Partisan Selective Sharing: The Biased Diffusion of Fact-Checking Messages on Social Media: Sharing Fact-Checking Messages on Social Media

Abstract

Using large Twitter datasets collected during the 2012 U.S. presidential election, we examined how partisanship shapes patterns of sharing and commenting on candidate fact-check rulings. Our results indicate that partisans selectively share fact-checking messages that cheerlead their own candidate and denigrate the opposing party's candidate, resulting in an ideologically narrow flow of fact checks to their followers. We also find evidence of hostile media perception in users' public accusations of bias on the part of fact-checking organizations. Additionally, Republicans showed stronger outgroup negativity and hostility toward fact checkers than Democrats. These findings help us understand “selective sharing” as a complementary process to selective exposure, as well as identifying asymmetries between partisans in their sharing practices.

Full text

Journal of Communication ISSN 0021-9916
ORIGINAL ARTICLE
Partisan Selective Sharing: The Biased
Diffusion of Fact-Checking Messages on
Social Media
Jieun Shin1& Kjerstin Thorson2
1 Department of Preventive Medicine, Keck School of Medicine, University of Southern California, Los Angeles,
CA 90032, USA
2 College of Communication Arts and Sciences, Michigan State University, East Lansing, MI 48824, USA
Using large Twitter datasets collected during the 2012 U.S. presidential election, we exam-
ined how partisanship shapes patterns of sharing and commenting on candidate fact-check
rulings. Our results indicate that partisans selectively share fact-checking messages that
cheerlead their own candidate and denigrate the opposing party’s candidate, resulting in an
ideologically narrow ow of fact checks to their followers. We also nd evidence of hostile
media perception in users’ public accusations of bias on the part of fact-checking organiza-
tions. Additionally, Republicans showed stronger outgroup negativity and hostility toward
fact checkers than Democrats. ese ndings help us understand “selective sharing” as a
complementary process to selective exposure, as well as identifying asymmetries between
partisans in their sharing practices.
Keywords: Fact Checking, Selective Sharing, Selective Exposure, Hostile Media, Partisan,
Election, Social Identity, Social Media.
doi:10.1111/jcom.12284
Political fact-checking, which focuses on verifying political actors’ claims, has grown
into an explosive phenomenon over the last half-decade. Unlike traditional journal-
ism, which emphasizes detached objectivity and adheres to the “he said, she said”
style of reporting, contemporary fact-checking directly engages in adjudicating fac-
tual disputes by publicly deciding whose claim is correct or incorrect (Graves, 2016).
Due to its unique format and contribution to the political sphere, the popularity of
fact-checking has been on the rise, generating a 300% increase in fact-checking stories
from 2008 to 2012 (Graves & Glaisyer, 2012). Today, nearly every national news out-
let incorporates fact-checking features into their political coverage (Graves, Nyhan, &
Reier, 2016).
Corresponding author: Jieun Shin; e-mail: jieunshi@usc.edu
Journal of Communication (2017) © 2017 International Communication Association 1

Sharing Fact-Checking Messages on Social Media J. Shin & K. Thorson
Despitetheupsurgeofinterest,ourunderstandingofthebroaderimpact
of the fact-checking phenomenon is still limited. Recent studies have evaluated
fact-checking as an eective intervention in improving political knowledge and
reducing belief in misinformation (Fridkin, Kenney, & Wintersieck, 2015; Nyhan &
Reier, 2015; Weeks, 2015; Wood & Porter, 2016). However, most of these studies
are experiments that require participants to read fact-checking messages that they
may not normally choose to consume. e literature on political polarization and
selective exposure casts doubt on the assumption that partisans seek out content
that challenges their views (Iyengar & Hahn, 2009; Stroud, 2008; Sunstein, 2001).
In addition, given the current media environment where exposure to news depends
signicantly on what your friends share on social media (Bakshy, Messing, & Adamic,
2015; Gottfried & Shearer, 2016), the visibility of fact-checking messages may be
aected by the phenomenon of selective sharing : the extent to which individuals
share primarily attitude-consistent content with their social networks. If the selective
sharing tendency is strong, the objective of fact-checking messages—providing the
public with accurate information on both sides of the political spectrum — becomes
less realistic in practice.
Moreover, examining fact-check recipients’ responses in their usual social settings
is critical to understanding the role of norms and perceptions shared among partisan
group members. In the United States, there is a signicant gap between members of
Republican and Democratic parties in their perceptions of the mainstream media and
fact checkers (Glynn & Huge, 2014; Nyhan & Reier, 2015). Although declining trust
in the media is a common concern across groups, Republicans have historically shown
much lower levels of condence in the media than Democrats (Swi, 2016). e sen-
timent toward fact checkers is no exception: Although Democratic politicians have
embraced the fact-checking movement and asked for more fact-checking in debates,
their Republican counterparts— including presidential candidate Donald Trump in
2016 and Mitt Romney in 2012— condemned fact checkers as biased. Such dier-
ent views of media between Democrats and Republicans could potentially produce
asymmetries in partisans’ responses to fact-checking messages.
To investigate this possibility, we turn to social media to observe “naturally
occurring” responses to fact-checking messages. We use a large set of data col-
lected from Twitter during the 2012 U.S. presidential election to look at how social
media users shared and responded to fact-checking messages about presidential
candidates from two rival parties. We draw on social identity theory (SIT; Tajfel &
Turner, 1979; Turner, Hogg, Oakes, Reicher, & Wetherell, 1987) as well as accounts
of partisan media perceptions to theorize the role of users’ partisanship in their
reaction to fact-check messages on social media. We show that partisans selectively
share fact-checking messages that “cheerlead” their own group and demoralize the
opposing group. Further, we examine public comments in response to fact-checking
messages on Twitter and nd similar patterns at work: Partisans are more likely to
publicly accuse fact-checking organizations of bias, even when the fact-check rulings
themselves are “neutral” toward the Twitter user’s own candidate.
2Journal of Communication (2017) © 2017 International Communication Association

J. Shin & K. Thorson Sharing Fact-Checking Messages on Social Media
We argue for the importance of selective sharing as a complementary process
to selective exposure. We also propose that asymmetries between partisans in their
sharing practices aect the extent to which fact-checking messages— and perhaps
other political messages—are made visible to wider audiences. Our ndings con-
tributetoagrowingbodyofliteratureconnectingmediabehaviorsgroundedin
group identication to broader societal eects on polarization, public opinion, and
misperceptions (Feldman, Myers, Hmielowski, & Leiserowitz, 2014; Garrett, Weeks,
& Neo, 2016; Slater, 2007).
Literature review
Partisan selective sharing
e majority of research on fact-checking has focused on the eects of exposure to
fact checking messages in experimental settings. Although some (e.g., orson, 2016)
found that fact checks have only limited eects on changing the recipients’ attitude
toward a candidate, in general, fact-checking messages have been shown to create
positive eects on the public such as increasing accurate understanding of political
issues (Fridkin et al., 2015; Nyhan & Reier, 2015; Wood & Porter, 2016). ese exper-
imental studies found no evidence that individuals’ prior political attitudes moder-
atedtheeectsoffact-checkexposureondemocraticoutcomes.atis,exposureto
fact-checking messages had a positive outcome on political knowledge even for those
who saw attitude-challenging messages.
Largely missing from this body of work is analysis of what shapes the likelihood
that individuals will encounter and disseminate fact-checking messages in their
dailylives.efewexistingstudiesconcernedwithfact-checkingonsocialmedia
showed that corrective messages oen fail to reach the target audience vulnerable to
misinformation and fall short of aecting the overall dynamics of rumor spreading
(Shin, Jian, Driscoll, & Bar, 2016; Friggeri, Adamic, Eckles, & Cheng, 2014; Hannak,
Margolin, Keegan, & Weber, 2014). Social media users— both elites and everyday
citizens—now have increased power to set news agenda and control ow of informa-
tion (Meraz & Papacharissi, 2013; Nahon & Hemsley, 2013; Singer, 2014). erefore,
itisimportanttoexaminehowtheseusersmakejudgmentsaboutwhatmessagesto
share or not to share with their own audience.
Partisan selective exposure— which has received a great deal of empirical study
(Garrett, 2009; Iyengar & Hahn, 2009; Knobloch-Westerwick & Meng, 2009; Stroud,
2008)—serves as a good reference point for theorizing partisan selective sharing. Just
as the selective exposure thesis argues, selective sharing may also be motivated by
partisan goals: People selectively share ideologically congenial information. Previ-
ous research on information sharing suggests this will be the case. Political bloggers
mainly share hyperlinks aligned with their own political spectrum rather than with
the opposing side (Adamic & Glance, 2005; Jacobson, Myung, & Johnson, 2015).
Twitter users are more likely to retweet messages from those sharing similar political
attitudes (Barbera, Jost, Nagler, Tucker, & Bonneau, 2015; Boutet, Kim, & Yoneki,
2012; Colleoni, Rozza, & Arvidsson, 2014; Conover et al., 2011).
Journal of Communication (2017) © 2017 International Communication Association 3

