News by Popular Demand: Ideological Congruence,Issue Salience,and Media Reputation in News Sharing 1

Abstract

Social media news sharing has become a central subject of scholarly research in communication studies. To test current theories, it is of an utmost importance to estimate the meaningful parameters of news sharing behavior from observational data. In this article, we retrieve measures of ideological congruence, issue salience, and media reputation to explain news sharing in social media. We describe how the proposed statistical model connects to different strands of the news sharing literature. We then exemplify the usefulness of the model with an analysis of the relationship between ideological congruence and issue salience. Results show that if ideology and salience correlate with each other, the preferences of ideologues (i.e., users who give higher weight to ideological congruence) will be overrepresented in observational data. This will result in the heightened perceptions of polarization. We test the performance of the model using data from Brazil, Argentina, and the United States.

Keywords

news sharing social media issue salience

Introduction

Why do users share news articles² with social media peers? How important are ideological considerations, issue salience, and the reputation of a news organization in the decision to embed and share news hyperlinks? News sharing is a complex phenomenon that combines attitudinal and reputational traits with the broad effects on the propagation of news, exposure to news, and on journalistic professional practices (Boehmer and Tandoc 2015; Choi and Lee 2015; Kümpel et al. 2015; Strömbäck et al. 2013). As interest grows to explain the mechanisms underlying news sharing, scholars require workable empirical strategies to measure sharing behavior using the observational data.

This article advances a statistical model to derive the news sharing behavior from the observational social media data. Our method decomposes news sharing in three sets of parameters: (i) user-specific issue salience, (ii) news media reputation; and (iii) ideological congruence. These three sets of parameters align with three different and well-established research areas in the communication’s literature—described in Kümpel et al. (2015)—which focus on the individual incentives to share content (Type I); the features of the content being shared (Type II); and group differences in network effects (Type III).³ A similar classification was presented by Scheufele and Nisbet (2013), who write “(i)n many ways, selectively attending to some messages over others, based on perceptions of source credibility⁴ [Type II], ideological congruence [Type III], or issue-specific interest [Type I], is what enables us to efficiently sift through large amounts of information” (Scheufele and Nisbet 2013: 45). In this article, we show how these three distinct and rich literatures can be empirically evaluated as features of a common sharing matrix in the observational data.

To exemplify the usefulness of our statistical approach, we use Twitter data and test the relationship between ideological congruence and issue salience (Weaver 1991). Researchers have documented a positive correlation between political position-taking and news sharing (Delli Carpini 2004; Strömbäck et al. 2013), tested using survey and experimental data (Oosterhoff et al. 2018). If ideological congruence and user specific issue-salience⁵ are positively correlated, ideologues would share news at higher rates than non-ideologues. The result would be that the preferences of ideologues would be overrepresented in observational data. We test for this relationship, finding support for a positive and statistically significant correlation between ideological congruence and issue salience in the observational data. We test our statistical approach with the data from the election of Bolsonaro in Brazil, the disappearance of the activist Santiago Maldonado in Argentina, and the TravelBan enacted by Donald Trump in the United States.⁶

The organization of the article is as follows: in the next section, Section 2, we describe the importance of retrieving sensible behavior information from social media embeds to test current theories of news sharing using the observational data. In Section 3, we present a formal description of the statistical model and exemplify how to interpret the parameters of interest. We then, in Section 4, discuss the current research that argues why we should expect a positive correlation between ideological congruence and issue salience, a hypothesis that we can test using our estimation strategy. In Section 5, we describe our parameter estimates. Finally, in Section 6, we report a support for the hypothesized relationship between ideology and issue salience. We show that ideologues are not only more likely to share content from a narrower set of news outlets but they are also unconditionally more likely to share more news.

Our results contribute to the ongoing debate about the formation and dynamics of echo chambers in social media (Barbera 2020; Flaxman et al. 2016; Guess 2021; Sikder et al. 2020). The positive correlation between ideological congruence and issue salience explains why the content preferred by ideologues is overrepresented in social media networks. Even if partisan sorting is modest (Guess 2021), users may still perceive echo chambers when ideology and salience correlate with each other (Bail 2021).

Theory: A Statistical Description of News Sharing in Social Media Data

Why Is it Important to Develop a Statistical Model of News Sharing?

cIn a recent article, Kümpel et al. (2015) conduct a meta-analysis of 461 articles on news sharing published between 2004 and 2014. The authors note that the number of news sharing articles published every year in peer-reviewed journals increased from ten in 2004–2005 to over two hundred in 2013–2014. Nowadays, news sharing is a key phenomenon that shapes journalistic and editorial practices (Blanchett Neheli 2018; Russell 2019); forges reciprocal ties among journalists (Hanusch and Nölleke 2019), and is an important source for journalistic content (Von Nordheim et al. 2018). Indeed, with the rise of social media, news sharing and news sharing behavior have become central topics in the communication’s literature (Lee and Tandoc 2017).

The emerging research on news sharing centers on three distinct but interrelated phenomena is summarized by Kümpel et al. (2015) in their review of the literature. The authors group different strands of news sharing research into three types or families: First, there are user-level traits that explain the users’ preference for sharing content (“why do some users share more news than others?”). Second, Kümpel, Karnowski, and Keyling describe a burgeoning literature that focuses on content features that increase the likelihood of news being shared (“why is some content shared more frequently by users?”). Finally, a third strand of research focuses on group-specific features that segment news sharing (“why do these groups of users share these particular sets of news”?).

These different concerns yield a vast number of literature studies on the subjective, social, rational, and emotional factors that explain the individual behavior (Boehmer and Tandoc 2015; Boyd et al. 2010; Rudat et al. 2015; Scheufele and Nisbet 2013); the content-level features that facilitate news sharing (Karnowski et al. 2020; Macskassy and Michelson 2011; Suh et al. 2010; Trilling et al. 2017; Wang et al. 2012); and the group differences (ideology, partisanship, and network structures) that segment the audiences (Barbera 2020). As we propose below, these three different agendas correspond to distinct features of a news-sharing matrix of observational social media embeds, which may be jointly estimated and serve communication scholars working on the subject.

