulate the following hypothesis:
H2. FOI will be negatively associated with willingness to express support for a candidate (a) face-to-face, (b) on Facebook, and (c) in anonymous online settings.
Finally, we predict the association between opinion congruency and willingness to express support for a candidate to occur indirectly through FOI. This prediction draws on the conceptualization of FOI as a situational factor that can vary by opinion congruency rather than a personality trait that is relatively stable regardless of circumstances. Given SOS plays out in online contexts (e.g., Liu & Fahmy, 2011; Stoycheff, 2016), we expect this reasoning to persist across online and offline contexts. Based on this rationale, we formulate the following hypotheses:
H3a. Opinion congruency in society will be indirectly associated with willingness to express support for a candidate (a) face-to-face, (b) on Facebook, and (c) in anonymous online settings through FOI.
H3b. Opinion congruency on Facebook will be indirectly associated with willingness to express support for a candidate (a) face-to-face, (b) on Facebook, and (c) in anonymous online settings through FOI.
The Research Context: 2016 US Presidential Election
The 2016 US presidential election was a specific, protracted event in which voters directly or indirectly experienced heightened tension and conflict. Leading up to the election, both Donald Trump and Hillary Clinton had low favorability ratings (“Election 2016 Favorability Ratings,” n.d.). Their candidacies were polarizing with both strong support and opposition. After the election, Time Magazine named then-president-elect Donald Trump the 2016 “Person of the Year,” referring to him as the “President of the Divided States of America” on its cover (Scherer, 2016).
During the campaign, Trump made many controversial and insulting statements, including comments about immigration (e.g., “When Mexico is sending its people, they’re not sending their best …. They’re bringing drugs. They’re bringing crime. They’re rapists. And some, I assume are good people”), Muslims (e.g., “Donald J. Trump is calling for a total and complete shutdown of Muslims entering the United States”), women (e.g., “blood coming out of her wherever”), and other candidates (e.g., “Little Marco,” “Lyin’Ted,” “Crooked Hillary”), among others (Heilpern, 2017; The New York Times, 2016).
Clinton was also a controversial figure (The New York Times, 2016). She was viewed as a Washington insider among a public fed up with government, was criticized for her stance on trade, was the subject of a high-profile investigation into her alleged use of a private email server to communicate classified information, and was accused of favoring individuals who donated to the Clinton Foundation while serving as Secretary of State (Rogan, 2016). Even if her campaign did not use insulting rhetoric like Trump’s, Clinton made statements that many found questionable and offensive, not only the “basket of deplorables” phrase to refer to Trump supporters but also others such as “We are going to put a lot of coal miners and coal companies out of business” and “I love it …. Those messages disappear all by themselves [referring to her campaign’s use of Snapchat]” (Greve, 2016).
The highly contentious nature of this election lends itself to questions about voters’ willingness to express support for Trump or Clinton. Expressing support for these divisive candidates triggered strong emotional reactions from citizens, as evidenced, for example, by several instances of altercations against protestors and Clinton supporters at Trump’s campaign rallies (Mathis-Lilley, 2016) and frequent use of negative labels directed at Trump supporters as racists, sexists, xenophobes, and uneducated (Barabak & Duara, 2016).
While Kim (2012) proposed that a transitory FOI could serve as a potential mediator between opinion congruency and willingness to speak out, he noted challenges to address this subject through survey-based research, as a survey might not reliably trigger the transitory fear that compels people to withhold personal views. However, the timing of the data collection during a highly salient, controversial, and divisive election that was permeating all aspects of American life makes it reasonable to evaluate a transitory FOI through the survey method.
Method
Sample and Procedure
The data for this study came from a web survey of an online panel of participants recruited by Qualtrics, an online panel sample aggregator. Given the skewed demographic characteristics of online panel members, a sample was selected to match the age, gender, and race distribution of the US adult population. The survey was fielded during the 2 weeks prior to the 2016 presidential election. A total of 630 responses were collected. The median age of the sample was 45–54 measured on a 6-point scale ranging from 18–24 to 65 or above. The sample included 48.1% of male and 51.9% of female participants. Roughly, 73% of the sample were White, 14.1% of Black or African American, 4.4% of Hispanic, 7.6% of Asian or Asian American, and 3.2% of other-category participants; 97.5% were high school graduates or held higher degrees and 33.4% had some college and bachelor’s degree or higher. The median household income of the sample was US$35,000–US$49,999 measured on a 10-point scale item ranging from under US$10,000 to US$200,000 or more, with US$50,000 to US$74,999 as the sample mode. Although the sample cannot be considered representative of the general American public, limiting our ability to draw population-level inferences, the data allow for inferences in terms of the underlying process by which perceived opinion climate affects willingness to speak out through variations in FOI (Hayes, 2013).
