News Recommendations from Social Media Opinion Leaders: Effects on Media Trust and Information Seeking

Abstract

Polls show a strong decline in public trust of traditional news outlets; however, social media offers new avenues for receiving news content. This experiment used the Facebook API to manipulate whether a news story appeared to have been posted on Facebook by one of the respondent's real‐life Facebook friends. Results show that social media recommendations improve levels of media trust, and also make people want to follow more news from that particular media outlet in the future. Moreover, these effects are amplified when the real‐life friend sharing the story on social media is perceived as an opinion leader. Implications for democracy and the news business are discussed.

The abundance of media choice has resulted in audiences drifting away from mainstream media (Prior, 2007; Stroud, 2011). Exacerbating the problem of an eroding mainstream news audience is a growing credibility problem. Despite the mainstream media's role in a functioning democracy, public attitudes toward the news have reached historic lows. Research has demonstrated a steady decline of public trust in the institution of news (Gronke & Cook, 2007). In fact, Pew Research (2012) has shown that over the past decade public credibility ratings for news have dropped across media and for all types of outlets; declines are more pronounced for broadcast news. Even noncommercial outlets such as NPR are not immune to the trend.

In an increasingly complex media environment, news outlets are finding it more difficult to retain an audience. Nonetheless, social media offers new possibilities for exposure to news; Facebook is one of those possibilities, and is one of the fastest growing tools for gathering news. Pew Research (2014) finds that half of all users – and at least 30% of U.S. adults – are consuming news on Facebook. More importantly, 78% report exposure to news on Facebook while using the site for other (often social) purposes (Pew Research, 2014). Now that news sharing on Facebook is beginning to reach a critical mass, this peer‐to‐peer process of digital news exposure warrants a closer look. As more people receive news from peers on social media, what implications does the news sharing process hold for news trust? More specifically, does receiving news through a friend on social media improve trust in the news outlet? Does news posted by a friend on social media inform information‐seeking behaviors? And are either of these relationships affected by who, meaning which friends, share the news story?

Utilizing an experiment and news recommendations from subjects' real‐life Facebook friends, this study examines 1) the relationship between exposure to news posted on social media and media trust; 2) the relationship between exposure to news posted on social media and information‐seeking intent; and 3) whether either of these relationships are affected by the perceived opinion leadership of the person sharing the news on social media.

Literature Review

The increased competition among news outlets in the postbroadcast environment has heightened negativity in the news, interpretive reporting styles, and attention to partisan news sources (Prior, 2007). One byproduct of increasing media competition is the erosion of public trust in the news (Ladd, 2011). Scholars have observed aggregate level declines of public trust in the news over the last few decades, transforming a once revered news profession to a subject of disdain and dissatisfaction (Gronke & Cook, 2007; Ladd, 2011). Since media trust and the perceived credibility of the news influence public opinion on important policy matters (Page, Shapiro, & Dempsey, 1987), this trend is worthy of scholarly attention.

Institutional Media Trust

First, the public has become less attentive to mainstream news altogether (Gilens, Vavreck, & Cohen, 2007) and increased negativity and incivility exacerbate the media's credibility problem (McCombs, Holbert, Kiousis, & Wanta, 2011). On an aggregate level, surveys point to a steady decline of public trust in the institution of news (Gronke & Cook, 2007; Ladd, 2010). People of all political persuasions are growing more dissatisfied with the news, as declining levels of media trust are observed for Democrats and Republicans alike (Ladd, 2013).

Nonetheless, in today's era of media abundance, the news credibility problem does not uniformly affect attitudes across all news outlets; for example, the public is more trusting of their preferred outlets for news and more trusting of local news outlets (Arceneaux, Johnson, & Murphy, 2012; Gronke & Cook, 2007). In other words, media trust is somewhat dependent on the characteristics of the news outlet. A 2014 survey released by Public Policy Polling found that Fox News is considered both the most and least trusted news source. This suggests that news outlet credibility varies according to partisan predispositions. In an era of media choice, the public selects outlets congruent with one's ideological leanings, and this process of selective exposure makes people distrusting of noncongruent outlets and perhaps too trusting of their preferred news outlets (Arceneaux et al., 2012). Beyond partisan preferences, demographic variables and political knowledge can also play a role in determining which news outlets one perceives as credible (Stroud & Lee, 2013).