Modeling News Sharing Behavior: An Intuitive Description

We begin with a formal description of the three sets of parameters and their connections to the existing literature. Consider an individual user $u_{i}$ who observes a thread of social media posts with embedded hyperlinks to news articles. She/he has a natural “click” rate (i.e., a trigger finger) and wants to connect with her peers (Waruwu et al. 2020; Weeks and Holbert 2013). Before sharing a post, however, she/he asks⁷ the following three questions: (1) Is this content worthy of my attention?; (2) Is this news created by a reputable source?; and (3) Do I agree with the content of this news? Each positive response increases the likelihood that she/he will share the news while each negative response reduces this likelihood. The important question we hope to elucidate is, how much do each of these three incentives (i.e., respectively issue salience, reputation, and ideological congruence) matter for any given set of users that are active on social media and engaged during a particular time and on a given topic?

Much of the earlier research on news sharing focused on the individual level incentives to share news. The Users and Gratification Approach (UGA) is a prime example, later extended to more general theories describing the norms, motives, and attitudes of users (Karnowski et al. 2018). We define this dimension of news sharing as issue salience:

Def.1: Issue salient is the utility of sharing content on an issue that the user considers more important (row feature).

A different mechanism (and literature) explains news sharing by focusing on the characteristics that make some content more broadly liked (Type II literature). A prime determinant of sharing is the reputation of the news organizations (Karnowski et al. 2020; Macskassy and Michelson 2011; Suh et al. 2010; Wang et al. 2012). In their recent article, “Worth to Share?”, Karnowski et al. (2020) posit that “we need to learn more about the content characteristics that determine whether news is shared widely or is not disseminated at all” (pg. 60). In related work, Trilling et al. (2017) provide evidence of the effect that the importance of a news agency’s reputation has on news sharing. The reputation dimension of news sharing, borrowing the term widely used in peer-to-peer (P2P) networks, explain differences due to the brand name of different organizations:

Def.2: Reputation is the utility of sharing content from news outlets recognized as higher quality by peers (column feature).

Finally, a broad literature shows that different groups of users are more likely to share content that is ideologically congruent and from news organizations whose editorial lines are ideologically closer to them (Barberá 2020; Flaxman et al. 2016). We define this dimension of news sharing as ideological congruence:

Def.3: Ideological congruence is the utility a user derives from sharing content that is consistent with her/his prior beliefs (cell feature).

These three different sharing incentives remain unobserved for every user and news organization. However, we may obtain reasonable estimates by modeling aggregate responses from a large number of users. Figure 1 describes the underlying model, with each of the different sets of parameters described by row features (issue salience), column features (reputation), and cell features (ideological congruence). Table 1 presents a possible distribution of social media embeds to exemplify the different parameters of the model in observational data. The next section provides a more formal treatment.

How to Interpret Aggregate News Sharing Data

Consider a vector of social media users (rows) that embed hyperlinks to news published by media organizations (columns).⁸ Table 1 of Figure 1 provides an example, with each user $u_{i} \in I$ sharing news published by media organizations, $m_{j} \in J$ . For presentation purposes, to simplify the visualization of Table 1, let us assume that these media organizations are listed from left to right by conservatism, and the dominant dimensions are retrieved from the data in our case studies,⁹ so that $m_{1}$ is less conservative than $m_{2}$ , and $m_{1} < m_{2} < \dots < m_{j}$ . Sorting media organizations from left to right on Table 1 allow the reader to easily interpret the column and cell features we are interested in.

Figure 1.

Theoretical model for news sharing. Notes: Table with counts of news embeds by user (rows) and media organizations (columns) is used to extract parameter issue salience, reputation, and ideological congruence.

In Table 1 of Figure 1, user $u_{2}$ shares a high number of news embeds (twenty-five) while user $u_{1}$ shares very few (seven). We may consider this as an indication that user $u_{2}$ gives more weight to the issue reported in the news compared with $u_{1}$ . Readers may also note that $m_{2}$ is shared by all but one of the users while $m_{3}$ is shared by only one user. That is, the news content published by $m_{2}$ is shared more broadly than that by $m_{3}$ , which we consider as an indication that $m_{2}$ is more reputable (higher quality, more investment, easier access, etc.). Finally, we may see that $u_{3}$ is sharing more progressive news (news published by outlets on the left) while $u_{4}$ is sharing more conservative news (news published by outlets on the right).

As political communication scholars, we value this information because knowing that some users really want to share content on an issue (rows) is conceptually different than knowing that content from some outlets is widely shared (columns). This is also conceptually different than knowing that some users share information from only some outlets (cells). Demand for news on an issue by users and prevalence of news by a source are conceptually different (Karnowski et al. 2020; Kümpel et al. 2015; Scheufele and Nisbet 2013; Trilling et al. 2017).

The Statistical Model: Nuts and Bolts

We now provide a detailed description of the statistical model. Consider a utility function where each social media user $u_{i}$ minimizes ideological dissonance on issue k, given her preferred ideological position, $x_{i}^{k}$ , and an editorialized set of news, $L_{j}^{k}$ , that is created or published by a news organization $m_{j}$ . We define editorialized news as content that is posted by a news organization and that has an ideological charge readily observed by the user. News organization $m_{j}$ may be a newspaper, a candidate, a political group, a social media peer, a friend, or any entity that publishes information in social media that is accessed by user $u_{i}$ . At this time, we only assume that media entities have a separate online page that can be accessed (and shared) through hyperlinks that may be inserted in a social media post.

