Professional Documents
Culture Documents
The Blue Check of Credibility: Does Account Verification Matter When Evaluating News On Twitter?
The Blue Check of Credibility: Does Account Verification Matter When Evaluating News On Twitter?
Abstract
Downloaded by East Carolina University from www.liebertpub.com at 03/12/19. For personal use only.
The increased reliance on social network sites for news and the proliferation of partisan news have refocused
scholarly attention on how people judge credibility online. Twitter has faced scrutiny regarding their practices
in assigning the ‘‘verified’’ status to Twitter accounts, but little work has investigated whether users apply this
cue in making assessments for information quality. Using an experimental design, we test whether the Twitter
verification mark contributes to perceptions of information and account credibility among news organizations.
We additionally consider how account ambiguity and account congruence with political beliefs condition this
relationship. Our results suggest little attention is paid to the verification mark when judging credibility, even
when little other information is provided about the account or the content. Instead, account ambiguity and
congruence dominate credibility assessments of news organizations. We propose that Twitter may need to
revise their verification badges to increase their salience or provide more information to users. Currently, users
appear to rely on other cues than the verification label when judging information quality.
1
Medill School of Journalism, Northwestern University, Evanston, Illinois.
2
Department of Communication, George Mason University, Fairfax, Virginia.
1
2 EDGERLY AND VRAGA
themselves as news. Twitter users may look for this norm Turk Prime interface on February 5, 2018.* Participants were
violation, much like they look for misspellings or the design offered $0.60 for participating in a 6-minute survey. Participants
layout of websites,7 as a quick cue about the source or were 36 years old (M = 36.13, SD = 11.42), 50% male, and 80%
message credibility. Given this, we expect: white, with 49% of the sample having at least a 2-year degree.
After answering a short pretest questionnaire, participants
H1: The presence of a verification check will result in higher were randomly assigned to 1 of 12 conditions across three
evaluations of (a) tweet credibility and (b) account credibility.
experimental factors. Across all conditions, participants
We also consider whether verification intersects with two viewed a single tweet, with the content held consistent and
additional source-related factors. First, we expect verification to ideologically neutral, and then answered a series of questions
have a differential impact based on level of source ambiguity. about that specific tweet (Fig. 1). First, we manipulated ac-
Under conditions of low ambiguity—when Twitter users have count verification of the source of the tweet, as indicated by
ample knowledge of an account—verification should have the presence (or absence) of the blue check mark used by
limited influence. The credibility of prominent mainstream Twitter. Second, we manipulated the ideological positioning
news sources, for example, will be uninfluenced by verification of the account that posted the tweet such that three sources
because these sources have name recognition and reputations favored Republicans, whereas three sources favored Demo-
useful in assessing credibility.8 It is under conditions of in- crats.{ We then matched these sources with participants’ party
creased ambiguity that verification should matter. High ambi- affiliation (N = 319 Democrats, N = 167 Republicans), as re-
guity exists when Twitter users encounter an unknown source; ported on a 7-point scale in the pretest questionnaire, to create
Downloaded by East Carolina University from www.liebertpub.com at 03/12/19. For personal use only.
while medium ambiguity exists when a source is unknown, but a measure of congruence (e.g., source matched participants’
the account name provides cues about the source (i.e., the name affiliation) versus incongruence. We classified those who
‘‘Conservative Tribune’’ gives users information about the ‘‘leaned’’ toward a party as partisans, and excluded true In-
source, even if users are unfamiliar with it). dependents (N = 130) from our analyses. Third, we manipu-
lated the ambiguity of the account that posted the tweet: low
H2: The presence of a verification check will have a larger ambiguity, defined as mainstream partisan sources (Fox
impact on evaluations of (a) tweet credibility and (b) ac- News, MSNBC), medium ambiguity, defined as lesser-known
count credibility when combined with medium and high sources with partisan cues (Conservative Times, Forward
ambiguity accounts, than when it is combined with low
ambiguity accounts. Progressives), and high ambiguity, defined as lesser-known
sources without partisan cues (Daily Caller, Palmer Report).