Scholars have conceptualized media trust in a variety of ways, including trust in news content, trust in those delivering the news, and trust in media ownership (Williams, 2012). Since proximity and political predispositions play a role in influencing public assessment of news credibility, it is important to conceptualize media trust at an outlet‐level rather than at an institutional level. Meyer (1988) explored media trust by measuring public perceptions of accuracy, fairness, bias, story context, and trustworthiness. In other words, the study assessed the believability of the news organization. More recent research suggests that when conceptualizing media trust it is also important to consider how trust informs information‐seeking behaviors.

Trust and Information‐Seeking Behaviors

Research has demonstrated that news credibility matters beyond mere attitudinal trust and can also inform behavior, particularly information‐seeking behavior. For example, perceived news credibility has been observed as a mediating variable for cable news consumption (Stroud & Lee, 2013). Perceived trust in news also influences attention to news; Williams (2012) found that public trust in the reporter holds a positive relationship with attention to print news; and trust in the media corporation owning a given network holds a positive relationship with attention to its broadcast news programs. Given these patterns, it is reasonable to expect that trust in the person sharing the news story would also affect information‐seeking behaviors and how the receiver engages with the content.

News trust can also influence political behavior. As aggregate levels of media distrust increases, the public is more reliant on partisan cues when casting a vote (Ladd, 2010). Furthermore, when the public holds low levels of media trust it is more likely to abandon mainstream news outlets for partisan sources. Those most distrusting of the media are also the most prone to resisting messages conveyed in mainstream news (Ladd, 2011/2013). Consequently, declining levels of media trust fosters a heightened perception that the current political climate is a polarizing one.

Just as media trust and perceived credibility of an outlet affects attentiveness to news, the credibility of those who share news also plays an important role in how that information is received. This area of influence is best explained by the literature concerning the two‐step flow model of communication and the influence opinion leaders hold over the general public.

Two‐Step Flow and Opinion Leadership

Two‐step flow emerged from Lazarsfeld, Berelson, and Gaudet's (1948) pioneering survey fieldwork conducted during the 1940 presidential election. Lazarsfeld et al.'s (1948) goal was to determine individual, interpersonal, and media factors that influenced “how and why people decided to vote as they did” in 1940 (p. 1). While the researchers did not formally predict how political information would pass from sources of mass communication—like newspapers—to and among message receivers, Lazarsfeld and colleagues (1948) discovered that such information diffused in a unique, “two‐step” process.

First, campaign information was received mainly by a minority of politically involved and knowledgeable “opinion leaders.” Opinion leaders were interested in the presidential campaign, routinely used media, and were perceived to be influential, trustworthy people to whom others could turn to for information and advice; thus, the second step of two‐step flow involved transmission of political information from opinion leaders to less engaged “opinion followers.” As Lazarsfeld and colleagues (1948) explain, two‐step flow is the process whereby “[political] ideas [sic] flow from radio and print to the opinion leaders and from them to the less active sections of the population” (p. 151, italics original). This research (Katz & Lazarsfeld, 1955), which found that people are more influenced by interpersonal communication than media content is especially prescient today given that the public is becoming less attentive to news. Moreover, since social uses of sites such as Facebook trump information‐seeking uses, interpersonal communication and perceived opinion leadership may help “opinion followers” make sense of and evaluate the news content exchanged through social media.

As research on two‐step flow has evolved, the conceptualization and operationalization of “opinion leadership” has been refined. Lazarsfeld et al. (1948) suggested that opinion leaders were politically interested, engaged, knowledgeable, and trusted sources of information within their local social networks. Subsequent research suggested individuals could be defined as opinion leaders only if they were similar in social position as those they influenced, perceived to be competent about the topic under discussion, and held many diverse contacts and frequent discussions within his or her own social network (Katz, 1957; Katz & Lazarsfeld, 1955). These underlying dimensions of opinion leadership were often assessed using questionnaires and self‐reports. For example, opinion leadership was measured using items that tapped “self‐designated” influence, such as asking whether the respondent recently tried to persuade a friend's vote choice or consumer decision (Katz & Lazarsfeld, 1955; Lazarsfeld et al., 1948).

More recent measures of opinion leadership incorporate indicators of issue‐specific discussion, self‐perceived ability to persuade, “personality strength” and frequency and depth of opinion‐giving(Childers, 1986; Feick & Price, 1987; Weimann, 1991; 1994; Weimann, Tustin, van Vuuren, & Joubert, 2007; for a review, see Nisbet & Kotcher, 2009). In general, opinion leaders are defined as such if they are engaged and competent individuals who are viewed as honest and trustworthy by opinion followers, with whom they frequently discuss issues.