The utility of user $u_{i}$ also increases with the perceived reputation $R_{j}^{k}$ of news organization $m_{j}$ . That is, users want to minimize cognitive dissonance when sharing news, but cognitive congruence is valuable insofar as information is trustworthy. For example, if the National Enquirer and Fox News publish the exact same article, we expect that users will value the National Enquire lower than Fox News. Therefore, we expect a higher likelihood of sharing the Fox News article.

By assumption, ideological dissonance is negative while reputations are positive. That is, users are less likely to share posts that disagree with their ideological beliefs and see a declining utility from news that is further removed or openly challenges those beliefs. While it is possible that users “ironically” share news that is cognitively dissonant, there is no empirical evidence that shows systematic sharing of dissonant content in social media. However, if users shared news that is cognitively dissonant (across the aisle), the model would describe such behavior.

Users also receive a positive utility for information they agree with if it is published by reputable news organizations. The reputation and the ideological leaning of news organizations may or may not be correlated with each other. Readers of Fox News, for example, may consider that its publications are of high reputation precisely because they minimize ideological dissonance.¹⁰ Readers of the New York Times, on the other hand, may perceive that Fox News is both biased and of low quality because news published by this organization fail to align with the individual’s preferences. Other readers, however, may perceive that ideology and reputation are separate dimensions and are orthogonal to each other. For example, a conservative reader may perceive that the NYT and Fox News are of high reputation and that the New York Post, while conservative and congruent with her/his beliefs, is of low reputation. The extent to which ideology and reputation are interrelated is something that we can test for empirically.

Both ideology and reputation are issue-dependent. That is, users may perceive the New York Times as leftist when reading world news, but see this same organization as centrist when reading real state News. Readers may also perceive that reputation varies by issue, considering the book reviews of the New Yorker as being of higher reputation than those of the New York Times, even if they do not differ in ideological terms. Therefore, ideological proximity and reputation may vary by issue as well as by organizations.

Issues may also be more or less important to each user. We identify this behavioral parameter as the variable $A_{i}^{k}$ in user’s utility function. Therefore, a reader $u_{i}$ will perceive utility from sharing news on issue k by organization j as described in Equation (1):

U_{(i j)}^{k} = - α_{i}^{k} (x_{i}^{k} - L_{j}^{k})^{2} + A_{i}^{k} + R_{j}^{k} + γ_{i j}^{k} .

(1)

In Equation (1), the quadratic term

- α_{i}^{k} (x_{i}^{k} - L_{j}^{k})^{2}

describes the disutility of a publication by media

m_{j}

on issue k, with ideological leaning L that is further removed from the reader’s preferred ideological position,

x_{i}^{k}

. For every unit of increase in ideological dissonance, the utility of user i declines by

α

. The parameter

α

also has a natural interpretation as the weight that a user attaches to the ideological leaning of a media organization on issue k. When browsing for news about Donald J. Trump, for example, ideology may weigh more heavily on the user’s decision to activate content than when browsing news about Justin Bieber,

α_{i}^{B i e b e r} < α_{i}^{T r u m p}

. As in Zheng et al. (2006),

α

, x, and L may be interpreted as latent and unobserved parameters.

Equation (1) also shows that news published by a more reputable actor, $R_{j}^{k}$ , increase the utility of user $u_{i}$ . The importance of reputation varies by issue k. For example, reputation may matter more when reading about Donald Trump than when reading about Justin Bieber, $R_{i j}^{B i e b e r} < R_{i j}^{T r u m p}$ . As we will show, reputation will also vary by the location of users in different regions of a network. Finally, users may also give different salience to an issue, $A_{i}^{k}$ , sharing a higher or lower average number of posts with their social media peers. As in Delli Carpini (2004), issue salience is heterogeneous and varies with user and issue.

Equation (1) also includes a stochastic term that captures overdispersion, $γ_{i j}^{k}$ , by user and media outlet. Readers may recognize Equation (1) as a multilevel specification with a random slope, $α$ , and two random intercepts, A and R. The random slope captures the weight that users attach to ideological congruence, while the random intercepts describe the importance of issue salience and reputation.

Sharing news can take many forms, such as reading (clicking), liking, or sharing content (retweeting). For simplicity, we consider the number of times users share content as the dependent variable. Multilevel estimation of the proposed model proceeds as in Zheng et al. (2006). Readers may readily observe that Equation (1) has a very large number of parameters and imposes significant computational demands. In the Supplementary Information File, we describe a strategy to reduce computational demands, binning issue salience—the row parameters—in quantiles.¹¹

Testing the Model: Are Ideologues Over Represented in Observational Data?

Table 2 summarizes our parameter definitions and measurement strategy. Readers may immediately notice that interesting questions emerge when considering the relationship between different parameter sets. For example, if the importance or weight that a user gives to ideology increases, will she/he also perceives the issue as more salient? Readers will immediately notice that if salience and ideological congruence are positively related to each other, ideologues would be more active and more readily observed in social media data.¹² Therefore, the content they share will be overrepresented, resulting in a network that appears more politicized (and likely polarized) than its average user. If ideologues preferences are over represented in news sharing data, therefore, the network will display heightened perceptions of polarization among the public. As it was described by Scheufele and Nisbet (2013): “Our social networks, that is, the people we are surrounded by in most of our daily activities, tend to be extremely like-minded and homogenous in their demographic and ideological makeup.” (Scheufele and Nisbet 2013: 46). Such perception, however, may be the result of how different users differ in their behavior rather than being a feature of their social compositions and network prevalence.

Table 1.

News Sharing Model - Parameters, Theory and Measurement.