Sharing Fact-Checking Messages on Social Media J. Shin & K. Thorson
Yet, despite the strong connection between exposure and sharing (Weeks & Hol-
bert, 2013), partisan selective sharing seems to be more robust and consistent than
selective exposure. Previous research on selective exposure shows occasional circum-
stances in which people expose themselves to opinion-challenging information due
to the desire to gain useful information (Knobloch-Westerwick & Kleinman, 2012;
Valentino, Banks, Hutchings, & Davis, 2009). However, existing research suggests that
sharing opinion-challenging information is much more rare. One reason for this dif-
ferencecouldbethattheactofsharingismorevisibletosocialothersthanisexposure.
To use Goman (1959), while sharing oen takes place on a frontstage where sharers
are conscious of their audience and their action, many forms of media consumption
occur in a backstage setting where an audience is not present.
Insum,sharingisfundamentallyasocialactivityintendedforormotivatedbythe
imagined audience (Marwick & Boyd, 2011). erefore, we pay attention to media
users’ social identity (i.e., partisanship) to understand various aspects of their shar-
ing behavior. In this paper, we draw on SIT (Tajfel & Turner, 1979; Turner & Oaks,
1986; Turner et al., 1987) to develop hypotheses and research questions. SIT pro-
poses that, when individuals are prompted by cues indicating one’s important social
group(e.g.,race,religion,orpoliticalparty),theycategorizethemselvesandothers
into either ingroup or outgroup based on shared similarities. As a result they exhibit
ingroup favoritism or outgroup discriminating behaviors to enhance their self-esteem
and reduce uncertainty about their world. For instance, political scientists found that
identication with political party can lead to a feeling of “us against them” and pro-
duce social stereotyping that maximally dierentiates between the two parties (Green,
Palmquist, & Schickler, 2002; Iyengar, Sood, & Lelkes, 2012).
is process of intergroup bias activation requires some type of stimuli to trig-
ger group identication, and contexts that dene the ingroup in relation to other
groups (Oaks, 1987; Slater, 2007; Turner et al., 1987). In this sense, media messages
play an important role not only in linking members of the same group by increasing
the salience of a certain identity (e.g., political party) over alternative other identities
(e.g., gender), but also in priming prototypes, a set of perceived attributes that repre-
sent each group. Slater (2007) argues that strong identiers tend to gravitate toward
media outlets or content that communicates positive prototypes of their own group,
and in turn, selective consumption of media can have reinforcing eects on group
identication and group-oriented behavior.
As such, SIT is particularly relevant to examining social media users’ reaction to
politicalfact-checkingmessages,whicharedesignedtoadjudicatepresidentialcan-
didates’ statements. We theorize that rulings on political candidates’ statements such
as “False,” “True,” and “Pants on Fire” serve as cues to invoke partisan identity by
increasing salience of group boundaries and priming certain prototypes. In this sense,
sharing fact-checking messages that elevate the status of an in-group is a way for par-
tisans to support their own group. Additionally, partisan social media users may use
fact-checking messages —or other political content—to display loyalty to their group
and gain trust from group members (Hogg, 2001; Hogg & Reid, 2006). In particular,
4Journal of Communication (2017) © 2017 International Communication Association

J. Shin & K. Thorson Sharing Fact-Checking Messages on Social Media
central group members tend to behave in a more group-serving manner than others
because they are expected to enhance group solidarity (Hogg & Reid, 2006).
Informed by SIT, we view retweeting as a function of partisanship. We hypothe-
size that fact-checking messages that give a positive distinctiveness of a certain group
wouldberetweetedbythemembersofthatgroupsignicantlymorethanbythemem-
bers of the other group. at is, messages favorable to the Democratic Party are more
likely to be retweeted by Democrats than Republicans, whereas messages favorable to
the Republican Party are more likely to be shared by Republicans than Democrats.
H1: Fact-checking messages that are relatively advantageous to the ingroup are more likely to
be retweeted by the ingroup members than the outgroup members.
We further investigate whether Democrats and Republicans dier in their sharing
behavior. Some studies on social identity suggest that relative status of a group can
aect intergroup dynamics (Branscombe & Wann, 1994; Voci, 2006). For instance,
perceived threat to group status can lead to increased outgroup derogation in an
attempt to restore damaged self-esteem among those who strongly identify with the
group. Branscombe and Wann (1994) showed that those whose collective identity
was threatened tended to display strong derogation toward a competing outgroup
as a means of defending their social identity. Brewer (1999) argued that asymmetry
of ingroup favoritism and outgroup antagonism are to be expected across dierent
groups.
In exploring partisan group dierences in the fact-checking context, we start
with the assumption that Republicans perceived their group to be disadvantaged in
comparison to Democrats, for several reasons. First, Republicans have long claimed
that media are biased against their side (Swi, 2016). In 2012, Republican presidential
candidate Romney indicated that he “didn’t expect a fair ght in the media” (Murray,
2012), a sentiment echoed by Republican candidate Trump in 2016. Previous research
has found that Republicans, especially those embedded in homogeneous social net-
works or who consume conservative media, are more likely to perceive media as
biased (Eveland & Shah, 2003). Second, Republican politicians fared worse in the
fact-checking arena compared to Democratic politicians in recent election cycles.
According to the Center for Media and Public Aairs at George Mason University
(2013), Politifact rated Republican claims to be false three times more oen than
it rated Democratic claims during Obama’s second term. Ostermeier (2011) also
showed similar patterns. Such a consistently low performance of the Republican Party
in fact-checking led Romney’s chief pollster (Neil Newhouse) to publicly state in
2012 that “we are not going to let our campaign be dictated by fact-checkers.” Finally,
the Republican Party was the challenger’s party, out of power, in 2012. Incumbent
candidates (e.g., Obama in 2012) are known to enjoy advantages in terms of their
name recognition, funding, and government resources (Mayhew, 2008).
Given the foregoing, we investigate the possibility of asymmetric responses to
fact-checking of the candidates from Republican versus Democrat social media
users. If Republicans felt that their group was disadvantaged for the reasons above,
Journal of Communication (2017) © 2017 International Communication Association 5

Sharing Fact-Checking Messages on Social Media J. Shin & K. Thorson
Republicans may be more alert to and nd more value in sharing fact-checking
messages that derogate the outgroup leader (Obama) than messages that compliment
the ingroup leader (Romney) in comparison to the Democrats. Hence, we explore
asymmetric patterns in intergroup bias in selective sharing between the Republican
users and the Democratic users:
RQ1: Is outgroup negativity (a desire to derogate the outgroup leader) a more important
predictor of retweeting for Republicans than for Democrats?
Hostile media perception in user comments
Our data allow us to look at Twitter users’ reactions to fact-checking messages not
only in terms of sharing behaviors, but also through analysis of comments posted
in response to fact-checking messages. We use the framework of hostile media per-
ception (HMP; Gunther & Schmitt, 2004; Vallone, Ross, & Lepper, 1985) to predict
that hostile responses to fact-checking messages on Twitter will also be a function
of partisanship. HMP refers to a phenomenon in which partisans perceive neutral
media reports to be biased against their own view. Although HMP focuses on parti-
sans’ biased reactions to objectively “neutral” media coverage, it also includes actually
slanted media coverage (Gunther & Chia, 2001). For instance, liberals perceive more
bias in Fox News (conservative-leaning channel), whereas conservatives perceive more
bias in Comedy Central’s e Daily Show (liberal-leaning program) (Coe et al., 2008).
Previous research (Rojas, 2010) found that HMP leads to reactive political behav-
ior termed “corrective action.” Rojas found that people who perceive media bias were
more likely to engage in correcting public opinions through actions such as post-
ing comments in online discussion forums and commenting on news stories. ese
actions not only indicate individuals’ attempt to correct media bias, but also reveal
group-based anger and underlying political struggle for power between groups (van
Zomeren, Postmes, & Spears, 2008).
We theorize that posting hostile comments toward fact checkers is a way for
partisanstodealwithapotentialthreattothelegitimacyoftheirgroup.Asa
result, when partisans read a fact-checking message that does not depict their
group positively, they may discredit the source as a means of elevating the status
of their ingroup— undertaking an expressive form of corrective action. We have
two relevant hypotheses. First, we expect that fact-checking messages containing
relatively neutral rulings will receive comments expressing media bias from both
partisan groups. Second, we expect that fact-checking messages that give favorable
rulings to one party’s candidate will receive hostile comments mainly from outgroup
members.
H2: Neutral fact-checking messages are more likely to receive hostile comments
from both Republicans and Democrats than nonpartisan users.
H3: Fact-checking messages that are relatively advantageous to one party are more
likely to receive hostile comments from the outgroup members than the ingroup
members and nonpartisans.
6Journal of Communication (2017) © 2017 International Communication Association