New Theoretical Challenges to Two‐Step Flow

Social and technological changes have created new theoretical challenges to the two‐step flow model broadly and the role of opinion leadership specifically (Bennett & Manheim, 2006). One theoretical challenge stems from the advent of the Internet and online microtargeting techniques that allow professional communicators to produce and disseminate messages that appeal directly to individuals. This change, according to Bennett and Manheim (2006), obviates the role opinion leaders play in receiving and interpreting messages for opinion followers. At the same time, Internet‐based microtargeting shifts more control over message content to communication professionals, who often have a strategic media effects goal in mind.

Second, Bennett and Manheim (2006) suggest that individuals are becoming increasingly socially isolated from one another, reducing the agency of opinion leaders in the two‐step flow process, and raising the potential for mass media to have powerful and uniform effects. If individuals are more physically isolated, it is more likely that isolated audience members select their own unique blend of media content and rely on their own issue interpretations. Rather than a two‐step communication process, Bennett and Manheim (2006) suggest a more fitting description would be a “one‐step” communication process in which media messages have direct effects on isolated receivers, bypassing opinion leaders altogether.

Yet this perspective and its concomitant theoretical challenges may not fully address the type of peer influence that transpires on Facebook and other popular social networking sites (Mutz & Young, 2011). While it may be the case that individuals are more physically separated than in previous decades, social network sites facilitate nuanced mediated‐interpersonal communication over vast distances, providing novel settings for leader‐follower interactions. For example, social networking site users often rely on friends for product recommendations (Forbes & Vespoli, 2013). Considering that today's media environment presents news consumers with expansive news choices, “one extremely important way [individuals] decide what to pay attention to is through recommendations that reach them through their online social networks” (Mutz & Young, 2011, p. 1038). This finding suggests that in a fragmented media environment, opinion‐leaders play increasingly important roles in facilitating exposure to news.

Opinion Leadership and Processing News on Facebook

News media content is becoming an increasingly salient component of Facebook and other social network sites. Approximately 47% of Facebook users – or 30% of the U.S. adult population – say they receive news on the site (Pew Research, 2014). Importantly, Facebook users can receive news either through subscribing to a news organization's feed, or through Facebook friends who create posts with linked news content. This is a unique feature of news delivery on Facebook in that both professional journalists and friends act as information gatekeepers, vetting the significance and relevance of news content, and raising the possibility of new opinion leader‐follower dynamics that remain unexplored in the communication and psychology literatures. To date, few studies have investigated opinion leadership and message processing in mediated‐interpersonal contexts. Specifically, evaluating the credibility of social media remains difficult due to the myriad of sources content can come from (Johnson & Kaye, 2014). Thus, little is known about how users evaluate posts with linked news content.

Emerging research has shown that sharing news on Facebook has some desirable engagement effects for the messenger; these effects are enhanced when content is shared publically on the “wall” (versus privately messaged) and when the messenger poses a question to his/her network when sharing the content (Oeldorf‐Hirsch & Sundar, 2012). Thus, social mechanisms of the website seem to drive news engagement and influence. Although research has shown that information credibility on social media improves when the person sharing the content holds values and predispositions which match the recipient (Metzger et al., 2010), Facebook offers an especially unique domain to explore news credibility and opinion leadership given that Facebook networks are less homogenous than other social networking sites (Hanson et al., 2010). Therefore, one way that users may judge Facebook posts with linked news content is by evaluating the message sender.

Perceptions that a message sender is honest and expert may act as heuristic cues (Chaiken, 1980), telling the user whether content received from that source is useful and trustworthy. Since Facebook users are often confronted by abundant social and news information from a wide variety of sources of varying credibility within the same news feed, they may feel the need to employ cues to reduce the cognitive burden of deciding how much to trust these sources. One potentially useful and always available cue about the trustworthiness of these media sources is the trustworthiness of the friend who shared the link. If that friend is believed to be highly knowledgeable and trustworthy about public affairs, these positive evaluations may transfer to the media source. If a friend posts a link to a story from a particular media source, one can assume at least that the friend uses that source, perhaps also that they trust it, and perhaps even that the source is oriented towards people similar to the friend in some way.