Parameters	Theoretical definition	Measurement strategy
Issue salience	The utility of sharing content more important or salient	Random intercepts for each user (row) in the embeds matrix
Reputation	The utility of sharing content that is also shared by peers and, consequently, recognized as high quality	Random intercepts for media (column) in the embeds matrix
Ideological congruence	The utility of sharing content that is consistent with prior beliefs	Distance in the first dimension of the network between user $u_{i}$ and media $m_{j}$

Note: All the three parameters are estimated using a multilevel specification with random sloped by equally sized quantiles in the first dimension of the network. The use of random slopes reduces computational burden for the model and also provides measures to verify variation in the weights of each parameter in different parts of the network.

The over-representation of ideologues in social networks is a variation on the well-known Friendship Paradox (Feld 1991), with more connected and more active nodes resulting in subjective perceptions of polarization that differ from mean polarization. If more extreme users differ from less extreme users on how frequently they share news, the result will be observational data that is more polarized than its users.

A broad literature on affective polarization has shown that intense ideologues are also unconditionally more motivated to participate in politics and in social media (Barberá 2020; Guess 2021; Huddy et al. 2015; Mason 2018). Similar findings—this time related to sharing fake news—in a recent study by Osmundsen et al. (2020), show negative partisan effects yielding large increases in the likelihood of sharing news. If negative and positive evaluations of political events result in users seeking and delivering information that is consistent with their preferences, motivated users will be both more enthusiastic as well as more attuned to particular types of evidence, which will positively correlate ideological beliefs and issue salience (Weaver 1991). Therefore, the hypothesis to be tested states the following:

$H_{1}$ : More ideological users, who share news that aligns with their preferences, will also be unconditionally more issue-motivated than non-ideological users .

The corollary to this hypothesis is that the relationship between ideology and issue salience will be multiplicative rather than additive. Therefore:

$H_{2}$ : A positive correlation between ideological congruence and issue salience will bias the frequency with which researchers will observe high levels of polarization .

Three Social Media Events: #Bolsonaro, #Maldonado, and #TravelBan

We provide evidence of the usefulness of the proposed model considering three different social media events in Brazil, Argentina, and the United States.¹³ All three events took place in deeply divided political contexts and garnered significant political attention. In all three cases, we have a larger showing by users with more progressive leanings, who are protesting against right-wing shifts in the status quo.

The data collection for the three cases followed similar procedures. We collected data using both APIs available on Twitter: the forward stream and the backward search. The streaming API collects live data, letting users capture a portion of tweets in real time. The search API allows users to access a repository of tweets published seven days prior to the query. Our search used both APIs with three search terms: Bolsonaro, Maldonado, and TravelBan. By collecting data from both APIs we increase our sample (both in terms of size and source), therefore avoiding risks of messages’ removals and algorithmic bias made arbitrarily by Twitter for each API (Timoneda 2018).

After collecting the data, we limit our sample to a network of all retweets from the original data, and retrieve information about the edges (retweet), the author of the tweet (authority) and the users who retweeted the original message (hubs). We then retained the largest connected cluster of the network, eliminating users with less than two retweets (out-degree> = 2). With the thinned network, we draw users’ [x,y] coordinates implementing the Fruchterman-Reingold algorithm in igraph-R (Csard et al. 2006). We then ran the walk.trap algorithm to identify the main clusters in the network. We label these clusters as communities in our network, and we validate them with qualitative analysis of the main authorities in each cluster (see appendix C, Supplementary Information file). We describe below these communities, and the data collection for each case.

First, we consider the Bolsonaro network in Brazil, using 2,943,993 tweets published by 162,107 high activity accounts the week prior to the election of President Jair Bolsonaro, from September 26 through October 2, 2018. Bolsonaro is widely considered as a fringe right-wing candidate, who has stacked his administration with military officers, celebrated the use of torture by the 1964–1985 military regime, and introduced extreme legislation to reverse social policies in areas such as LGBT rights and welfare insurance. Jair Bolsonaro has been an extremely divisive political figure and, more important for this research, used a vast network of intelligence and “fake news mills” to support his presidential candidacy (Aruguete et al. 2021). Consequently, there are significant differences in the reputation of traditional media outlets and new extremely conservative ones.

The upper-left plot of Figure 2 describes the basic layout of the Bolsonaro network, with pro-Bolsonaro users in blue and anti-Bolsonaro users in red. Of the more than 2.9 million retweets analyzed in the data, 432,591 (14.7 percent) included hyperlinks. The number declines to 387,841 (13.2 percent) if we do not consider hyperlinks to other tweets. The most frequent news outlet embedded in the data is the pro-Bolsonaro O’ Antagonista, which represents 63,862 hyperlinks, 14.7 percent, and is intensively retweeted by core supporters of the current president.

Figure 2.

Visualization of all retweets in the bolsonaro network (upper-left), maldonado (upper-right), and travelBan (lower). Notes: Layout (Fruchterman-Reingold) and community detection (Random Walk) for the Bolsonaro, Maldonado, and TravelBan networks. Igraph in R 3.6.

In the case of Maldonado in Argentina, the upper-right plot in Figure 2, we analyze 5,325,240 tweets posted by 196,066 high activity accounts in the seventy-eight days that followed the disappearance of activist Santiago Maldonado, from August 1 to October 18, 2017. The disappearance of Maldonado was a deeply polarizing event. Different media outlets aligned for and against the government, which the opposition portrait as responsible. Of the more than five million retweets analyzed in the data, 816,694 (15.3 percent) included hyperlinks. The number declines to 513,659 (9.6 percent) if we eliminate hyperlinks to other tweets.

In the case of the TravelBan in the United States, lower-left plot of Figure 3, we analyze 2,031,518 retweets from 241,271 high activity accounts on January 30 and 31, 2017, following the decision of the Trump administration to restrict travel from seven majority Muslim countries. The basic layout of the TravelBan network shows pro-TravelBan users in blue and anti-TravelBan users in red. Of more than two million retweets analyzed in the data, 641,719 (31 percent) included hyperlinks. The number declines to 485,560 (23.9 percent) if we do not consider hyperlinks directed to other tweets.¹⁴

Figure 3.