Measures
Opinion congruency
Following prior work (Ho & McLeod, 2008; Ho, Chen, & Sim, 2013), a three-step procedure was taken to measure opinion congruency. First, respondents were asked how much they supported or opposed Hillary Clinton and Donald Trump, respectively, on a 5-point scale (1 = strongly oppose; 5 = strongly support). This item was recoded on a 5-point scale from −1 to 1. Next, respondents were asked to indicate what percentage of Americans they thought supported Hillary Clinton and Donald Trump, respectively, on a 100-point scale. Respondents who scored 0–49 were coded as −1, while those who scored 50–100 were coded as 1. Then, the score for respondents’ support for Clinton and Trump, respectively, was multiplied by the score for their perception of Americans’ support for each candidate. The resulting scores, opinion congruency for Clinton and Trump, ranged from −1 to 1, with higher positive scores indicating higher opinion congruency (M = .37, standard deviation [SD] = .71 for Clinton and M = .53, SD = .64 for Trump). We measured opinion congruency for Clinton and Trump separately instead of doing so on a single continuum, because one’s lack of support for Clinton does not necessarily mean his or her support for Trump, and vice versa.
Using the same procedure, we created an opinion congruency measure for Clinton and Trump on Facebook. We asked respondents to rate what percentage of their friends on Facebook they thought supported each candidate on a 100-point scale. Respondents who scored 0–49 were coded as −1 and those who scored 50–100 were coded as 1. The score for respondents’ support for Clinton and Trump, respectively, was multiplied by the score for perceived support of their friends for each candidate. The resulting score, opinion congruency for Clinton and Trump on Facebook, ranged from −1 to 1, with higher positive scores indicating higher opinion congruency on Facebook (M = .40, SD = .69 for Clinton and M = .45, SD = .70 for Trump).
FOI
FOI was measured by five items adapted from prior work (Ho & McLeod, 2008). Respondents were asked to rate how strongly they agreed or disagreed with the following statements on a 5-point Likert-type scale: (1) I worry about being isolated if people disagree with me about the candidate I support, (2) I avoid telling other people which candidate I support when there is a risk, (3) I do not enjoy getting into arguments about the candidate I support, (4) I enjoy a good argument over the candidate I support (recoded), and (5) I try to avoid getting into arguments about the candidate I support. Responses were averaged on a 5-point scale (M = 3.03, SD = .78, α = 70).
Willingness to express support for a candidate
We measured three types of willingness to express support for a candidate. Willingness to express support face-to-face was measured by asking respondents, on a 5-point scale (1 = very unwilling; 5 = very willing), how willing they would be to express support for their preferred presidential candidate in the following settings, if such a topic came up: at work, among strangers at a restaurant, at a community meeting, at an informal gathering among family and friends, and in a TV interview. Responses were averaged on a 5-point scale (M = 3.15, SD = 1.13, α = .92).
Willingness to express support on Facebook was measured by four items. Respondents were asked how willing they would be to express support for their preferred presidential candidate on their Facebook status updates, through likes and shares of news articles, videos, or images that praise their preferred candidate, by writing to the comment section of a news post about their preferred candidate, and in a discussion with their friends on Facebook. Responses were averaged on a 5-point scale (1 = very unwilling; 5 = very willing) (M = 3.05, SD = 1.35, α = .96).
Willingness to express support in anonymous online settings was measured by three items. Respondents were asked how willing they would be to express support for their preferred presidential candidate (1) on social media using a fictional user name and profile, (2) in a blog’s or news website’s comment section where they could post anonymously, and (3) on anonymous online forums or discussion boards. Responses were averaged on a 5-point scale (M = 2.87, SD = 1.34, α = .93).
Statistical controls
To assess the unique influence of the proposed predictors on willingness to express support for a candidate, we controlled for several variables, including age, gender, education, income, race, political interest, party affiliation, and political ideology. Age was measured on a 6-point scale (median = 4, 45–54). Gender was coded with female respondents as 1 and male respondents as 0 (female = 51.9%). Education was measured by an 8-point scale item ranging from less than high school to postgraduate or professional degree (median = 4, some college). Income was measured by a 10-point scale item ranging from under US$10,000 to US$200,000 or more (median = 5, US$35,000–US$49,999). Race was coded with Caucasians as the high value (73.2%). Political interest was measured by a 7-point scale item (1 = not interested at all; 7 = very interested) asking how interested respondents were in the 2016 US presidential election (M = 5.43, SD = 1.77). Party affiliation was measured by a 7-point scale item (1 = strong Republican; 7 = strong Democrat) (M = 4.21, SD = 1.95). Political ideology was measured by a 7-point scale item (1 = very conservative; 7 = very liberal) (M = 3.79, SD = 1.73).