J. Shin & K. Thorson Sharing Fact-Checking Messages on Social Media
We further investigate whether Republicans were more likely to take expressive
corrective action than Democrats due to the perceived disadvantages mentioned ear-
lier. Previous studies (Hartmann & Tanis, 2013; Huge & Glynn, 2010) found evidence
that perceived group status played a role in activating HMP among partisans. For
instance, Huge and Glynn (2010) found that, during a gubernatorial election cam-
paign where the Democratic Party was projected to win the election, Republicans
exhibited more HMP than Democrats. Based on this reasoning, we examine the fol-
lowing research question.
RQ2: Is the extent to which partisans express HMP through comments greater
among Republicans than among Democrats?
Methods
Data collection
We investigate how Twitter users shared and commented on fact-checking messages
posted to three major fact-checking Twitter accounts— e Ta m p a Bay Ti m es’
Politifact, Factcheck.org, and e Washington Post’s Fact Checker— in October 2012.
October was chosen because three presidential debates and one vice presidential
debate led to the highest fact-checking activity that year (Graves & Glaisyer, 2012),
andthetimepriortoanelectionusuallygarnersthegreatestpublicattentionacross
the nation (Knobloch-Westerwick & Kleinman, 2012). Each of the three fact-checking
sites had an account on Twitter and published their fact checks about the presidential
candidates during this month. Because this project draws on SIT, the scope of the
study is limited to messages pertaining to each party’s leader (Barack Obama and Mitt
Romney), who is the most prototypical member of the group.
Data analysis centered on a large collection of political tweets (N =298,894,327)
collected during the 2012 presidential election in the United States (October 2011 to
December 2012). is dataset (hereaer larger political dataset) provides a compre-
hensive picture of the political Twittersphere in 2012, as it includes all the publicly
available tweets containing at least one political keyword from the master list of 427
keywords oen found in political Twitter conversations (e.g., the names of candidates,
political parties, and issue-specic terminology).1edatawerecollectedinreal-time
using the Gnip PowerTrack service, which provides access to “the Twitter rehose.”
Unlike other streaming services, the rehose provides 100% of the publicly available
tweets, along with metadata about the tweet.
Using each fact checker’s username, the study retrieved 194 original fact checks
(tweets)2posted by Politifact (n=126), Factcheck.org (n=48), and Factchecker
(n=20) that mentioned Obama or/and Romney during October 2012 from the
larger political dataset. ese messages were retweeted or commented on 93,578
times, by 55,869 unique Twitter users.
Measurement
Fact-check ruling
In order to determine whether or not a fact-check ruling was favorable to one party or
the other, we conducted a content analysis of all the fact-check rulings in the dataset.
Journal of Communication (2017) © 2017 International Communication Association 7

Sharing Fact-Checking Messages on Social Media J. Shin & K. Thorson
e three fact-checking websites used slightly dierent styles in assessing political
statements. Politifact’s scale3was comprised of “True,” “Mostly True,” “Half True,”
“Mostly False,” “False,” and “Pants on Fire.” Fact checker rated political statements
on a range of one to four Pinocchios, with one Pinocchio for statements containing
slight omissions or exaggerations and four Pinocchios for complete lies. On the other
hand, Factcheck.org did not utilize a standardized rating scale and instead used vari-
ous linguistic markers such as “False,” “Not possible,” “Accurate,” and “Yes” to indicate
their rulings.
To obtain comparable rulings, this study recoded each fact check for three vari-
ables: (1) the political party that gained a relative advantage from the fact check,
(2a) the valence of the fact check toward Obama, and (2b) the valence of the fact
check toward Romney. e rst variable indicated the extent to which a fact check
was explicitly adjudicating the claim regarding the candidates and eventually gave
an advantage to one party over the other. is variable had three categories of val-
ues: “advantageous to the Democratic Party,” “advantageous to the Republican Party,”
and “Neutral.” If a fact check indicated that the statement under investigation was
clearly inaccurate or accurate, then it was coded either “advantageous to the Demo-
cratic Party” or “advantageous to the Republican Party.” For instance, the fact check of
“Obama ad says Romney hiked nursing home fees eight times in Mass. Mostly False.
http://t.co/b17McNST” was coded as “advantageous to the Republican Party.” e
fact check of “Obama said Romney refused to say if he’d sign Ledbetter equal pay
act. Mostly True. http://t.co/PjG1PG2y #debate” was coded as “advantageous to the
DemocraticParty.”Allotherfactchecksthatdidnotclearlyindicatewhetherthestate-
ment was false or true were coded as “Neutral.”
Second, each fact-checking tweet was coded for its valence toward each candidate:
positive, negative, or neutral for the target candidate. ese variables were necessary
to directly assess a valence of the fact check toward Obama and Romney rather than
measuring a relative advantage of the fact check. Some fact checks pitted one candi-
date’s statement against another and thus were relevant to both candidates. However,
other fact checks were only concerned about one candidate without referencing the
other. For example, a fact check of “Obama said he would cut taxes for small busi-
ness, middle class. Mostly True. Details: http://t.co/BCIzQnqs #debate” is positive to
Obama without containing negative mention of Romney.
ree undergraduate coders coded every message in the data set. Krippendor’s
αwas .87 for the relative-advantage variable, and .90 and .92 for the Obama valence
and the Romney valence variables (n=194). Any remaining disagreements among
the three coders were discussed and resolved by consensus.
User’s party preference
In order to test our hypotheses, we needed to identify the partisanship of each
Twitter user who commented on or retweeted a fact-checking message. Classifying
the partisanship of social media users on a large scale is a major challenge for
political communication research using these kinds of data sets because users do
8Journal of Communication (2017) © 2017 International Communication Association