News sharing in bolsonaro (top), maldonado (middle), and travelBan (bottom).

For each of the three networks we retrieve the matrix of users (rows) and media organizations (columns), keeping only the twenty-four most frequently embedded news outlets. We retrieve the first dimension value for each user (horizontal axes in figures each plot of Figure 2), as a proxy for $x_{i}^{k}$ in Equation 3, and the quantile indices $q (i)$ . To estimate the media locations, $L_{j}^{k}$ , we borrow from the model developed by Bond and Messing (2015) and estimate the average $x_{i}^{k}$ position for the user $u_{i}$ , weighted by the number of media $m_{j}$ embeds.¹⁵

A Visual Inspection of Media Embedding

Figure 3 presents twenty-four plots that describe the areas of the network activated by the top eight media outlets in Bolsonaro, Maldonado, and the TravelBan. The other forty-eight media outlets can be found in the Supplementary Information file. In all three of our cases, the activated nodes describe the region of the network where news sharing is more active.

The top plot of Figure 3 shows activation in Brazil, with O’ Antagonista and Conexão Politica activated more readily by users on the right of the political spectrum. The former was recently founded by three prominent conservative journalists that abandoned the weekly news magazine Veja. Meanwhile, Folha, Veja, and Globo are more readily embedded by users on the center and center-left of the political spectrum. Readers may also note a much larger share of links to Twitter on the left and more frequent links to YouTube on the right of the political spectrum.

Such is the result of the decision by Twitter to suspend accounts from the conservative group Movimiento Brasil Libre, who engaged from within YouTube on a very active campaign of misinformation. The decision by Twitter was mirrored by Facebook, who suspended over 100,000 WhatsApp accounts on what is without a doubt the largest astroturfing campaign in any election in the region.

The middle plot in Figure 3 shows distinct activation by Argentine outlets on the left (Página/12) or right (TN) of the political spectrum. However, other outlets such as La Nación are embedded by most of the conservative users but also by a significant number of moderates on the center of the political spectrum. Finally, in the case of the TravelBan, lower plot in Figure 4, we see outlets with a higher than average readership on the left of the political spectrum (New York Times) as well as those with a wider right leaning readership (Fox News).

Figure 4.

Ideology by quantile (left) and issue salience by quantile (right). Note: Parameters of ideology and salience by quantile estimated from Equation (1).

More important for our purpose, all twenty-four plots provide evidence of significant variation within and across networks, with some media outlets being more widely shared by all users, some media outlets more intensely shared by users in a particular region of the network, as well as other users more actively sharing links on the Bolsonaro, Maldonado, and the TravelBan networks.

Ideological Congruence, Issue Salience, and Reputation

Figures 2 and 3 describe the data captured in the three RxC matrices of news embeds of Bolsonaro, Maldonado, and the TravelBan. Using Equations 1 and 2, as well as the binning strategy proposed in Section 3.1, we proceed to estimate all three sets of parameters (ideological congruence, issue salience, and reputation). Figure 4 provides a visual comparison of the ideology and issue salience parameters by quantile for each of our three cases.

Because ideological distance is cognitively costly, larger negative values indicate that ideological congruence matters more. In the particular case of Maldonado, for example, users in the 8th, 9th, and 10th quantile, which corresponds to the core of the pro-Government sub-network, display large negative estimates, reflecting a high demand for congruent news. As it is also the case with survey data, ideological congruence tends to be more modest for users in the center of the network and it increases centrifugally as we move to the extremes.

If we compare our results with similar estimates in survey data, we will see that ideological congruence has a significantly larger weight in these networks. In effect, most estimates of ideological congruence in survey data fall in the range of [-.05, -.12], four times smaller than the estimates in observational social media data (Calvo and Hellwig 2011).

The right plot in Figure 4 provides estimates of issue salience by quantiles, with larger values indicating a higher propensity to embed links on the collection terms we use for each case. Consistent with the visual inspection of the data in the previous section, we see that users on the left and right of the political spectrum are more likely to pay attention to #Bolsonaro, #Maldonado, and the #TravelBan. Particularly interesting is the very high issue salience of users to the right of the political spectrum in #Maldonado, with activity that is orders of magnitude above the rest of the network. While users on the right of the political spectrum were fewer in number in the #Maldonado network, they still shared news on the issue at much higher rates.

Figure 5 provides estimates of the model’s reputation parameter. As described in section three, these parameters capture that propensity of users to embed links to media organizations once we control for issue salience and ideology. As expected, readers can observe that niche organizations on the left and right of the political spectrum are at the bottom of the list, given that most of the news sharing incentives is explained by the users’ ideological affinity with the media. By contrast, most traditional news organizations are more broadly shared once we control for the other factors.

Figure 5.

Reputation parameters by news organization in, Brazil, Argentina, and United States. Notes: Estimates describe the $R_{j}^{k}$ parameters in Equation (1). Higher parameter values describe media that is more broadly shared by users, holding issue salience and ideology constant.

In the next section, we use the estimated parameters to evaluate the hypothesis that test for a positive correlation between ideology and issue salience.

Are More Intense Users Over Represented in Observational Data?

Once we estimate the importance of ideology, issue salience, and reputation in observational data, we may use our parameter estimates to test the hypotheses in Section 4, which asked whether more ideological users were over-represented in observational data. As it was presented, if issue salience and ideology are positively correlated, then political news shared in social media will appear to be more polarized than they actually are. That is, the preferences of intense ideologues would be over-represented in observational data.