Given what we know about the public's growing distrust for the media (see Gronke & Cook, 2007), compared to receiving a Facebook post directly from a traditional news source, individuals may be more likely to perceive information as trustworthy if it is shared through a Facebook friend. Following the patterns noted in two‐step flow, if the user considers the friend to be an expert and honest opinion leader, the user may then be more likely to judge the linked news content of the post as being trustworthy. Lastly, given the strong relationship between news trust and information‐seeking behavior (see Stroud & Lee, 2013; Ladd, 2011), we would expect these assessments to guide future engagement with the specific news outlet being shared. Stated formally, this study posits the following predictions:

H1: A news story recommended by a friend on social media will increase trust in the news outlet, compared to receiving the same story directly from the traditional news outlet where it appears.

H2: There will be an interaction between a friend's news recommendation and perceptions of the friend as an opinion leader predicting trust in the news outlet.

H3: A news story recommendation from a friend on social media will increase intent to seek future information from the news outlet, compared to receiving the same story directly from the traditional news outlet where it appears.

H4: There will be an interaction between a friend's news recommendation and perceptions of the friend as an opinion leader predicting intent to seek information from the news outlet.

Method

We conducted an online survey experiment intended for multiple purposes beyond this paper. The data collection consisted of a 2 X 3 X 2 experimental design: a fully crossed 2 (social recommendation of news vs. mainstream media) by 3 (adjudication in favor of Republicans, Democrats or no adjudication) by 2 (polarized elite or bipartisan elite). Following the pretest questionnaire and exposure to one of the stimuli, respondents answered posttest questionnaires measuring dependent variables and the moderator. The entire design was repeated in two issue contexts, which we used to check whether any effects were contained to one topic of news. Both contexts were state‐level political issues without very strong national partisan divisions. The two contexts were tax credits for the film industry in the state where the study took place and a ban on smoking in casinos in the state. These issues were selected for their perceived salience to younger, local news audiences; these topics required little political sophistication to follow and the dominant political parties have yet to seize issue ownership on either topic.

Sample

Given that younger news audiences are turning to social media to follow current events at a higher rate than other demographic groups (Holcomb, Gottfried, & Mitchell, 2013), we recruited 364 undergraduate students at a large, public southern U.S. university using a subject pool to take a survey‐based experiment. The sample was primarily comprised of females (80.0%), with fewer males (19.7%) and other gender (0.3%) participants. Age of the sample ranged from 18 to 36 years (M = 20.20, SD = 1.80). Participants were asked to identify their year in school from freshman to graduate student (M = 1.6, SD =1.13, range = 0 to 4). The sample was mainly comprised of White students (82.6%), with African American (8.0%), Asian (3.0%) Latino (2.2%), Pacific Islander (0.6%), and other race or ethnicity (3.6%) comprising the remainder of the sample. Participants were also asked to self‐report their political orientations on 7‐point Likert scales, from strong Democrat to strong Republican (M = 4.50, SD = 1.29, range = 1 to 7) and strong liberal to strong conservative (M = 4.22, SD = 1.29, range = 1 to 7).

Social Recommendation Manipulation

Before pretest measures, participants were asked to “login with Facebook.” We used Facebook's application programming interface (API) to obtain participants' permission to gather publicly available information from their Facebook accounts. We obtained this permission from all participants before randomly placing participants into one of two categories: Social recommendation (n = 191) and no social recommendation (n = 181). Participants who received the social recommendation were exposed to a fabricated news story about one of two local news contexts and told that it was recommended by one of their real‐life Facebook friends. The fabricated news story was provided to them with the following prompt: “One of your Facebook friends shared a link to a news story relevant to this study recently. Please click the link they posted below to read the story.” Under this prompt was a link with the selected friend's actual Facebook picture and the friend's (fabricated) comment, “People should pay more attention to this kind of thing.” The real‐life Facebook profile photo, comment and link to the story were ‘laid out’ to resemble a Facebook post (see Appendix A).

For all participants, we chose a real‐life Facebook friend, but included the recommendation only for participants in the stimulus condition. Participants in the no‐recommendation condition saw only a URL link, with no recommendation to the news story (see Appendix B). These participants did not see the picture and name of their Facebook friend until after outcomes were measured. For participants in the social recommendation treatment, we used Facebook's API to choose the particular friend who “recommended” the story by selecting a friend who had a previous history of frequent Facebook interactions with the participant. The person with the most likes and comments from the participant, and who had their address listed as within the same U.S. state the experiment was conducted, was selected as the purported recommender.11 Although the Facebook API would have allowed us to directly collect a variety of information about participants as well as the purported recommender, such data would have been problematic for participant privacy and data quality. The proportion of missing values is high in Facebook profile data retrieved from the API, because users are not required to answer all profile fields and because of privacy settings that cause profile fields to appear blank even when they are not. Further, if data from the friend's profile is assumed to accurately reflect that person's actual traits, participants' perceptions of that friend are more theoretically relevant in terms of accepting a cue from that person. Instead of attempting to use the API to collect data about the purported recommender, we rely on the participant's self‐reported perceptions of that person.
Using a person who has frequent social media interactions with the participant was intended to increase the external validity of the manipulation. Using a person residing in the same state was deemed necessary to make the stimulus believable, since the news story stimulus was specific to that state.