We also controlled for traditional media use and Facebook use for news. Traditional media use was measured by six items. Respondents were asked how much attention they paid to public media, network television news, and newspapers offline and online, respectively. Responses were averaged on a 7-point scale (M = 2.54, SD = 1.01, α = .85). Facebook use for news was measured by four items. Respondents were asked how often they used Facebook to get news about the election, stay informed about the election, learn others’ opinions on the election, and get election news from traditional news media. Responses were averaged on a 5-point scale (1 = never; 5 = very often) (M = 2.59, SD = 1.34, α = .96).
Analytical Strategy
To test H1 and H2, we estimated ordinary least squares (OLS) regression models predicting FOI and willingness to express support face-to-face, on Facebook, and in anonymous online settings. To test H3, we used the SPSS PROCESS macro (Hayes, 2013), whereby FOI was specified as a mediator linking opinion congruency and willingness to express support for a candidate.
Results
The first column in Table 1 presents an OLS regression model predicting FOI. Of the four opinion congruency variables, opinion congruency in society for Clinton was positively related to FOI (β = −.119, p < .01). That is, those who more strongly perceived their opinions on Clinton were in the majority had lower levels of FOI. Opinion congruency in society for Trump was not significantly related to FOI. Opinion congruency on Facebook for Clinton was inversely related to FOI (β = −.084, p = .05), with the relationship nearly significant at p < .05. Opinion congruency on Facebook for Trump was also negatively related to FOI (β = −.145, p < .01). That is, those who more strongly felt their opinions on Clinton and Trump, respectively, were in the majority on their Facebook networks reported lower levels of FOI. On balance, the data show partial support for H1a and H1b. In addition, the results indicated that political ideology and political interest were negatively related to FOI and that party affiliation had a positive relationship with FOI.
Predictors of Fear of Isolation.
Entries are standardized regression coefficients (N = 626).
p < .001, **p < .01, *p < .05, +p < .10.
The second and third columns in Table 1 show OLS regression models predicting willingness to express support for a candidate face-to-face. Model 1 in the second column shows that opinion congruency in society was not associated with willingness to express support for a candidate face-to-face. In contrast, opinion congruency on Facebook for Clinton and Trump had a positive relationship with willingness to express support for a candidate face-to-face (β = .146, p < .001 and β = .096, p < .05, respectively). That is, those who more strongly perceived their opinions on Clinton and Trump, respectively, were in the majority on their Facebook networks were more willing to express support for a preferred candidate.
Model 2 in the third column introduces FOI to the model. The data show that FOI was negatively related to willingness to express support for a candidate face-to-face (β = −.298, p < .001). Respondents who had higher levels of FOI were less willing to express support for a preferred candidate face-to-face. It is important to note that when FOI was added to the model, opinion congruency for Trump on Facebook was statistically insignificant. The relationship of opinion congruency for Clinton on Facebook with expressing support for a candidate face-to-face remained significant (β = .121, p < .01). Also, age, political interest, and Facebook use for news were positively associated with this outcome variable.
The first two columns in Table 2 show OLS regression models predicting willingness to express support for a candidate on Facebook. Model 1 in the first column shows that Opinion congruency for Clinton (β = .080, p < .05) and Trump (β = .077, p < .05) on Facebook were positively related to willingness to express support for a candidate on Facebook. That is, those who more strongly perceived their opinions on Clinton and Trump, respectively, were in the majority on their Facebook networks were more willing to express support for a preferred candidate on Facebook. Opinion congruency in society for Clinton was marginally related to willingness to express support for a candidate on Facebook (β = .060, p < .10). Model 2 in the second column introduces FOI to the first model. FOI was negatively associated with willingness to express support for a candidate on Facebook (β = −.232, p < .001). When FOI was controlled, the observed effects of opinion congruency for Clinton and Trump on Facebook became statistically insignificant. Political interest and Facebook use for news were also positively related to the outcome variable. Respondents who were more interested in the election and who more often used Facebook for news reported greater willingness to express support for a preferred candidate on Facebook.
Predictors of Willingness to Express Support for a Candidate Face-to-Face and on Facebook.
Entries are standardized regression coefficients (N = 626).
p < .001, **p < .01, * p < .05, +p < .10.