J. Shin & K. Thorson Sharing Fact-Checking Messages on Social Media
not consistently self-identify as partisans. Our measure builds on the work of Vargo,
Guo, McCombs, and Shaw (2014). We combined three computational approaches to
develop an accurate measure of users’4partyidentication,eachofwhichreliedon
inferring partisanship by classifying the partisan sentiment of users’ other tweets in
the larger political dataset. To determine the sentiment of those tweets, we (a) used
a preexisting sentiment analysis lexicon, (b) extracted relevant sentiment terms from
the larger political dataset to extend the original lexicon, and (c) used information
about the partisan nature of users’ Twitter network.
Our starting point was to use SentiStrength, a sentiment analysis tool, which draws
on a lexicon of 2,310 sentiment words for assigning positive or negative scores to
detect the valence of a given text. Although this tool has been shown to produce
near-human-level accuracy, it performs less well for political content or controversial
topics where sarcasm or jokes are prevalent (elwall, 2013). erefore, we supple-
mented this dictionary-based approach to identifying partisan sentiment in tweets by
including partisan network measures: that is, whether a given tweet is a retweet of
Twitter accounts that Democrats and Republicans tend to follow (Barbera et al., 2015;
Vargo et al., 2014), and an additional dictionary of partisan keywords for Democrats
and Republicans relevant to the 2012 election extracted from the current data set.
We used these new dictionaries to classify the sentiment of Twitter users’ previous
tweets that mentioned either the Democratic or Republican Party. Users whose sen-
timents toward the Democrats were signicantly more positive than toward Repub-
licans were classied as Democrats, and vice versa. Users who did not have sucient
data(atleasttwoprevioustweetsthatmentionedDemocratsandtwotweetsthatmen-
tioned Republicans), as well as those whose sentiments toward the two parties were
notsignicantlydierent,werelabeledas“nonpartisanusers”astheylackedastrong
party preference on Twitter5(see Appendix S1, Supporting Information for additional
details about how we developed this method).
e accuracy of the nal data (i.e., political preference of each Twitter user) was
assessed with a random sample of 380 users (0.68%)— a sample size recommended
by McIntire and Miller (2007)— drawn from the entire population of those who had
commented on or retweeted a fact-checking message (n=55,869). Two undergradu-
ate coders independently coded each user’s party preference in the sample, informed
by the user’s entire tweets posted during the data collection period. e human-coded
political preference (i.e., single value derived by consensus coding) agreed with the
nal model 92% of the time.
Replies expressing media bias
To identify replies containing the user’s concerns about media bias, this study devel-
oped a codebook based on previous studies measuring HMP (Gunther & Leibhart,
2006; Gunther & Schmitt, 2004). e coding procedure involved three undergradu-
ate coders independently coding the same messages, a total of 1,591 replies for 194
fact checks. e coders categorized each reply as 1 if the user mentioned bias of the
factcheck(content)orthefactchecker(journalist).Ifnot,thereplywascodedas0.
Journal of Communication (2017) © 2017 International Communication Association 9

Sharing Fact-Checking Messages on Social Media J. Shin & K. Thorson
e codebook specied that the reply should be coded as 1 when it complained about
the fact checker’s analysis of political statements, even if a reply did not use a spe-
cic term such as “bias.” For example, a reply of “@politifact you are lying. Who fact
checks politifact???” was also coded as 1. On the other hand, expressions of resentment
or anger toward the target (e.g., Obama or Romney) as opposed to the fact checker
were not considered “media bias.” Intercoder reliability, measured by Krippendor ’s
αcoecient, was .79.
Results
Preliminary analysis
Fact-checking tweets
42.3% of the 194 fact-check (n=82) tweets posted by the three accounts in October
2012 contained rulings that were advantageous to the Democratic Party (i.e., either
positive to Obama or negative to Romney), while 23.7% of them (n=46) were advan-
tageous to the Republican Party (i.e., either positive to Romney or negative to Obama).
e remaining 34% (n=66) were neutral, as their statements contained either a con-
textualizedanalysisoraneutralanchor.
Inadditiontotherelativeadvantageofthefactchecks,thevalenceofthe
fact-checking tweet toward each candidate was also analyzed. Of the 194 fact checks,
34.5% (n=67) were positive toward Obama, 46.9% (n=91) were neutral toward
Obama, and 18.6% (n=36) were negative toward Obama. On the other hand, 14.9%
(n=29) of the 194 fact checks contained positive valence toward Romney, 53.6%
(n=104) were neutral toward Romney, and 31.4% (n=61) were negative valence
toward Romney.
Fact-checking users
Among the users—that is, those who retweeted or replied to a fact-checking message
in October 2012—we identied 73.79% (n=41,225) as Democrats, 9.79% (5,472)
as Republicans, and 16.42% (9,172) as nonpartisan users. e number of Democrat
users was 7.5 times larger than Republican users. ese users are a mix of regular and
eliteusers,whichincludedtheObamacampaignaccount(butnotRomney’s),media
organizations (e.g., Chicago Tribune), journalists (e.g., Terry Moran), celebrities (e.g.,
Pitbull), and political pundits (e.g., Ana Marie Cox).
When these fact-check users (n=55,869) are compared with all the other Twit-
ter users in the political dataset who have retweeted or made a comment at least
once during October 2012 (n=6,487,356), we found the following characteristics.
e fact-check users followed more accounts (Mdn =216), started using Twitter
much earlier (median tenure =139 weeks), and posted more tweets during October
(Mdn =28) than other political users (median number of followees=182, median
tenure =76 weeks, median number of tweets in October=4).Yet,fact-checkusers
did not have more followers (Mdn =116) than other users (Mdn =128). In sum,
fact-check users tend to be more active and engaged in Twitter than other political
users.
10 Journal of Communication (2017) © 2017 International Communication Association

J. Shin & K. Thorson Sharing Fact-Checking Messages on Social Media
Table 1 Results of Chi-Square Tests and Distributions of ree Political Groups Among
Retweeters of Obama Favorable and Romney Favorable Fact-Checking Messages
Democrats Republicans Nonpartisans χ2
Fact checks advantageous to
the Democratic Party
32,701 (83.06%) 591 (1.50%) 6,076 (15.43%) 22,402.0***
Fact checks advantageous to
the Republican Party
2,501 (32.13%) 3,602 (46.27%) 1,682 (21.61%) 360.27***
Note: Numbers in parentheses indicate row percentages.
***p<.001.
Hypothesis testing
Retweeting as an intergroup phenomenon
H1 stated that fact checks advantageous to the ingroup are more likely to be retweeted
by ingroup members than outgroup members. To test this hypothesis, we rst iden-
tied unique retweeters within each of three fact checks categories in terms of
relative advantage. For instance, even if a Democrat retweeted multiple fact checks
that were advantageous to the Democratic Party, we counted this person as one
Democrat. is approach makes our analysis conservative. Our data show that fact
checks advantageous to the Democratic Party were retweeted mainly by Democrats
(83.06%, n=32,701), followed by nonpartisans (15.43%, n=6,076), and Republicans
(1.50%, n=591). In contrast, fact checks advantageous to the Republican Party were
retweeted the most by Republicans (46.27%, n=3,602), followed by Democrats
(32.13%, n=2,501), and nonpartisans (21.61%, 1,682). e neutral fact checks
were retweeted by 74.04% Democrats (n=3,399), 9.89% Republicans (n=454), and
16.07% nonpartisans (n=738).
Using the distribution of Democrats, Republicans, and nonpartisan users for each
type of fact check, we ran a series of χ2tests. For the fact checks that were advanta-
geoustotheDemocraticParty,theanalysisrevealedthatthethreepoliticalgroups
of retweeters were not equally distributed, indicating the eect of partisanship on
selective sharing, χ2(2, N=39,368) =22,402, p<.01. Post hoc comparisons further
conrmed that the proportion of Democrats was signicantly higher than that of
Republicans, χ2(1, N=33,292) =19,249, p<.01.Similarly,thefactchecksadvanta-
geous to the Republican Party also showed a signicant eect of partisan group, χ2(2,
N=7,785) =360.27, p<.01, with the proportion of Republicans signicantly higher
than that of Democrats, χ2(1, N=6,103) =91.80, p<.01 (Table 1).6
Groupdierencesinretweeting
To examine whether discrediting the outgroup member is more important than
cheering their ingroup member for Republicans than for Democrats (R1), the study
compared the strength of outgroup negativity relative to ingroup positivity on
attracting retweets from Democrats and Republicans. e regression results (Table 2)
revealed that fact checks containing positive valence toward the ingroup leader
Journal of Communication (2017) © 2017 International Communication Association 11