Visual inspection of Figure 4 in the previous section hinted of a relationship between ideological preference and issue salience. The salience parameters in Bolsonaro, Maldonado, and the TravelBan closely align with the ideology parameters. It is worth reminding that while issue salience increases sharing behavior, ideological distance measures dissonance and, consequently, reduces sharing behavior. To simplify the interpretation of the results, we display $α$ instead $– α$ , showing a positive correlation between the weight users attach to ideology and to issue salience.

Results displayed in Figure 6 strongly support Hypothesis 1, with a close fit between issue salience and ideology in all three countries. Country correlations of 0.82 for the United States, 0.86 for Brazil, and 0.92 for Argentina validate the hypothesis for all three cases, showing clear support in observational social media data.

Figure 6.

Relationship between the ideology and issue salience parameters, all three networks. Notes: The figures use augmented data from the parameter estimates in Equation (1). To provide a more intuitive understanding, we inverted the value of the ideology parameter; therefore, positive values for issue salience and ideological congruence means higher weights on both dimensions. The Pearson correlation between ideology and issue salience using the augmented data is 0.84, 0.92 and 0.86 for each network, respectively.

The joint effect of ideology and issue salience in Figure 6 provide a clear mechanism to explain the appearance of high ideological sorting in social media data (Aruguete and Calvo 2018; Barberá 2020; Flaxman et al. 2016; Osmundsen et al. 2020), which reflects the outsized weight of ideologues’ news sharing behavior on politically salient issues.

Figure 6 provides support for $H_{1}$ , showing that ideology and issue salience are closely connected. We may also evaluate the expected content shared by ideologues compared to non-ideologues by substituting proper values for the parameters in Equation (1). Consider for example a news organization that publishes posts exactly at the location of two users with $α_{1}^{k} > α_{2}^{k}$ , so that user 1 is an ideologue and user 2 is not an ideologue. Given that $- α_{1, 2}^{k} (0)^{2} + A_{1, 2}^{k}$ , we find that the content shared by user 1 is larger than the content shared by user 2 by $\exp (A_{1}^{k} - A_{2}^{k})$ . Given that both $A_{1}^{k}$ and $A_{2}^{k}$ are linearly related to $α_{1}^{k}$ and $α_{2}^{k}$ , the content shared by ideologues is overrepresented in the observational data as expected by $H_{1} .$

Concluding Remarks

This paper describes a statistical strategy to study news sharing behavior using observational social media data. We propose a simple model that takes a matrix of embeds as inputs and estimates the importance of ideological congruence, issue salience, and reputation in social media data. Our model provides a path to test existing theories of news sharing using observational social media embeds. We exemplify this method with Twitter data from three major social media events in Argentina, Brazil, and the United States, although the model could use as input any $R x C$ matrix of embeds.

The model allows researchers to estimate meaningful parameters of interest from observational data. While there have been extraordinary computational advances in the study of large social networks, designs that answer practical communication theory problems receive considerably less issue salience. Our analysis combines computational tools with multilevel modelling to fill this gap, focusing on the behavioral determinants of news sharing.

As it pertains to the cases of Bolsonaro, Maldonado, and the TravelBan, results show that users on the left and right of the political spectrum are both more attentive to the issues and more likely to share ideologically congruent news. The proposed model allows us to test for a positive correlation between issue salience and ideology, which explains why ideologues are overrepresented in social media data.

The parameters retrieved from the matrix of news embeds may also be used to test other important communication problems, such as the editorial incentives to cater to extreme users (gatekeeping). Alternative specifications of the proposed model may also be used to explore the linkages between ideology and reputation. Finally, future extensions of this model may expand the matrix formulation to issue dimensions, with users (rows), media (columns), cognitive congruence (cells), and issues (layers). These are computationally tractable extensions that will allow researchers to work with flexible and scalable models to better understand news sharing behavior.

Supplemental Material

sj-pdf-1-ijpp-10.1177_19401612211057068 - Supplemental material for News by Popular Demand: Ideological Congruence, Issue Salience, and Media Reputation in News Sharing¹

Supplemental material, sj-pdf-1-ijpp-10.1177_19401612211057068 for News by Popular Demand: Ideological Congruence, Issue Salience, and Media Reputation in News Sharing¹ by Natalia Aruguete, Ernesto Calvo and Tiago Ventura in The International Journal of Press/Politics

Footnotes

Correction (May 2023):

This article has been updated with style corrections since its original publication.

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) received no financial support for the research, authorship and/or publication of this article.

ORCID iDs

Ernesto Calvo

Tiago Ventura

Supplemental Material

Supplemental material for this article is available online.

Notes

Author Biographies

Natalia Aruguete (PhD UNQ, 2000) is Professor of the Universidad Nacional de Quilmes and the Universidad Austral, and a member of CONICET (Argentina’s Federal Agency for the development of Science and Technology). She has published three books and fifty articles studying the relationship between traditional media and social media, with particular attention to the effect of political agendas, media agendas, and the public.

Ernesto Calvo (PhD, Northwestern University 2001), is Professor of Government and Politics at the University of Maryland and Director of the interdisciplinary Lab for Computational Social Science (iLCSS). His research focuses on the study of comparative political institutions, representation, and social networks. He is the author of Fake News, Trolls y otros Encantos (Siglo XXI, 2020) with Natalia Aruguete; Non-Policy Politics (Cambridge U.P, 2019) with M. Victoria Murillo; Legislator Success in Fragmented Congresses in Argentina (Cambridge U.P: 2014), and over seventy publications in Latin America, Europe, and the United States.

Tiago Ventura is a Ph.D. Candidate in the Department of Government and Politics at the University of Maryland, College Park, and a researcher at the Interdisciplinary Laboratory for Computational Social Science (iLCSS). He is a scholar in computational social science, and comparative political behavior. His work focuses on political communication, survey digital experimentation, content activation, misinformation and polarization on social media. His research has appeared at top academic journals in political science and communication, such as the Journal of Communication, Electoral Studies, Digital Journalism, Journal of Elections, Public Opinion and Parties, and Latin American Politics and Society.