The social media friend recommended the story in this manipulation, but the story itself was made to appear as if it came from a local newspaper, with the newspaper's masthead. Participants who were randomized into the no social recommendation group were taken to the same story as the social recommendation group, but saw only the masthead of the local news organization above the text of the story, with no mention of social media. After the study's post‐test, we debriefed participants, informing them that the news article and their friend's comment were both fabricated by the researchers.

Other Experimental Manipulations

Two other factors were part of the experimental design but are outside the scope of this analysis. The first of these factors manipulated adjudication of factual disputes, given that adjudication in news has been found to influence perceptions of news quality (Pingree et al., 2014). This manipulation included three possible treatments: in favor of Republican claims, in favor of Democratic claims, and no adjudication. The second of these factors manipulated elite polarization vs. bipartisanship manipulation, following Druckman et al. (2013). We controlled for these two factors in the analysis and found no effects; other work discusses their effects on other outcomes.

Dependent Variable Measures

Looking to the research of Meyer (1988), who conceptualized media trust at the outlet level, we develop a 5‐question scale assessing participant attitudes toward the news outlet providing the stimuli. We adopt measures from Meyer (1988) using a 7‐point Likert type scale (1 = strongly disagree, 7 = strongly agree). These measures assess whether the news outlet is: 1) trustworthy (M = 4.94, SD = 1.13), 2) accurate (M = 4.99, SD = 1.04), 3) unfair (M = 4.87, SD = 1.25), 4) tells the whole story (M = 4.30, SD = 1.23), and 5) biased (M = 4.51, SD = 1.32). Measures for the Outlet Trust Scale (M = 4.72, SD = .91) demonstrate strong inter‐item reliability, α = .817.

Because the work of Ladd (2011, 2013) has shown that perceived news credibility can influence behavior and because research has also demonstrated a relationship between news trust and news‐seeking habits (Williams, 2012; Stroud & Lee, 2013), we include a second dependent variable measuring the intent to seek future information from the news outlet using two measures. Using a 7‐point Likert type scale (1 = Very Unlikely, 7 = Very Likely), we create an information‐seeking measure (M = 4.79, SD = 1.21, r = 0.73) based on items that asked:

1. Based on the story you saw, how likely would you be to seek out local news from this news outlet in the future? (M = 4.74, SD = 1.29).
2. If you saw another story from this same news outlet, how likely would you be to read it? (M = 4.85, SD = 1.31).

Moderator: Respondents' Perception of a Friend's Opinion Leadership

The perceived opinion leadership of a respondent's Facebook friend was measured with a two‐item index of measures with the prompt “Please answer the following questions about [name of a real‐life friend here], your Facebook friend.” (Please see the section above regarding the social recommendation manipulation for information about how we chose which of the respondent's social media friend to feature for this question.) We solicited the respondent's perception of the opinion leadership of the same friend that we used for the social recommendation manipulation. For those respondents who were randomized into the no social recommendation condition, we still measured opinion leadership of the friend that we would have used in the case that the respondent had been in the other condition. This was necessary to allow this variable to be used as a moderator ofthe experimental effects. Otherwise, missing values for the moderator would cause all participants in the no social recommendation condition to be dropped from analyses.

The two items in the opinion leadership measure (M = 4.62, SD = 1.33, r = 0.67) asked respondents to indicate agreement with the statements “This person is well informed about politics and current events” (M = 4.40, SD = 1.58) and “This person is honest about politics and current events” (M = 4.84, SD = 1.33). While there are a variety of alternative methods to assess opinion leadership (see Nisbet & Kotcher, 2009), these two items—in combination with frequency of discussion—are foundational to almost every measurement of the opinion leadership construct.