Sharing Fact-Checking Messages on Social Media J. Shin & K. Thorson
Table 2 Regression Models Investigating Whether Valence Toward a Certain Candidate
Predicts Partisanship of Retweeters of Each Fact-Check
Proportion of Democrats Proportion of Republicans
B(SE)Sig. B(SE)Sig.
Ingroup positivity (ref. neutral) 14.17 (2.49)*** 12.96 (2.45)***
Ingroup negativity (ref. neutral) −30.57 (2.67)*** −3.43 (2.32)
Outgroup positivity (ref. neutral) −13.75 (2.61)*** −6.68 (2.21)**
Outgroup negativity (ref. neutral) 2.87 (2.61) 28.89 (2.37)***
A url (ref. none) 0.06 (4.67) 2.45 (4.15)
A hashtag (ref. none) −0.56 (2.13) 0.63 (1.90)
A mention (ref. none) −4.07 (5.74) −3.16 (5.10)
A live-tweet (ref. none) 1.25 (2.22) −1.26 (1.98)
Number of retweets 2.86 (0.99)** −0.77 (0.89)
Factcheck.org (ref. politifact) 4.11 (2.15) −2.15 (1.91)
Fact-checker (ref. politifact) 33.08 (7.75)*** −7.25 (6.90)
Observations N=174 N=174
Adjusted R20.78 0.73
Note: All variables are dummy coded except for the number of retweets, which was
log-transformed in the model. is analysis excluded fact-checking messages retweeted by less
than 10 unique retweeters.
**p<.01, *** p<.001.
wasastatisticallysignicantpredictorforbothDemocrat(B=14.17, p<.01) and
Republican retweeters (B=12.96, p<.01). However, the eect of outgroup negativity
was only signicant for predicting Republican retweeters (B=28.89, p<.01), not
Democrat retweeters (B=2.87, p>.5).
As a robustness check, we also examined the relative importance of two
variables—negative valence toward the outgroup leader and positive valence toward
the ingroup leader— in explaining variance within each model. To accomplish this
goal, we used a measure proposed by Silber, Rosenbaum, and Ross (1995) to estimate
the relative contributions of two variables. e results conrmed the stronger eect
of outgroup negativity than ingroup positivity for Republicans. More precisely, the
study found that negative valence toward the outgroup leader (Obama) explained
more variation (ω=2.40, 95% CI [1.53–3.77]) in predicting the proportion of
Republican retweeters than positive valence toward the ingroup leader (Romney).
On the contrary, negative valence toward the outgroup leader (Romney) contributed
less (ω=0.34, 95% condence interval [CI] [0.12–0.98]) to predicting the proportion
of Democrat retweeters for fact checks than positive valence toward the ingroup
leader (Obama).
Hostilecommentsasanintergroupphenomenon
e fact checks had a relatively small number of replies (n=1,591)—just a fraction
(1.73%) of the total number of retweets (n=91,987). Of these 1,591 replies, 32.4%
12 Journal of Communication (2017) © 2017 International Communication Association

J. Shin & K. Thorson Sharing Fact-Checking Messages on Social Media
Table 3 Results of Chi-Square Tests and Distributions of ree Political Groups Among
HostileCommentersofObamaFavorableandRomneyFavorableFact-CheckingMessages
Democrats Republicans Nonpartisans χ2
Fact checks advantageous to
the Democratic Party
31 (10.99%) 237 (84.04%) 14 (4.96%) 152.79***
Fact checks advantageous to
the Republican Party
99 (81.15%) 7 (5.74%) 16 (13.11%) 59.07***
Note: Numbers in parentheses indicate row percentages.
***p<.001.
(n=515) contained concerns about the bias of the fact check or the fact checker.
ese include replies directly using the words “bias” or “unfair” in their comment such
as “@politifact Horrible, biased interpretation. No proof Obama said he would ‘cre-
ate daylight.’ Your interpretation is NOT a FACT check.” Additionally, these include
remarksthatcomplainedabouttherulingofthefactcheckandsuggesttheirownevi-
dence such as “http://t.co/MVhnM7Q1 If I may bring this up. anks for reading and
educating yourself. @politifact.”
We identied unique hostile media commenters within each category of fact
checks in terms of relative advantage (see Tables 3 and 4 for more information). H2
and H3 were concerned with the extent to which fact checks received comments
expressing concern about media bias. A series of χ2tests were performed to determine
whether the hostile responses from Republicans, Democrats, and nonpartisans were
signicantly dierent. H2 predicted that neutral fact checks would be more likely
to receive comments containing accusations of media bias from both Republicans
and Democrats than nonpartisans. Of those who replied to neutral fact checks
mentioning bias, 49.38% were Democrat (n=40), 41.98% were Republicans (n=34),
and 8.64% were nonpartisan users (n=7). A chi-square test showed that hostile
commenters were not equally distributed among the three political groups, χ2(2,
N=81) =15.09, p<.001. Post hoc comparisons revealed that both the proportions
of Democrats and Republicans were signicantly higher than that of nonpartisan
users, yielding χ2(1, N=47) =12.34, p<.01 and χ2(1, N=41) =9.61, p<.05. Yet,
the proportions of Democrats and Republicans were not signicantly dierent, χ2(1,
N=74) =0.08, p>.05.
H3 stated that fact-checking messages that are relatively advantageous to one polit-
ical party are more likely to receive hostile comments from the outgroup members
than the ingroup members and nonpartisans. For the Democratic Party favorable
fact checks, the proportions of hostile commenters among three groups were signi-
cantly dierent, χ2(2, N=282) =52.79, p<.001. e post hoc analyses showed that
the proportion of Republicans was signicantly higher than that of Democrats and
nonpartisans for the fact checks that were advantageous to the Democratic Party, χ2
(1, N=268) =81.10, p<.01 and X2(1, N=251) =111.97, p<.01. Similarly, for the
Journal of Communication (2017) © 2017 International Communication Association 13

Sharing Fact-Checking Messages on Social Media J. Shin & K. Thorson
Table 4 Distributions of ree Political Groups Among Hostile Commenters of Neutral
Fact-Checking Messages in Comparison to Retweeters
Hostile Commenters of Neutral Fact
Checks Versus Retweeters
Democrats Republicans Nonpartisans
Hostile commenters 40 (49.38%) 34 (41.98%) 7 (8.64%)
Retweeters (base line) 34,339 (74.42%) 4,145 (8.98%) 7,656 (16.59%)
Note: Numbers in parentheses indicate row percentages.
Republican Party favorable fact checks, the distribution of the three groups of hos-
tile commenters was again signicantly dierent, χ2(2, N=122) =59.07, p<.001.
Post hoc tests revealed that the proportion of Democrats were signicantly higher
than that of Republicans and nonpartisans, χ2(1, N=106) =43.54, p<.01 and χ2(1,
N=115) =28.59, p<.01.
Group dierences in hostile comments
RQ2 examined whether hostile media comment toward fact checkers was stronger
among Republicans than Democrats. e post hoc tests of H2 already indicated that
there was no signicant dierence between the proportions of Democrats (49.38%)
and Republicans (41.98%) in terms of posting hostile comments to neutral fact checks.
In other words, these neutral fact-checking messages were accused of “being partial”
by the approximately same number of Democrats and Republicans.
Yet, this analysis assumed the equal distribution of the three groups (33%
Democrat, 33% Republican, and 33% nonpartisan) and examined whether hostile
commenters deviated from such distribution. A more realistic comparison should
be based o the distribution of retweeters, which was disproportionally domi-
natedbyDemocrats.Assuch,whenweexaminedtheobserveddistributionof
corrective commenters against the distribution of retweeters of the 194 fact checks,
the results painted a dierent picture. e tests revealed that the proportion of
Republican hostile commenters was signicantly higher than that of Democrats
and nonpartisans, χ2(1, N=74) =90.96 p<.01 and χ2(1, N=41) =38.82, p<.01.
is means that the extent to which Republicans perceived neutral fact checks as
hostile (41.98%) considering their retweeting participation (8.98%) is greater than
the extent to which Democrats engaged in HMP (49.38%) given their retweet-
ing participation (74.42%). Results are presented in Table 4. We also found fewer
overlaps between fact check retweeters and hostile commenters among Repub-
licans (15.6% of Republican hostile commenters also retweeted fact checkers)
than Democrats (47.3% of Democratic hostile commenters also retweeted fact
checkers).
Additionally, to examine the extent to which power users (e.g., political pundits,
journalists) inuenced the results, we ran supplementary analyses excluding the top
14 Journal of Communication (2017) © 2017 International Communication Association