References

Aruguete

Calvo

. 2018. “Time to# Protest: Selective Exposure, Cascading Activation, and Framing in Social Media.” Journal of Communication 68 (3): 480–502. https://doi.org/10.1093/joc/jqy007.

Aruguete

Calvo

Ventura

. 2021. “News Sharing, Gatekeeping, and Polarization: A Study of the# Bolsonaro Election.” Digital Journalism 9 (1): 1–23. https://doi.org/10.1080/21670811.2020.1852094.

Bail

2021. Breaking the Social Media Prism: How to Make Our Platforms Less Polarizing. Princeton, USA: Princeton University Press.

Barberá

2015. “Birds of the Same Feather Tweet Together: Bayesian Ideal Point Estimation Using Twitter Data.” Political Analysis 23 (1): 76–91. https://doi.org/10.1093/pan/mpu011.

Barberá

Pablo

. 2020. “Social media, echo chambers, and political polarization.” Social Media and Democracy: The State of the Field, Prospects for Reform 34.

Blanchett Neheli

2018. “News by Numbers: The Evolution of Analytics in Journalism.” Digital Journalism 6 (8): 1041–51. https://doi.org/10.1080/21670811.2018.1504626.

Boehmer

Jan

Tandoc

Edson C

. 2015. “Why we retweet: Factors influencing intentions to share sport news on Twitter.” International Journal of Sport Communication 8 (2): 212–232.

Bond

Messing

. 2015. “Quantifying Social Media’s Political Space: Estimating Ideology from Publicly Revealed Preferences on Facebook.” American Political Science Review 109 (1): 62–78. https://doi.org/10.1017/S0003055414000525.

Boyd

D. M.

Golder

Lotan

. 2010. “Tweet, Tweet, Retweet: Conversational Aspects of Retweeting on Twitter.” In 2010 43rd Hawaii International Conference on System Sciences, 1–10. Honolulu, HI: IEEE. https://doi.org/10.1109/HICSS.2010.412.

10.

Calvo

Hellwig

. 2011. “Centripetal and Centrifugal Incentives Under Different Electoral Systems.” American Journal of Political Science 55 (1): 27–41. https://doi.org/10.1111/j.1540-5907.2010.00482.x.

11.

Choi

Lee

J. K.

. 2015. “Investigating the Effects of News Sharing and Political Interest on Social Media Network Heterogeneity.” Computers in Human Behavior 44:258–66. https://doi.org/10.1016/j.chb.2014.11.029.

12.

Conover

M. D.

Ratkiewicz

Francisco

Gonçalves

Menczer

Flammini

. 2011. “Political polarization on twitter.” In Fifth international AAAI conference on weblogs and social media.

13.

Delli Carpini

M. X.

2004. “Mediating Democratic Engagement: The Impact of Communications on Citizens’ Involvement in Political and Civic Life.” In Handbook of Political Communication Research, ed. L. L. Kaid, 395–434. Mahwah, NJ: Lawrence Erlbaum.

14.

Feld

S. L.

1991. “Why Your Friends Have More Friends Than you do.” American Journal of Sociology 96 (6): 1464–77. https://doi.org/10.1086/229693.

15.

Flaxman

Goel

Rao

J. M.

. 2016. “Filter Bubbles, echo Chambers, and Online News Consumption.” Public Opinion Quarterly 80 (S1): 298–320. https://doi.org/10.1093/poq/nfw006.

16.

García-Perdomo

Salaverría

Kilgo

D. K.

Harlow

. 2018. “To Share or not to Share: The Influence of News Values and Topics on Popular Social Media Content in the United States, Brazil, and Argentina.” Journalism Studies 19 (8): 1180–201. https://doi.org/10.1080/1461670X.2016.1265896.

17.

Guess

A. M.

2021. “(Almost) Everything in Moderation: New Evidence on Americans’ Online Media Diets.” American Journal of Political Science 65: 1007–1022.

18.

Hanusch

Nölleke

. 2019. “Journalistic Homophily on Social Media: Exploring Journalists’ Interactions with Each Other on Twitter.” Digital Journalism 7 (1): 22–44. https://doi.org/10.1080/21670811.2018.1436977.

19.

Huddy

Mason

Aarøe

. 2015. “Expressive Partisanship: Campaign Involvement, Political Emotion, and Partisan Identity.” American Political Science Review 109 (1): 1–17. https://doi.org/10.1017/S0003055414000604.

20.

Kahneman

2011. Thinking, Fast and Slow. New York: Macmillan.

21.

Karnowski

Leiner

D. J.

Sophie Kümpel

Leonhard

. 2020. “Worth to Share? How Content Characteristics and Article Competitiveness Influence News Sharing on Social Network Sites.” Journalism & Mass Communication Quarterly 98(1): 59–82. https://doi.org/1077699020940340.

22.

Karnowski

Leonhard

Kümpel

A. S.

. 2018. “Why Users Share the News: A Theory of Reasoned Action-Based Study on the Antecedents of News-Sharing Behavior.” Communication Research Reports 35 (2): 91–100. https://doi.org/10.1080/08824096.2017.1379984.

23.

Kümpel

A. S.

Karnowski

Keyling

. 2015. “News Sharing in Social Media: A Review of Current Research on News Sharing Users, Content, and Networks.” Social Media+ Society 1 (2): .1–15. https://doi.org/10.1177/2056305115610141.

24.

Lee

E.-J.

Tandoc

E. C.

Jr.

2017. “When News Meets the Audience: How Audience Feedback Online Affects News Production and Consumption.” Human Communication Research 43 (4): 436–49. https://doi.org/10.1111/hcre.12123.

25.