Analysis

Main effects were tested with OLS regression models that did not include the interaction term (model 1 in both tables), and interactions were tested using otherwise identical models adding the interaction term (model 2 in both tables). We used the PROCESS macro for SPSS (Hayes, 2012) to test the interactions and to output effect estimates and confidence intervals at each level of the moderator (used to construct the interaction plots in figures 1 and 2). Although the moderator was measured in the posttest, it is an exogenous variable that is not affected by the experimental manipulations. In a model predicting perceived opinion leadership, none of the manipulated factors were related, including social recommendation (Beta = 0.148, p = .292).

Figure 1

Interaction Effect: Opinion Leadership and Social Recommendation on News Trust

Figure 2

Interaction Effect: Opinion Leadership and Social Recommendation on Information Seeking

Results

The data support Hypothesis 1 and 2. We found a main effect of a social media cue, compared to a traditional news source cue, on outlet trust (Beta = −.8395, p = .0160). There was a significant interaction effect between receiving a social cue from a friend about the news and opinion leadership (Beta = .1885, p = .0093). Please see table 1. The nature of this interaction reveals that recommendations from friends who are perceived as very poor opinion leaders have a negative effect on respondents' trust in the news outlet that was recommended. On the other hand, recommendations from those perceived to be excellent opinion leaders have a positive effect on news outlet trust (see figure 1).

Table 1. OLS Regression Predicting Trust in News Outlet

	Model 1		Model 2
	b	(SE)	b	(SE)
Adjudication	−0.03	(0.06)	−0.02	(0.06)
Polarization	0.06	(0.10)	0.07	(0.09)
Story Context	0.08	(0.10)	0.08	(0.09)
Recommendation	0.03	(0.10)	−0.84** p < .05. Between‐group means tests indicated no significant demographic effects—for gender or age—on the outcome variable.	(0.35)
Opinion Leader	0.13****** p < .001	(0.04)	0.05	(0.05)
Recommendation X Opinion Leader	—		0.19**** p < .01	(0.07)
Constant	4.09****** p < .001	(0.19)	4.43****** p < .001	(0.23)
N	364		364
R^2	0.04		0.06

• Note:
• *** p < .001
• ** p < .01
• * p < .05. Between‐group means tests indicated no significant demographic effects—for gender or age—on the outcome variable.

The data also support Hypothesis 3. Controlling for the interaction, there was a significant main effect of receiving a news recommendation from a friend on likelihood to seek information from that same outlet (Beta = −1.3498, p = .0035). Please see table 2. We also find support for Hypothesis 4: We observe a significant interaction effect between receiving the story through a social media cue and receiving the cue from a friend perceived as an opinion leader (Beta = .2829, p = .0033). The nature of this interaction reveals that recommendations from friends who are perceived as strong opinion leaders are associated with an increase on respondents' desire to engage in additional information‐seeking behaviors. On the other hand, recommendations from those perceived to be poor opinion leaders have the inverse effect on information‐seeking intent (see figure 2).

Table 2. OLS Regression Predicting Intent to Seek Additional News

	Model 1		Model 2
	b	(SE)	b	(SE)
Adjudication	−0.14	(0.08)	0.01	(0.07)
Polarization	0.13	(0.13)	0.15	(0.13)
Story Context	0.14	(0.13)	0.14	(0.13)
Recommendation	−0.04	(0.13)	−1.35**** p < .01	(0.46)
Opinion Leader	0.15**** p < .01	(0.05)	0.02	(0.06)
Recommendation X Opinion Leader	—		0.28**** p < .01	(0.10)
Constant	4.03****** p < .001	(0.26)	4.55****** p < .001	(0.31)
N	364		364
R^2	0.032		0.055

• Note:
• *** p < .001
• ** p < .01
• * p < .05. Between‐group means tests indicated no significant demographic effects—for gender or age—on the outcome variable.

The data do not show an effect of story context on the results. There is no significant effect of whether the manipulations were on stories about film tax credits or on stories about a casino smoking ban. More importantly, there is no difference in the shape of the interaction between opinion leadership and social news recommendation in the two contexts. Functionally, the film tax credit context replicates the results found in the casino smoking ban study. The data do not show any effects of the adjudication or bipartisanship manipulations. Results for these manipulations were very sensitive to model specification. No one‐tailed significant effects of either polarization or adjudication in news articles emerged. We also tested alternative model specifications that included respondents' partisanship as a control and as an interaction with the manipulations: no pattern emerged. Note that the effects in the study come from characteristics of where the article was found (social media recommendation of mainstream news article vs. plain old mainstream news article) and who recommended it (opinion leader or not) rather than the content of the news article (context, adjudication, and polarization).