J. Shin & K. Thorson Sharing Fact-Checking Messages on Social Media
1% users based on the number of followers (n=1,170). We found that the results
were not substantially dierent from the main analysis.
Discussion
is study examined two types of reactions among social media users to political
fact-checking messages. Unlike previous studies focused on the eects of consuming
fact checks on political knowledge (Fridkin et al., 2015; Nyhan & Reier, 2015; Wood
& Porter, 2016), the current study emphasized fact-check consumers’ voluntary shar-
ing and commenting behavior in a public forum. is is an important task, because
while only a small number of Internet users regularly visit fact-checking sites, those
who publicly share fact-checking messages can inuence others by increasing visibil-
ityofthesemessages,aswellasaectinginterpretationsoftheoriginalfactchecks
(Cappella, Kim, & Albarracín, 2015).
We drew on SIT as a route to theorize the role of partisanship in selective shar-
ing and the expression of bias perceptions. Based on this framework, we found that
the fact-checking messages served as a tool for partisans to celebrate their own group
and denigrate the opposing group. Fact checks that were advantageous to a candi-
date from the ingroup party were shared signicantly more by the ingroup mem-
bers than the outgroup members, a process of partisan selective sharing. is nding
has implications for the extant literature on political polarization in the contempo-
rary media environment. Studies of selective exposure and political polarization have
mainlyfocusedonpartisanmediaoutletssuchasFoxorMSNBC,ratherthanmain-
stream media outlets that emphasize balance and fairness (Iyengar & Hahn, 2009;
Levendusky, 2013; Sunstein, 2001). Our study of selective sharing on social media
adds another dimension, showing that partisans cherry-pick favorable media con-
tent from relatively balanced media outlets when they are deciding what content to
make visible to their personal social networks. In this scenario, media consumers
themselves play the role of partisan sources in social media by selectively feeding
information to their followers.
Further analysis of the users in our data revealed that these individuals (both
retweeters and repliers of fact checks) are also more active and engaged than other
users who tweeted about politics during the 2012 election cycle. is nding suggests
thatthevisibilityoffact-checkingmessagesmayheavilydependonasmallnum-
ber of active users. ese users can control the ow of fact-checking information
by reassessing the initial editorial decisions and ltering what is newsworthy for
their followers (Singer, 2014). In our dataset, the original fact checkers’ messages
potentially reached an audience 354 times larger because of retweeting and com-
menting. Users who commented on or shared these fact-checking messages had a
total of 78,726,217 followers, whereas the three fact checkers had 222,513 followers.
is means that many of those who do not directly follow fact checkers may con-
sume one-sided fact-checking information selectively disseminated by their Twitter
contacts.
Journal of Communication (2017) © 2017 International Communication Association 15

Sharing Fact-Checking Messages on Social Media J. Shin & K. Thorson
Our ndings also highlight fascinating asymmetries between Democrats and
Republicansintheiruseoffact-checkingmessages.Overall,thereweremanymore
Democrats than Republicans in our fact check user dataset, which suggests that more
Democrats than Republicans follow fact check organizations on Twitter. Further, we
found that Democrats and Republicans diered in the extent to which they shared
fact checks that were negative toward the outgroup party. Fact checks that were
negative toward Obama were more likely to attract Republican retweeters, whereas
fact checks negative toward Romney did not attract Democrat retweeters. In addition,
there was a relatively larger eect of outgroup negativity than ingroup positivity for
Republicans, but not for Democrats.
Although we drew on SIT to predict selective sharing behavior, we believe the
implications of our ndings go beyond conrming predictions of SIT. A growing
body of literature in communication has theorized connections between patterns
of media behavior grounded in social identity and long-term, society-level eects
on political polarization, political knowledge, and misperceptions (e.g., Feldman
et al., 2014; Garrett et al., 2016; Slater, 2007). Slater’s (2007) reinforcing spirals
model proposed that media behavior consistent with a particular social identity is
involved in a mutually reinforcing relationship with that identity: Identity-relevant
media use increases identity salience and identity salience motivates further use
ofidentity-relevantmedia,producingaspiralofeects.roughthisprocessof
reinforcing spirals, the media behavior of particular individuals is translated into
broadinuenceoverthecultureofsocialgroups.Slaterpointsoutthatreinforcing
spirals are particularly likely among social groups that are more closed or suspicious
ofinuencesoutsidetheirsocialgroup,suchasfromthemainstreammedia.
Ourdatadonotallowustotestalongitudinalchainofinuence,butwesuspect
thatselectivesharing,aectivepolarization,andconservativeperceptionsofmedia
bias are related to each other in just such a reinforcing spiral. Selective sharing of
fact-check messages is motivated by social identity, which itself has been shaped
by exposure to media messages that emphasize polarization between parties, and
negativity (Iyengar et al., 2012). In the case of conservatives, social identity has been
shaped by immersion in an ideological media environment that promotes a “culture
of mistrust toward other viewpoints” (Feldman et al., 2014). at the Republicans in
our data engaged in greater negativity in their retweets and responses to fact checkers
is evidence both of this longstanding culture of mistrust and the social psychological
processes that we believe contribute to the development— and reinforcement—of
this culture over time.
Consistent with the SIT framework and previous research on HMP, we observed
that expressing hostility toward the fact checkers was also a function of the Twitter
user’s partisanship. Neutral fact-checking messages received hostile comments both
from Democrats and Republicans more than nonpartisans. In addition, fact checks
that were advantageous to one party received signicantly more hostile comments
from the outgroup members than ingroup members. ese ndings suggest that bias
perceptions are in the eye of the beholder. For example, in response to fact-checking
16 Journal of Communication (2017) © 2017 International Communication Association

J. Shin & K. Thorson Sharing Fact-Checking Messages on Social Media
messages rendering a “half true,” “mixed,” or “complicated” ruling to Obama or
Romney, both Democrats and Republicans posted similar hostile comments, such as
“@politifact No, Your fact check as a fact is a lie. Obama is correct, period! Rachel
Maddow is right about your organization!” (from a Democrat user) and “@politifact
yourchoicesofwhoandwhattofactcheckclearlydemonstrateyourlewinged
bias” (from a Republican user). Such a mirroring pattern was aptly recognized by one
nonpartisan user who commented, “is is why no one likes you.”
We also observed some evidence that Republicans expressed concerns about
media bias in fact checks more than Democrats did. Although there was no sig-
nicant dierence in the number of hostile commenters between the two groups,
considering a disproportionately large number of Democratic retweeters in the
current dataset, the level of engagement of Republicans in posting hostile comments
reects their greater resentment toward the fact checkers. is nding is consis-
tent with previous studies (Hartmann & Tanis, 2013; Huge & Glynn, 2010), which
observed a stronger hostile media eect for those whose group had a relatively lower
standing, as well the “culture of mistrust” of media among conservatives.
ese ndings have critical implications for fact-checking practitioners. Read-
ers’ comments about media content— especially negative opinions—can inuence
other viewers’ evaluation of a news story (orson, Vraga, & Ekdale, 2010; Lee &
Jang, 2010). erefore, users’ comments depicting fact checkers as biased sources may
not only hinder the fact-checking organization’s ability to eectively challenge misin-
formation by damaging their reputation, but also can lead to avoidance of reading
messages produced by that source. Writ large—and as fact-checking expands as a
media practice— hostile responses to fact-check messages may further contribute to
public perceptions of media bias through the very mechanisms we identify in this
paper (we thank an anonymous reviewer for this suggestion).
Our study has a number of limitations. First, we only focused on the manifest
behaviors of social media users such as retweeting and commenting, rather than their
exposure to fact checks. We recommend that future research examine a potential dis-
crepancy between selective exposure and selective sharing of fact checks. Second,
we focused on partisans’ responses to fact-checking messages as opposed to those
of nonpartisans. Although a majority of fact-checking users in the current dataset
(83.58%) were identied as either Democrats or Republicans, examining how nonpar-
tisans consume fact-checking messages would provide useful insights for practition-
ers. ird, this study categorized a number of dierent rulings (e.g., False and Pants on
Fire) into the same category. Future research dierentiating linguistic markers would
help us better understand the eects of message framing on partisanship activation.
In addition, we acknowledge that users may have retweeted fact checks without actu-
ally reading the full content, and our dataset may contain automated bots set up by
political or commercial entities.
Journal of Communication (2017) © 2017 International Communication Association 17