The third and fourth columns in Table 2 show OLS regression models predicting willingness to express support for a candidate in anonymous online settings. The opinion congruency variables were not significantly related to willingness to express support for a candidate in anonymous online settings. The exception was opinion congruency in society for Trump. It had a nearly significant association (β = .077, p = .050). Much like the preceding results, FOI had an independent negative association with willingness to express support for a candidate in anonymous online settings (β = −.079, p < .05), yet the magnitude of the relationship was weaker. When FOI was taken into account, the influence of opinion congruency in society for Trump was statistically significant (β = .078, p < .05). The data thus provide support for H2. In addition, age, political interest, and Facebook use for news were significantly related to the outcome variable, with those who were younger, more politically interested, and active news consumers on Facebook more willing to express support for a preferred candidate anonymously online.
Finally, we assessed whether the associations between opinion congruency and willingness to express support for a candidate would occur indirectly through FOI. Opinion congruency for Clinton in society had indirect associations with willingness to express support for a candidate face-to-face (point estimate = .056, confidence intervals [CIs = 017, .100]), on Facebook (point estimate = .052, CIs [.017, .098]), and in anonymous online settings (point estimate = .018, CIs [.020, .049]) through FOI. Opinion congruency for Trump in society was not indirectly related to any of the dependent variables. Thus, support for H3a is mixed, with differences across candidates (Table 3).
Indirect Associations between Opinion Congruency and Willingness to Express Support for a Candidate that Occur through Fear of Isolation.
CIs: confidence interval.
Opinion congruency for Clinton on Facebook had an indirect relationship with willingness to express support for a candidate face-to-face (point estimate = .038, CIs [.001, .085]) and on Facebook (point estimate = .013, CIs [.0001, .041]) through FOI. Interestingly, opinion congruency for Trump on Facebook was related to willingness to express support for a candidate face-to-face (point estimate = .070, CIs [.030, .119]), on Facebook (point estimate = .065, CIs [.027, .114]), and in anonymous online settings (point estimate = .022, CIs [.003, .056]) indirectly through FOI. Thus, support for H3b is mixed, with differences across candidates.
Discussion
The above-presented results extend the SOS literature in important respects. First, consistent with the literature (e.g., Neuwirth et al., 2007; Noelle-Neumann, 1974), on balance, opinion congruency had a negative association with FOI, particularly on Facebook. The result seems to support the role of reference groups in the SOS relative to societal majority (e.g., Moy et al., 2001; Oshagan, 1996). Facebook is a platform where users connect with family, friends, colleagues, and acquaintances, and hence maintaining relational harmony and a good image may be a priority. It may be important for users to determine a Facebook opinion climate. An incongruent Facebook opinion climate may cause people to worry about ruining relationships if they express opinions they think others will likely oppose.
Interestingly, opinion congruency for Trump in society was not related to FOI. This implies that those strongly in favor of, or against, Trump were not necessarily afraid of how society would think about them, even if they felt a societal majority opposed Trump (for Trump supporters) or a societal majority supported him (for Trump opponents). Considering controversies surrounding the Trump campaign, the prospect of social sanction may have been higher if people had expressed unpopular opinions on Facebook where they know others relatively well than in general social settings where people do not know others’ identity.
Second, consistent with our prediction, FOI was negatively related to willingness to express opinions in support of either Clinton or Trump in both offline and online settings. The result is consistent with the extant research showing the negative association between FOI and willingness to speak out face-to-face (Ho & McLeod, 2008; Kim, 2012; Matthes et al., 2012; Scheufele, 1999), and support for this role of FOI extends to social media in anonymous and non-anonymous settings (Fox & Warber, 2015; Gearhart & Zhang, 2015; Liu & Fahmy, 2011).
It is noted, however, that although significant, the linkage between FOI and willingness to express support for a candidate was weaker in anonymous online settings than face-to-face and on Facebook. While expressing unpopular opinions face-to-face can certainly have negative social consequences, the observed difference between Facebook and anonymous online settings may be due to different affordances (boyd, 2010; boyd & Ellison, 2007; Sundar & Limperos, 2013). Compared to anonymous online settings, opinions posted on Facebook are publicly viewable, searchable, and permanent barring deletion by the user. As posts have increased in visibility because of site notifications and algorithms and as Facebook relies heavily on users’ true identity and interaction with personal networks, users may actively self-monitor to avoid possible social sanction (Stoycheff, 2016). The result is in line with the possibility that online anonymity does not necessarily erase the FOI that may arise from expressing unpopular views, yet people may not be as fearful as when their identity is known to others (Liu & Fahmy, 2011). As other social media platforms gain popularity, such as messaging apps (e.g., WhatsApp), future SOS research should consider how the present findings extend to such platforms as research suggests varying affordances can explain variations in political outcomes and processes (Rathnayake & Winter, 2017).