Discussion

Using an experiment that tested the effect of recommendations to read a news story from a person's real‐life social media friend, we find that social recommendation is related to the likelihood of seeking news from that outlet in the future. In particular, social recommendations from people perceived as quality opinion leaders led to an increase in outlet trust. Alternatively, social recommendations from people perceived as poor opinion leaders decreased the reader's outlet trust. These results extended beyond trusting a news outlet to indicators of future behavior. Receiving a cue from a strong opinion leader increased attitudes related to future information seeking behavior. The opposite effect was found amongst readers who received a cue from a person perceived as a poor opinion leader. These findings are important for democracy for several reasons.

The increasing diversification of the modern media market splits media consumers into atomized niches. However, social network sites such as Facebook provide a place where users may see many types of content, both entertainment‐based and information‐based content. This is important given that attention to public affairs news is often the result of incidental exposure (Prior, 2007). The finding that news shared by a friend on Facebook is perceived as more trustworthy than stories received directly from the media outlet may inform how news organization elect to grapple with increasingly inattentive news audiences in the age of digital and social media. Thus, understanding the effects of social recommendations on Facebook, assessments of news trust in particular, gives insights into how the 1.26 billion Facebook users interact with news organizations and one another.

Social interactions between members of the public also facilitate democracy (Mutz, 2005). Social media users interact with other users in a politically heterogeneous way at higher rates than social media interactions between representatives of news organizations (Conover et al., 2011). Consequently, social recommendations can help to mitigate the effects of moderate groups receiving less prominent coverage in mainstream news (McCluskey & Kim, 2012). Moreover, these news‐sharing interactions may not only improve levels of media credibility but also hold implications for information‐seeking habits.

In line with the research of Ladd (2011, 2013), this study suggests that news trust guides behavior. We find that exposure to news on social media increases intention to continue following news from the respective media outlet and increases likelihood to read future stories from the outlet. In other words, socializing news habits have direct implications for improving not only public trust in the news but democratically desirable behaviors. This research also reinforces the importance of opinion leadership in the new media environment; those perceived as high on opinion leadership amplify the effect of increasing intention to follow future news covered by the traditional media outlet. In other words, both news professionals and opinion leaders shoulder the burden of informing and educating the public in the age of digital journalism. Communication technology provides new affordances for news‐sharing that may help disseminate public affairs information deemed more credible and more worthy of attention than content coming directly from traditional news outlets. Should these patterns continue to hold, opinion leadership may help solve the news media's worsening credibility problem.

From a theoretical perspective, this study adds to the emerging literature on news sharing on social media and situates this phenomenon to existing models of communication: the two‐step flow. To date, most research in this area has studied those sharing news content online and the predictors of news‐sharing behavior with little exploration into how that information is received. This study demonstrates that interpersonal communication matters to news credibility and that perceived opinion leadership bolsters news credibility and information‐seeking intent. It also suggests a more dynamic interpretation of the two‐step flow in a media context with numerous sources by emphasizing the diffusion of trust in media sources. Trust is typically conceptualized as static in the two‐step flow, and is treated as part of the definition of who is an opinion leader. Our findings suggest that individuals may learn to trust certain media outlets through opinion leaders, but then may directly use those outlets in the future, potentially then acting as opinion leaders themselves.

Our findings do not support Bennett and Manheim's (2006) one‐step flow model of communication. This model contends that in a communication environment with high levels of choice more people receive information through a one‐step flow using niche media; however, this work predates most social media. The research of Singer (2014) found that more news outlets are offering sharing mechanisms for social media through their digital news sites and that these affordances have created a “two‐step gatekeeping process” in which news audiences re‐disseminate content to “secondary” audiences. In other words, users are not only sharing content but are encouraged to do so, and in ways which underscore the relevancy of two‐step flow. As technology emerges, scholarship should revisit existing theoretical constructs to make sense of the evolving pathways of computer‐mediated information. Broadly, this research reinforces the importance in revisiting foundational models in mass communication to understand how these processes – including two‐step flow – play out across new technologies. In an era of information abundance and digital news sharing, social media users resort to social recommendation cues and perceptions of opinion leadership to assess information credibility; however, we should note that this study presents a simplified conceptualization of opinion leadership.