Sharing Fact-Checking Messages on Social Media J. Shin & K. Thorson
Conclusion
Previous research has generally evaluated the eects of fact-checking on recip-
ients’ political knowledge through experiments and concluded that exposure to
fact-checking is eective regardless of the recipient’s prior partisan belief. Our study
demonstrates that real-world exposure to fact checks may not be as random as in the
experimentalsetting.Wefoundthatpartisanfact-checkuserssharedfact-checking
messagesselectivelyonTwittersuchthattheyretweetedonlymessagesthatwere
favorable to a candidate from their own political party and ltered fact checks that
supported the opposing party. is partisan selective sharing—undertaken by a
small but highly active partisan group of Twitter users — served to bias the visibility
of fact-check messages to a much wider audience.
Partisan selective sharing is a process whereby partisan groups reassess the initial
editorial decisions of the media and dene the organization’s content for their fol-
lowers. We propose that partisan selective sharing is a phenomenon that is ripe for
further study within political communication and journalism studies, particularly as
social media platforms grow in importance as sites for exposure to news and politi-
cal information. e phenomenon of partisan selective sharing may serve to further
polarize audiences as well as undercut trust in the process of fact-checking.
We observed that while Democrats make up the majority of fact-check sharers,
Republicans exhibit stronger hostility toward fact checkers. Such asymmetry between
partisan groups in their media behavior and perception raises concerns over a
reinforcing spiral: A partisan individual’s group identity aects his or her use of
identity-relevant media and such media use increases identity salience for themselves
and their social contacts. As such, partisan selective sharing proposes a fruitful
approach to examine political polarization in social networking sites by emphasizing
relationships among media, active partisan users, and broader audiences.
Acknowledgments
We are grateful to Lian Jian, Margaret McLaughlin, and François Bar for their thought-
ful and helpful comments. We also thank the editor and three anonymous reviewers
for their constructive feedback and support during the review process. is project
was supported by the Annenberg School for Communication & Journalism, Univer-
sity of Southern California.
Notes
1 elistofkeywordscanbeprovideduponrequest.
2 is method excludes retweets or promotional messages (e.g., A tweet from Politifact: “If
you hear something during the debate that you’d like us to fact-check, tweet it with
#PolitiFactis”).
3 ey also use “Promise Kept,” “Compromise,” “Promise Broken,” “No Flip,” “Half Flip,”
and “Full op.”
18 Journal of Communication (2017) © 2017 International Communication Association

J. Shin & K. Thorson Sharing Fact-Checking Messages on Social Media
4 Users are dened as those who have retweeted or posted a comment to one of the 194
fact-checking messages at least once. A total of 55,869 such users posted 25,983,635 tweets
between January and December 2012.
5 is label did not mean that they were truly independent users; rather, it indicated at the
veryleastthattheseusersdidnotuseTwitterasaplatformtopubliclyexpresstheir
political party preference.
6 ese results hold when tested in a multivariate context as well, controlling for the
originating fact-check organization as well as message variables such as whether it
contained a hashtag (#) or a url, whether it mentioned someone using “@username”
convention, and whether it was a live-tweet posted during the presidential and vice
presidential debates. Analysis available upon request.
References
Adamic, L., & Glance, N. (2005). e political blogosphere and the 2004 US election: Divided
they blog. In Proceedings of the 3rd international workshop on Link discovery (pp. 36–43).
New York, NY: ACM Press.
Bakshy, E., Messing, S., & Adamic, L. (2015). Exposure to ideologically diverse news and
opinion on Facebook. Science,348(6239), 1130–1132.
Barbera, P., Jost, J. T., Nagler, J., Tucker, J. A., & Bonneau, R. (2015). Tweeting from le to
right: Is online political communication more than an echo chamber? Pyschological
Science,26, 1531– 1542.
Boutet, A., Kim, H., & Yoneki, E. (2012). What’s in Twitter: I know what parties are popular
and who you are supporting now! In Proceedings of the 2012 International Conference on
Advances in Social Networks Analysis and Mining, New York, NY, pp. 132– 129.
Branscombe, N. R., & Wann, D. L. (1994). Collective self-esteem consequences of outgroup
derogation when a valued social identity is on trial. European Journal of Social Psychology,
24, 641–657.
Brewer, M. B. (1999). e psychology of prejudice: Ingroup love and outgroup hate? Journal
of Social Issue,55(3), 429 – 444.
Cappella, J. N., Kim, H. S., & Albarracín, D. (2015). Selection and transmission processes for
information in the emerging media environment: Psychological motives and message
characteristics. Media Psychology,18(3), 396 –424.
Center for Media and Public Aairs at George Mason University. (2013). Study: Media
fact-checker says Republicans lie more. Retrieved from http://cmpa.gmu.edu/study-
media-fact-checker-says-republicans-lie-more
Coe, K., Tewksbury, D., Bond, B. J., Drogos, K. L., Porter, R. W., Yahn, A., & Zhang, Y. (2008).
Hostile news: Partisan use and perceptions of fable news programming. Journal of
Communication,58(2), 201–219.
Colleoni, E., Rozza, A., & Arvidsson, A. (2014). Echo chamber or public sphere? Predicting
political orientation and measuring political homophily in Twitter using Big Data. Journal
of Communication,64(2), 317–332.
Conover, M.D., Ratkiewicz, J., Francisco, M., Goncalves, B., Flammini, A., & Menczer, F.
(2011). PoliticalpolarizationonTwitter.Fih International AAAI Conference on Weblogs
and Social Media. Barcelona, Spain: AAAI Press.
Eveland, W., & Shah, D. (2003). e impact of individual and interpersonal factors on
perceived news media bias. Political Psychology,24(1), 101 –117.
Journal of Communication (2017) © 2017 International Communication Association 19

Sharing Fact-Checking Messages on Social Media J. Shin & K. Thorson
Feldman, L., Myers, T., Hmielowski, J., & Leiserowitz, A. (2014). e mutual reinforcement of
media selectivity and eects: Testing the reinforcing spirals framework in the context of
global warming. Journal of Communication,64(4), 590 – 611.
Fridkin, K., Kenney, P. J., & Wintersieck, A. (2015). Liar, liar, pants on re: How fact-checking
inuences citizens’ reactions to negative advertising. Political Communication,32(1),
127–151.
Friggeri, A., Adamic, L. A., Eckles, D., & Cheng, J. (2014). Rumor cascades. In Proceedings of
the Eighth International AAAI Conference on Weblogs and Social Media (pp. 101 –110).
Ann Arbor, MI: AAAI Press.
Garrett, R. K. (2009). Politically motivated reinforcement seeking: Reframing the selective
exposure debate. Journal of Communication,59(4), 676–699.
Garrett, R. K., Weeks, B. E., & Neo, R. L. (2016). Driving a wedge between evidence and
beliefs: How online ideological news exposure promotes political misperceptions. Journal
of Computer Mediated Communication,21(5), 331–348.
Glynn, C., & Huge, M. (2014). How pervasive are perceptions of bias? Exploring judgments
of media bias in nancial news. International Journal of Public Opinion Research,26(4),
543–553.
Goman, E. (1959). e presentation of self in everyday life.NewYork,NY:AnchorBooks.
Gottfried, J. A., & Shearer, E. (2016, 26 May). News use across social media platforms 2016.
Pew Research Center. Retrieved from http://www.journalism.org/2016/05/26/news-use-
across-social-media-platforms-2016/
Graves, L. (2016). Decidingwhat’strue:eriseofpoliticalfact-checkinginAmerican
journalism.NewYork,NY:ColumbiaUniversityPress.
Graves, L., & Glaisyer, T. (2012, February). e fact-checking universe in Spring 2012.New
America. Retrieved from https://www.newamerica.org/oti/policy-papers/the-fact-
checking-universe-in-spring-2012/.
Graves, L., Nyhan, B., & Reier, J. (2016). Field experiment examining motivations for
fact-checking. Journal of Communication,66(1), 102–138.
Green, D., Palmquist, B., & Schickler, E. (2002). Partisan hearts and minds: Political parties
and the social identities of voters. London, England: Yale University Press.
Gunther, A. C., & Chia, S. C. (2001). Predicting pluralistic ignorance: e hostile media
perception and its consequences. Journalism & Mass Communication Quarterly,78(4),
688–701.
Gunther, A. C., & Leibhart, J. L. (2006). Broad reach or biased source? Decomposing the
hostile media eect. Journal of Communication,56, 449–466.
Gunther, A. C., & Schmitt, K. (2004). Mapping boundaries of the hostile media eect. Journal
of Communication,54(1), 55–70.
Hannak, A., Margolin, D., Keegan, B., & Weber, I. (2014). Get back! You don’t know me like
that: e social mediation of fact checking interventions in Twitter conversation. In
Proceedings of the Eighth International AAAI Conference on Weblogs and Social Media.
Palo Alto, CA: AAAI Press.
Hartmann, T., & Tanis, M. (2013). Examining the hostile media eect as an intergroup
phenomenon: e role of intergroup identication and status. Journal of Communication,
63, 535–555.
Hogg, M. (2001). A social identity theory of leadership. Personality and Social Psychology
Review,5(3), 184 – 200.
20 Journal of Communication (2017) © 2017 International Communication Association