RQ1: Will a verification check influence (a) tweet cred-
Measures
ibility and (b) account credibility more for high versus
medium ambiguity accounts? Tweet credibility. A series of 7-point semantic differen-
tials asked participants to evaluate the ‘‘content of the tweet’’
Second, people often judge sources and messages that are as: complete/incomplete, accurate/inaccurate, unbiased/bi-
congruent with their political ideology to be more credible.6 ased, trustworthy/not trustworthy, credible/not credible, and
We explore how this process interacts with verification sta- tells the whole story/not tells the whole story to measure
tus. On one hand, conditions of congruency and incon- tweet credibility. These items were averaged to form a scale
gruency may reach a sufficing principle—these are primary (M = 3.92, SD = 1.13, a = 0.82).
and powerful heuristics that allow people to make credibility
judgments.9 As such, verification status may not matter. On Account credibility. Participants next evaluated the
the other hand, subtle ‘‘proximal cues’’ that accompany news account that posted the tweet as: unbiased/biased, trustworthy/
stories (e.g., timestamps) can increase judgments of credi- untrustworthy, unprofessional/professional, not credible/
bility, but only for low-credibility news sources.8 Extending credible which we averaged (M = 4.11, SD = 1.31, a = 0.87).
this logic, a verification check may increase credibility, but
only for incongruent sources (typically seen as less credible). Manipulation checks
Given competing expectations, we ask:
To ensure people were paying adequate attention, we
RQ2: Will a verification check influence (a) tweet cred- asked participants at the end of the posttest questionnaire to
ibility and (b) account credibility more for congruent versus report the account that posted the tweet they just saw out of
incongruent accounts? nine possible sources. Overall, 79.4% of the sample correctly
identified the source of the tweet. A logistic regression
Our final research question tests whether verification de- suggests that correct identification of the source was not
pends both on ambiguity and congruence. predicted by verification or congruence, but low ambiguity
sources were more often identified than high ambiguity
RQ3: What influence do various combinations of ac-
count cues (verification status, ambiguity, congruency) *This sample was extracted from a larger experimental design,
have on tweet and account credibility? which included three non-partisan news sources that were outside of
the scope of this article. These conditions were not crossed with the
design here and were excluded from all analyses (N = 322).
{
The ideological positioning of the sources was taken from
Methods the media bias chart, available here: www.allgeneralizationsare
false.com/media-bias-chart-3-1-minor-updates-based-constructive-
To test these hypotheses, data were collected from 616 par- feedback. The sources were all classified as ‘‘most extreme’’ or
ticipants recruited from Amazon Mechanical Turk using the ‘‘hyper-partisan’’ in liberal or conservative positioning.
THE BLUE CHECK OF CREDIBILITY 3
Downloaded by East Carolina University from www.liebertpub.com at 03/12/19. For personal use only.
sources (B = 0.96, SE = 0.29, p = 0.001, odds ratio = 2.62). (B = -2.27, SE = 0.64, p = 0.000, odds ratio = 0.104). For
We limit all subsequent analyses to those who correctly example, over 95% of participants had heard of the low
identified the source of the tweet (N = 489). ambiguity accounts, and under 10% had heard of the low and
A second manipulation check asked whether participants medium ambiguity accounts.{
had heard of the account source before. Logistic regression
confirms that low ambiguity accounts were more well known
than the high ambiguity accounts (B = 5.22, SE = 0.58, {
The one outlier among medium and high ambiguity accounts
p = 0.000, odds ratio = 184.54), and that medium ambiguity was The Daily Caller, in which -27% of participants had heard of
accounts were less known than the high ambiguity accounts the source previously.
4 EDGERLY AND VRAGA
A final manipulation check asked participants whether moderate the effects of verification (RQ2), nor is there a
the account had a verified Twitter account. A logistic re- three-way interaction between these factors (RQ3).** Instead,
gression confirms that our manipulation of verification in- there is a main effect of congruence and an interaction with
creased the likelihood that participants correctly identified ambiguity.{{
the account as verified (B = 1.07, SE = 0.22, p = 0.000, odds
ratio = 2.92), from 39% when it was unverified to 61%
when it was. Discussion
This study tests the prevailing assumption that Twitter
Results account verification badges influence how people make
We test H1 using an analysis of covariance, which en- credibility judgments. Our results consistently show that
tered verification as the experimental factor while control- verification, on its own or combined with other source-
ling for the six sources examined in this study. Our first related cues, did not influence evaluations of a news tweet or
hypothesis is unsupported. Verification did not exert a main its source. Even when people were unfamiliar with the source
effect on tweet credibility, F(1,489) = 0.07, p = 0.79, partial (and were provided no partisan cues), the presence of a
g2 = 0.000, or account credibility, F(1,489) = 0.05, p = 0.82, verification mark did not alter credibility. Instead, account
partial g2 = 0.000. ambiguity and congruency were more powerful cues in as-
The next hypothesis and research question test whether the sessing credibility.
effects of verification are conditioned by characteristics of These findings may underscore the lost value of Twitter
the source itself—namely, the ambiguity of the account and verification badges. The number of Twitter accounts granted
its congruence. A series of three-way analysis of variance verification status doubled, from 150,000 to 300,000 accounts,
with verification, ambiguity, and congruence as factors in <3 years.10 Such growth parallels Twitter opening (and then
confirm that verification still does not have a main effect on closing) the application process to the general public. What was
tweet or account credibility (Table 1). Next, we test H2, once a useful heuristic for assessing credibility may now be
which predicts that verification matters more for high and unhelpful due to its prevalence. Of course, this assumes people
medium ambiguity accounts, compared with low ambiguity
accounts; and RQ1, which compares high and medium am-
**
biguity accounts. We find verification is not conditioned by We also ran these analyses excluding the high ambiguity
source ambiguity for either type of credibility.x conditions because these conditions lacked cues about congruence.