Third, opinion congruency for Clinton in society and for Trump on Facebook were associated with willingness to express support for a candidate across three different types (i.e., face-to-face, on Facebook, and in anonymous online settings) indirectly through FOI. Opinion congruency for Clinton on Facebook also had indirect relationships with willingness to express support on Facebook and in anonymous online settings. Given that the direct linkages between opinion congruency and willingness to express support were reduced to non-significance or their magnitude was weakened when FOI was taken into account, this finding adds more support for the role of FOI as a key mediating mechanism by which opinion congruency influences opinion expression (Kim, 2012; Noelle-Neumann, 1977; Noelle-Neumann & Peterson, 2004).
Finally, the observed differences in SOS between Clinton and Trump may be associated with their campaign rhetoric. As noted, during his campaign, Trump made controversial statements that some commentators equated with being sexist, racists, or which were otherwise publicly maligned. A casual supporter of Trump may have been less willing to express their support when they feared isolation, so-called “Shy Trumpers” (Guilford, 2016), whereas hardcore Trump supporters may still have spoken out. This possibility aligns with the notion of hardcore noncoformists who have high attitude certainty or speak out despite holding minority opinions (Matthes et al., 2010; Moy et al., 2001). Such reasoning may add to some interpretations of the difference between pre-election polling that pointed to a likely Clinton victory and the election outcomes.
Several limitations hamper definitive conclusions. First, we acknowledge the cross-sectional data analyzed in this study. Although the hypotheses were theoretically based on the SOS theory, causal inferences are limited, and we cannot rule out possible alternative models. For example, as citizens express support for a candidate, they might feel less afraid of being isolated from peers. Such opinion expression effects are plausible (Pingree, 2007). To offer more causally robust evidence, future work should consider multi-wave data collection, tracking the same individuals at multiple points in time. Second, the data were based on members of an opt-in online panel, and thus the extent to which the results can be generalizable to the larger American public is limited.
Third, our dependent variables, measured as willingness to express support for a preferred candidate, are not as fine-grained as they could be. Future work can improve the measures by examining willingness to express support for a specific candidate, instead of a preferred candidate, and also willingness to express opposition to a specific candidate. Fourth, the present study did not study antecedents of opinion congruency. For example, exposure to traditional and partisan media, and the structure of one’s social network on Facebook, such as how homogeneous or heterogeneous the network is (Stoycheff, 2016), might explain variations in opinion congruency, which in turn might affect FOI and willingness to express opinions.
Finally, although derived from previous research (Ho & McLeod, 2008; Ho et al., 2013), our measure of opinion congruency is not without limitation. As part of the calculation of the variable, respondents’ perceptions of the American public’s and their Facebook friends’ support for candidates were coded into one or the other category. For example, those who said 90% of Americans supported Trump and those who said 51% were both assigned one. Consequently, we could not capture a precise match between respondents’ levels of support for candidates and their perceptions of opinion climates. Future research should consider other measurement schemes such as the one proposed by Moy et al. (2001). It would also be useful to tap opinion climates in multiple contexts, not only the United States broadly but also the neighborhoods, cities, and/or states where people live, as their views of opinion climates may be shaped by the local and regional contexts they personally observe and experience on a day-to-day basis.
These limitations, however, should be balanced against the unique insights reported above. Our study provides support for the notion that SOS is alive face-to-face and is extended to online contexts. We have shown that FOI is a key mechanism underlying the process by which opinion congruency affects people’s willingness to express political opinions. The contentious nature of political campaigns, coverage, and discussions has major implications for American democracy. Fearing reprimand, citizens face a dilemma. They may engage in self-censorship or further isolate themselves from those with whom they disagree. Either route likely contributes to political polarization in US politics, further muting opinion expression among cross-cutting networks in favor of interaction among like-minded individuals (Pariser, 2012; Sunstein, 2001). While fear of hostile interaction with dissimilar others has long existed, citizens today can selectively expose themselves to confirmatory content (Tsfati, Stroud, & Chotiner, 2014), whether factually based or so-called “fake news.” Perhaps, no place is this truer than in today’s digitally networked society where people can “unfriend” those with whom they disagree, opting instead to join politically like-minded groups. The question remains as to what the larger effect of this trend will be on democracy in the so-called “post truth era.”
Footnotes
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: Funding for this study was provided in part by the Shepherd University Foundation.