Macskassy

S. A.

Michelson

2011, July. “Why do people retweet? Anti-Homophily Wins the Day!” In Fifth International AAAI Conference on Weblogs and Social Media.

26.

Mason

2018. Uncivil Agreement: How Politics Became our Identity. Chicago: University of Chicago Press.

27.

Oosterhoff

Shook

N. J.

Ford

. 2018. “Is That Disgust I see? Political Ideology and Biased Visual Attention.” Behavioral Brain Research 336:227–35. Is That Disgust I see? Political Ideology and Biased Visual Attention.

28.

Osmundsen

Bor

Vahlstrup

P. B.

Bechmann

Petersen

M. B.

. 2020. “Partisan Polarization is the Primary Psychological Motivation Behind “Fake News” Sharing on Twitter.” American Political Science Review 115 (3): 999–1015. https://doi.org/10.7910/DVN/AH6AZK.

29.

Rudat

Anja

Buder

Jürgen

. 2015. “Making retweeting social: The influence of content and context information on sharing news in Twitter.” Computers in Human Behavior 46: 75–84.

30.

Russell

F. M.

2019. “Twitter and News Gatekeeping: Interactivity, Reciprocity, and Promotion in News Organizations’ Tweets.” Digital Journalism 7 (1): 80–99. https://doi.org/10.1080/21670811.2017.1399805.

31.

Scheufele

D. A.

Nisbet

M. C.

. 2013. “Commentary: Online News and the Demise of Political Disagreement.” Annals of the International Communication Association 36 (1): 45–53. https://doi.org/10.1080/23808985.2013.11679125.

32.

Sikder

Smith

R. E.

Vivo

Livan

. 2020. “A Minimalistic Model of Bias, Polarization and Misinformation in Social Networks.” Scientific Reports 10 (1): 1–11. https://doi.org/10.1038/s41598-020-62085-w.

33.

Strömbäck

Djerf-Pierre

Shehata

. 2013. “The Dynamics of Political Interest and News Media Consumption: A Longitudinal Perspective.” International Journal of Public Opinion Research 25 (4): 414–35. https://doi.org/10.1093/ijpor/eds018.

34.

Suh

Hong

Pirolli

Chi

E. H.

. 2010, August. “Want to be Retweeted? Large Scale Analytics on Factors Impacting Retweet in Twitter Network.” In 2010 IEEE Second International Conference on Social Computing, 177–84. IEEE.

35.

Timoneda

J. C.

2018. “Where in the World Is My Tweet: Detecting Irregular Removal Patterns on Twitter.” PloS One 13 (9): e0203104. https://doi.org/10.1371/journal.pone.0203104.

36.

Trilling

Tolochko

Burscher

. 2017. “From newsworthiness to shareworthiness: How to predict news sharing based on article characteristics.” Journalism and Mass Communication Quarterly 94 (1): 38–60.

37.

Von Nordheim

Boczek

Koppers

. 2018. “Sourcing the Sources: An Analysis of the use of Twitter and Facebook as A Journalistic Source Over 10 Years in The New York Times, The Guardian, and Süddeutsche Zeitung.” Digital Journalism 6 (7): 807–28. https://doi.org/10.1080/21670811.2018.1490658.

38.

Wang

Can

Kazemzadeh

Bar

Narayanan

. 2012, July. “A System for Real-Time Twitter Sentiment Analysis of 2012 us Presidential Election Cycle.” In Proceedings of the ACL 2012 System Demonstrations, 115–20.

39.

Waruwu

B. K.

Tandoc

E. C.

Jr. Duffy

Kim

Ling

. 2020. “Telling Lies Together? Sharing News as A Form of Social Authentication.” New Media & Society 23(9): 2516–2533. https://doi.org/10.1177/1461444820931017.

40.

Weaver

1991. “Issue Salience and Public Opinion: Are There Consequences of Agenda-Setting?” International Journal of Public Opinion Research 3 (1): 53–68. https://doi.org/10.1093/ijpor/3.1.53.

41.

Weeks

B. E.

Holbert

R. L.

. 2013. “Predicting Dissemination of News Content in Social Media: A Focus on Reception, Friending, and Partisanship.” Journalism & Mass Communication Quarterly 90 (2): 212–32. https://doi.org/10.1177/1077699013482906.

42.

Zheng

Salganik

M. J.

Gelman

. 2006. “How Many People do you Know in Prison? Using Overdispersion in Count Data to Estimate Social Structure in Networks.” Journal of the American Statistical Association 101 (474): 409–23.

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.

0.00 MB

2.22 MB

News by Popular Demand: Ideological Congruence,Issue Salience,and Media Reputation in News Sharing 1

Abstract

Keywords

Introduction

Theory: A Statistical Description of News Sharing in Social Media Data

Why Is it Important to Develop a Statistical Model of News Sharing?

Modeling News Sharing Behavior: An Intuitive Description

How to Interpret Aggregate News Sharing Data

The Statistical Model: Nuts and Bolts

Testing the Model: Are Ideologues Over Represented in Observational Data?

Three Social Media Events: #Bolsonaro, #Maldonado, and #TravelBan

A Visual Inspection of Media Embedding

Ideological Congruence, Issue Salience, and Reputation

Are More Intense Users Over Represented in Observational Data?

Concluding Remarks

Supplemental Material

sj-pdf-1-ijpp-10.1177_19401612211057068 - Supplemental material for News by Popular Demand: Ideological Congruence, Issue Salience, and Media Reputation in News Sharing 1

Footnotes

Correction (May 2023):

Declaration of Conflicting Interests

Funding

ORCID iDs

Supplemental Material

Notes

Author Biographies

References

Supplementary Material

sj-pdf-1-ijpp-10.1177_19401612211057068 - Supplemental material for News by Popular Demand: Ideological Congruence, Issue Salience, and Media Reputation in News Sharing¹