However, these analyses are substantively similar to those re-
The final research questions explore whether verification ported in the paper, suggesting these conditions were not muting
depends on congruence, as well as account ambiguity. We our ability to perceive an interaction between verification and
find no support for these expectations; congruence does not congruence. Please contact the lead author for more details on
these analyses.
{{
Consistent with previous research, people rate the tweets and
accounts as more credible when they are congruent (tweets: M = 4.06,
x
There is a significant main effect of ambiguity on perceptions of SE = 0.08; accounts: M = 4.54, SE = 0.10) rather than incongruent
account credibility, with medium ambiguity accounts (i.e., unknown (tweets: M = 3.66, SE = 0.08; accounts: M = 3.54, SE = 0.09) with their
sources with partisan cues) seen as the least credible (M = 3.78, party affiliation. In addition, a significant interaction between ambi-
SE = 0.12) compared to high ambiguity accounts without partisan guity and congruence shows that low ambiguity congruent accounts
cues (M = 4.29, SE = 0.12, p = 0.01 using Bonferroni adjustment), were seen as the most credible, whereas low and medium ambiguity
with low ambiguity accounts falling between (M = 4.04, SE = 0.10). incongruent accounts were seen as least credible. Please contact the
This effect is not significant for tweet credibility. authors for more details on these analyses.
THE BLUE CHECK OF CREDIBILITY 5
organizations (e.g., nonprofits, government, media, news). nication Quarterly 1988; 65:567–574.
By making these account cues more prominent and infor- 6. Metzger MJ, Flanagin AJ, Medders RB. Social and heu-
ristic approaches to credibility evaluation online. Journal of
mative, Twitter can draw boundaries between sources and
Communication 2010; 60:413–439.
better appeal to users who process online information heu-
7. Flanagin AJ, Metzger MJ. The role of site features, user
ristically.6 attributes, and information verification behaviors on the
There are several study limitations that should be con- perceived credibility of web-based information. New Media
sidered. The use of an MTurk sample limits our ability to & Society 2007; 9:319–342.
generalize the findings to the population at large, although 8. Sundar SS, Knobloch-Westerwick S, Hastall MR. News
emerging research suggests experiments on MTurk often cues: information scent and cognitive heuristics. Journal of
produce similar results to general population samples.11 the American Society for Information Science and Tech-
Additionally, more research is needed to determine if our nology 2007; 58:366–378.
results apply to other social media sites like Facebook. 9. Chaiken S. (1987) The heuristic model of persuasion. In
Lastly, we focused on source-related factors and did not Zanna MP, Olson JM, Herman CP, eds. Social Influence:
manipulate tweet content. An overtly opinionated tweet or The Ontario Symposium 5. Hillsdale, NJ: Erlbaum, pp. 3–39.
one that appeals to certain news values may produce an in- 10. Kamps HJ. Who are Twitter’s verified users? Medium. May
fluence of verification on credibility. Ultimately, there is still 25, 2015. https://medium.com/@Haje/who-are-twitter-s-
much work to be done in this area. While people undoubtedly verified-users-af976fc1b032 (accessed February 4, 2019).
rely on heuristics in making credibility judgments online, it 11. Coppock A, Lepper TJ, Mullinix KJ. The generaliability of
appears that the verification mark on Twitter is currently not heterogenous treatment effect esitmates across samples.
one of those cues. PNAS 2018; 115:12441–12446.
Address correspondence to:
Acknowledgment Dr. Stephanie Edgerly
The authors thank the Medill School of Journalism at Medill School of Journalism
Northwestern University for providing funding for this study. Northwestern University
1870 Campus Drive
Author Disclosure Statement Evanston, IL 60208
{{
People were significantly more likely to report low ambiguity
accounts had a verified account as compared to high ambiguity
accounts (B = 0.74, SE = 0.26, p = 0.004, odds ratio = 2.10). Among
the medium and high ambiguity accounts (which on average 10%
had heard of previously), roughly 37% said the unverified account
was verified, compared to 55% when the account was verified.