This study assesses opinion leadership primarily at the “informational” level, using a 2‐item measure tapping into perceived level of politics and current affairs knowledge and honesty about politics and current affairs; however, earlier research considers multiple dimensions of opinion leadership (see Rogers & Cartano, 1962). Moreover, it is important to acknowledge that we measure opinion leadership broadly through a “polymorphic” lens which assumes opinion leaders are opinion leaders on multiple issues whereas earlier research measures this process according to issue‐specific domains – or a “monomorphic” approach (Merton, 1957). This is an important distinction given that the roles of opinion leaders and opinion seekers is more fluid than this research design suggests, meaning people may switch roles between leader and seeker depending on the issue. This limitation is somewhat mitigated by the consistency of findings across both issue domains: Interaction effects for opinion leadership and the social recommendation cue for the film tax credit story paralleled the interaction for the casino smoking ban. Given that our study also explored partisan adjudication, we carefully selected issues that neither party has seized ownership of. The dearth of media coverage and party attention to these issues hindered our ability to measure perceived opinion leadership within these issue‐specific domains. Nonetheless, future research should account for opinion leadership at both levels.

As Flynn et al. (1996) critique, opinion leadership may more accurately measure the propensity to participate in “social communication” rather than measuring opinion influence within a two‐step flow of communication. The measures used herein for capturing opinion leadership are by no means exhaustive. For example, using a student sample limited the ability to measure diversity of contacts, as participants came from similar life stages and a relatively homogenous public university. Capturing additional dimensions of opinion leadership is challenged by the validity problems of self‐reported data; we work around this by having participants report perceived leadership characteristics of their peers.

Given that this study uses stimuli reflecting local news issues covered by a local media outlet, future research could also consider exploring these relationships across national‐level news and national‐level outlets. Since the literature on media trust suggests that the public holds less trust for the institution of news and national‐level outlets relative to local news (Moy & Pfau, 2000; Gronke & Cook, 2007), future research could test whether these relationships hold when using stimuli from a national‐level news outlet. It is also worth noting that scholars have recently found that the recency of social media activity enhances the perceived credibility of the message sender (Westerman et al., 2014); although we elected to utilize a friend with frequent social media interactions with the participant to increase the external validity of the manipulation, future research should test this across interaction levels. Moreover, we caution that these results are not necessarily generalizable to all social networking sites, and although age is not always predictive of social networking credibility, the work of Johnson and Kaye (2014) suggests a user's reliance on social media does predict one's perceived credibility of information shared on these sites; this is a variable that should be controlled for in future research.

Nonetheless, this research lays important groundwork for assessing the problem of declining levels of public trust in the news and the implications of news‐sharing processes on social media. Moreover, using an experimental design delivering stimuli through real‐life friends of participants offers a novel approach to improving external validity. This research also emphasizes the importance of considering opinion leadership and revisiting elements of the two‐step flow of communication, as news‐sharing behaviors on Facebook not only hold implications for news outlet trust but also underscore the importance of interpersonal communication in creating a public more attentive to news. With a little help from our (opinion leader) friends, citizens may be able to reconnect with journalism outlets and improve their readiness to participate in an informed democracy.

Note

1 Although the Facebook API would have allowed us to directly collect a variety of information about participants as well as the purported recommender, such data would have been problematic for participant privacy and data quality. The proportion of missing values is high in Facebook profile data retrieved from the API, because users are not required to answer all profile fields and because of privacy settings that cause profile fields to appear blank even when they are not. Further, if data from the friend's profile is assumed to accurately reflect that person's actual traits, participants' perceptions of that friend are more theoretically relevant in terms of accepting a cue from that person. Instead of attempting to use the API to collect data about the purported recommender, we rely on the participant's self‐reported perceptions of that person.

A. Appendix: Social Recommendation Manipulation Example

B. Appendix: Nonsocial Recommendation Condition

Biographies

• Jason Turcotte is Assistant Professor in Communication, California State Polytechnic University, Pomona, 3801 W. Temple Ave., Pomona, CA, 91768. E‐mail: [e-mail]

• Chance York is Assistant Professor of Mass Communication, Kent State University, School of Journalism and Mass Communication, 550 Hilltop Drive, Kent, OH, 44242. E‐mail: [e-mail]

• Jacob Irving is a law student at Louisiana State University, Paul Hebert Law Center, 1 E. Campus Drive, Baton Rouge, LA, 70803. E‐mail: [e-mail]

• Rosanne M. Scholl is an assistant professor in the Manship School of Mass Communication, Louisiana State University, Baton Rouge, LA, 70803. E‐mail: [e-mail]

• Raymond J. Pingree is an assistant professor in the Manship School of Mass Communication, Louisiana State University, Baton Rouge, LA, 70803. E‐mail: [e-mail]