J. Shin & K. Thorson Sharing Fact-Checking Messages on Social Media
Hogg, M., & Reid, S. (2006). Social identity, self-categorization, and the communication of
group norms. Communication eory,16(1), 7–30.
Huge, M., & Glynn, C. J. (2010). Hostile media and the campaign trail: Perceived media bias
intheraceforgovernor.Journal of Communication,60, 165 –181.
Iyengar, S., & Hahn, K. S. (2009). Red media, blue media: Evidence of ideological selectivity
in media use. Journal of Communication,59, 19–39.
Iyengar, S., Sood, G., & Lelkes, Y. (2012). Aect, not ideology: A social identity perspective
on polarization. Public Opinion Quarterly,76(3), 405–431.
Jacobson, S., Myung, E., & Johnson, S. L. (2015). Open media or echo chamber: e use of
links in audience discussions on the Facebook pages of partisan news organizations.
Information, Communication & Society,19(7), 875– 891.
Knobloch-Westerwick, S., & Kleinman, S. B. (2012). Preelection selective exposure:
Conrmationbiasversusinformationalutility.Communication Research,39(2), 170–193.
Knobloch-Westerwick, S., & Meng, J. (2009). Looking the other way: Selective exposure to
attitude-consistent and counterattitudinal political information. Communication
Research,36(3), 426– 448.
Lee, E., & Jang, Y. (2010). What do others’ reactions to news on internet portal sites tell us?
Eects of presentation format and readers’ need for cognition on reality perception.
Communication Research,37(6), 825 – 846.
Levendusky, M. (2013). Why do partisan media polarize viewers? American Journal of
Political Science,57(3), 611–623.
Marwick, A. E., & Boyd, D. (2011). I tweet honestly, I tweet passionately: Twitter users,
context collapse, and the imagined audience. New Media & Society,13(1), 114– 133.
Mayhew, D. R. (2008). Incumbency advantage in U.S. Presidential elections: e historical
record. Political Science Quarterly,123(2), 201–228.
McIntire, S. A., & Miller, L. A. (2007). Foundations of psychological testing: A practical
approach. London, England: Sage.
Meraz, S., & Papacharissi, Z. (2013). Networked gatekeeping and networked framing on
#Egypt. e International Journal of Press/Politics,18(2), 138– 166.
Murray, S. (2012, April). Romney species deductions he’d cut. e Wall Street Journal.
Retrieved from https://www.wsj.com/articles/SB1000142405270230443270457734661
1860756628.
Nahon, K., & Hemsley, J. (2013). Going viral. Malden, MA: Polity.
Nyhan, B., & Reier, J. (2015). Estimating fact-checking’s eects: Evidence from a long term
experiment during campaign 2014. Retrieved from http://www.americanpressinstitute
.org/wp-content/uploads/2015/04/Estimating-Fact-Checkings-Eect.pdf
Oaks, P. J. (1987). e salience of social categories. In J. C. Turner, M. A. Hogg, P. J. Oakes, S.
D.Reicher,&M.S.Wetherell(Eds.),Rediscovering the social group: A self-categorization
theory (pp. 117–141). Oxford, England: Blackwell.
Ostermeier, E. (2011). Selection bias? Politifact rates Republican statements as false as 3 times
the rate of Democrats. Retrieved from http://editions.lib.umn.edu/smartpolitics/2011/02/
10/selection-bias-politifact-rate/
Rojas, H. (2010). Corrective actions in the public sphere: How perceptions of media and
media eects shape political behaviors. International Journal of Public Opinion Research,
22(3), 343–363.
Journal of Communication (2017) © 2017 International Communication Association 21

Sharing Fact-Checking Messages on Social Media J. Shin & K. Thorson
Shin, J., Jian, L., Driscoll, K., & Bar, F. (2016). Political rumoring on Twitter during the 2012
U.S. Presidential election: Rumor diusion and correction. New Media & Society.
Advance online publication. doi:10.1177/1461444816634054
Silber, J. H., Rosenbaum, P. R., & Ross, R. N. (1995). Comparing the contributions of groups
of predictors: Which outcomes vary with hospital rather than patient characteristics.
JournaloftheAmericanStatisticalAssociation,90(429), 7– 18.
Singer, J. B. (2014). User-generated visibility: Secondary gatekeeping in a shared media space.
New Media & Society,16(1), 55– 73.
Slater, M. D. (2007). Reinforcing spirals: e mutual inuence of media selectivity and media
eects and their impact on individual behavior and social identity. Communication
eory,17, 281 – 303.
Stroud, N. J. (2008). Media use and political predispositions: Revisiting the concept of
selective exposure. Political Behavior,30(3), 341–366.
Sunstein, C. (2001). Republic.com 2.0.Princeton,NJ:PrincetonUniversityPress.
Swi, A. (2016). Americans’ trust in mass media sinks to new low. Gallup. Retrieved from
http://www.gallup.com/poll/195542/americans-trust-mass-media-sinks-new-low.aspx.
Tajfel, H., & Turner, J. C. (1979). An integrative theory of intergroup conict. e Social
Psychology of Intergroup Relations,33(47), 74.
elwall, M. (2013). Heart and soul: Sentiment strength detection in the social web with
SentiStrength. Retrieved from http://sentistrength.wlv.ac.uk/documentation/
SentiStrengthChapter.pdf
orson, E. (2016). Belief echoes: e persistent eects of corrected misinfomration. Political
Communication,33(3), 460–480.
orson, K., Vraga, E., & Ekdale, B. (2010). Credibility in context: How uncivil online
commentary aects news credibility. Mass Communication and Society,13(3), 289 –313.
Turner, J. C., Hogg, M. A., Oakes, P. J., Reicher, S. D., & Wetherell, M. S. (1987). Rediscovering
the social group: A self-categorization theory. Oxford, England: Blackwell.
Turner, J. C., & Oaks, P. J. (1986). e signicance of the social identity concept for social
psychology with reference to individualism, interactionism, and social inuence. British
JournalofSocialPsychology,25, 237 –252.
Valentino, N., Banks, A., Hutchings, V., & Davis, A. (2009). Selective exposure in the Internet
Age: e interaction between anxiety and information utility. Political Psychology,30(4),
591–613.
Vallone, R. P., Ross, L., & Lepper, M. R. (1985). e hostile media phenomenon: Biased
perception and perceptions of media bias in coverage of the Beirut massacre. Journal of
Personality and Social Psychology,49(3), 577.
Van Zomeren, M., Postmes, T., & Spears, R. (2008). Toward an integrative social identity
modelofcollectiveaction:Aquantitativeresearchsynthesisofthreesocio-psychological
perspectives. Psychological Bulletin,134(4), 504– 535.
Vargo, C., Guo, L., McCombs, M., & Shaw, D. (2014). Network issue agendas on Twitter
during the 2012 U.S. Presidential election. Journal of Communication,64(2), 296 –316.
Voci, A. (2006). e link between identication and in-group favouritism: Eects of threat to
social identity and trust-related emotions. British Journal of Social Psychology,45(2),
265–284.
Weeks, B. (2015). Emotions, partisanship, and misperceptions: How anger and anxiety
moderate the eect of partisan bias on susceptibility to political misinformation. Journal
of Communication,65, 699–719.
22 Journal of Communication (2017) © 2017 International Communication Association

J. Shin & K. Thorson Sharing Fact-Checking Messages on Social Media
Weeks, B., & Holbert, L. (2013). Predicting dissemination of news content in social media: A
focus on reception, friending, and partisanship. Journalism & Mass Communication
Quarterly,90(2), 212–232.
Wood, T., & Porter, E. (2016). e elusive backre eect: Mass attitudes’ steadfast factual
adherence. Available from https://ssrn.com/abstract=2819073
Supporting Information
Additional supporting information may be found in the online version of this article:
Appendix S1. Measuring user’s party preference.
Journal of Communication (2017) © 2017 International Communication Association 23